Code

Idr solver#620

Code

Merged

fritzgoebel merged 33 commits intodevelopginkgo-project/ginkgo:developfrom IDR-solverginkgo-project/ginkgo:IDR-solverCopy head branch name to clipboardNov 26, 2020

+5,421-46

Commits

Commits on Nov 20, 2020

• Add IDR skeleton

  fritzgoebel

  committed
  8e104fc

  Copy full SHA for 8e104fc

• Add core implementation

  fritzgoebel

  committed
  c4594a8

  Copy full SHA for c4594a8

• Add reference kernels

  fritzgoebel

  committed
  8746304

  Copy full SHA for 8746304

• Finish reference implementation, add reference tests and omp implementation

  fritzgoebel

  committed
  78350c9

  Copy full SHA for 78350c9

• Add some omp tests

  fritzgoebel

  committed
  e89f558

  Copy full SHA for e89f558

• Add cuda implementation

  fritzgoebel

  committed
  1a9e6ba

  Copy full SHA for 1a9e6ba

• Add hip implementation and core / hip tests

  fritzgoebel

  committed
  745f13f

  Copy full SHA for 745f13f

• Write documentation

  fritzgoebel

  committed
  a6cc762

  Copy full SHA for a6cc762

• Add comments for factory parameters

  fritzgoebel

  committed
  34a5462

  Copy full SHA for 34a5462

• add IDR to benchmarks

  fritzgoebel

  committed
  fed0b36

  Copy full SHA for fed0b36

• Add logging to IDR

  fritzgoebel

  committed
  4899c07

  Copy full SHA for 4899c07

• Add orthogonalization for P

  fritzgoebel

  committed
  98ad686

  Copy full SHA for 98ad686

• Move g_k to separate vector instead of using it as a submatrix of g

  fritzgoebel

  committed
  ddec93b

  Copy full SHA for ddec93b

• Add multidot to optimize update_g_and_u

  fritzgoebel

  committed
  8340e04

  Copy full SHA for 8340e04

• Use multidot for m update

  fritzgoebel

  committed
  f217822

  Copy full SHA for f217822

• improve random initialization of P

  fritzgoebel

  committed
  bdc4223

  Copy full SHA for bdc4223

• Add hiprand binding

  fritzgoebel

  committed
  8012cbf

  Copy full SHA for 8012cbf

• Add option to deterministically generate subspace vectors for testing

  fritzgoebel

  committed
  32ae466

  Copy full SHA for 32ae466

• Suggestions from code review

  fritzgoebel

  committed
  7e7d7ba

  Copy full SHA for 7e7d7ba

• change images to hiprand ones

  fritzgoebel

  committed
  b7741f9

  Copy full SHA for b7741f9

• Suggestions from Code Review

  fritzgoebel

  committed
  af1288c

  Copy full SHA for af1288c

Commits on Nov 25, 2020

• Add option to use complex subspace for real problems

  fritzgoebel

  committed
  b72336c

  Copy full SHA for b72336c

• Add hip support for complex blas functions

  fritzgoebel

  committed
  6e6885e

  Copy full SHA for 6e6885e

• Add missing check for complex in hip and cuda

  fritzgoebel

  committed
  fd0f34e

  Copy full SHA for fd0f34e

• Disable HIP with CUDA < 9.2.

  

  tcojean

  authored and

  fritzgoebel

  committed
  c39d929

  Copy full SHA for c39d929

• Add DPC++ NOT_IMPLEMENTED IDR kernels.

  

  tcojean

  authored and

  fritzgoebel

  committed
  51502e6

  Copy full SHA for 51502e6

• Remove make_diag_dominant from CUDA test

  fritzgoebel

  committed
  d6c2174

  Copy full SHA for d6c2174

• Add math header to curand and hiprand bindings, use hiprand image for subdir_build

  fritzgoebel

  committed
  a726de6

  Copy full SHA for a726de6

• Suggestions from code review

  fritzgoebel

  committed
  9a37ebd

  Copy full SHA for 9a37ebd

• use default containers

  fritzgoebel

  committed
  e0669d7

  Copy full SHA for e0669d7

• fix dpcpp kernels, suggestions from code review

  fritzgoebel

  committed
  8dd248e

  Copy full SHA for 8dd248e

• update windows build

  fritzgoebel

  committed
  401b500

  Copy full SHA for 401b500

• slightly larger tolerance for multiple rhs tests

  fritzgoebel

  committed
  eb10002

  Copy full SHA for eb10002