Code

Optimize ompGemm_m multiply kernel#327

Code

Merged

tkoskela merged 45 commits intodevelopOrderN/CONQUEST-release:developfrom tk-optimize-multiply-kernelOrderN/CONQUEST-release:tk-optimize-multiply-kernelCopy head branch name to clipboardMay 22, 2024

+249-163

Commits

Commits on Dec 19, 2023

• Add ompGemm_m kernel parallelised over j in both min and max kernel. Does not work.

  tkoskela

  committed
  19d26e8

  Copy full SHA for 19d26e8

• Merge branch 'tk-ci-test-multiply-kernels' into tk-omp-experiments

  tkoskela

  committed
  04b83f1

  Copy full SHA for 04b83f1

• Add multiply kernel with parallelisation over j in both min and max kernels

  tkoskela

  committed
  6c4fd5d

  Copy full SHA for 6c4fd5d

• Add kernel where we thread over both i and j with collapse(2)

  tkoskela

  committed
  0242279

  Copy full SHA for 0242279

• Change schedule to runtime for testing. Fix omp end bug

  tkoskela

  committed
  914166a

  Copy full SHA for 914166a

Commits on Jan 10, 2024

• Add OpenAcc multiply kernel for testing

  tkoskela

  committed
  b7822b8

  Copy full SHA for b7822b8

• Merge branch 'develop' into tk-omp-experiments

  tkoskela

  committed
  4d19c11

  Copy full SHA for 4d19c11

Commits on Jan 16, 2024

• Copy A matrix in array syntax

  tkoskela

  committed
  7cf301d

  Copy full SHA for 7cf301d

• Do b copy using array syntax outside parallel loop

  tkoskela

  committed
  c94cf89

  Copy full SHA for c94cf89

• Remove commented out code

  tkoskela

  committed
  2e65616

  Copy full SHA for 2e65616

Commits on Jan 19, 2024

• Version of acc kernel that compiles but doesn't work.

  Show description for 8680442
  tkoskela

  committed
  8680442

  Copy full SHA for 8680442

• Use allocatables instead of pointers

  tkoskela

  committed
  9930b94

  Copy full SHA for 9930b94

Commits on Jan 25, 2024

• Shorten loop to copy C

  tkoskela

  committed
  06ed04c

  Copy full SHA for 06ed04c

• Remove unnecessary initializations

  tkoskela

  committed
  41d45a6

  Copy full SHA for 41d45a6

• Avoid loop carrier dependencies by precomputing indices

  tkoskela

  committed
  dfb30e6

  Copy full SHA for dfb30e6

• Copy back to C with a logical mask

  Show description for 90e9121
  tkoskela

  committed
  90e9121

  Copy full SHA for 90e9121

Commits on Jan 30, 2024

• Use pointer to A instead of data copy

  tkoskela

  committed
  1f29e90

  Copy full SHA for 1f29e90

• Use a pointer to B instead of data copy

  tkoskela

  committed
  d50b006

  Copy full SHA for d50b006

• Clean up and reorder things

  tkoskela

  committed
  704d3c5

  Copy full SHA for 704d3c5

Commits on Jan 31, 2024

• Less indices. Still too many indices

  tkoskela

  committed
  1f5a3a7

  Copy full SHA for 1f5a3a7

• Ignore more files in tests and benchmarks

  tkoskela

  committed
  b07653d

  Copy full SHA for b07653d

• Move index precomputation out of parallel loop

  tkoskela

  committed
  453ae68

  Copy full SHA for 453ae68

Commits on Feb 2, 2024

• Add syste.make file for myriad

  tkoskela

  committed
  d0f0db7

  Copy full SHA for d0f0db7

• Merge branch 'tk-add-system.make.myriad' into tk-optimize-multiply-kernel

  tkoskela

  committed
  ef65a38

  Copy full SHA for ef65a38

Commits on Feb 5, 2024

• Add XC_COMPFLAGS to compiler with libxc v4

  tkoskela

  committed
  411bdbf

  Copy full SHA for 411bdbf

Commits on Feb 8, 2024

• Fix copy c loop. Add comments

  tkoskela

  committed
  1a94b96

  Copy full SHA for 1a94b96

• Use right multiply kernel on myriad

  tkoskela

  committed
  80c6b53

  Copy full SHA for 80c6b53

Commits on Feb 13, 2024

• WIP: refactor m_kern_min

  tkoskela

  committed
  2da61e5

  Copy full SHA for 2da61e5

Commits on Feb 19, 2024

• Test ompGemm_m in main workflow

  tkoskela

  committed
  a076735

  Copy full SHA for a076735

Commits on Feb 21, 2024

• Refactor index computations out of m_kern_min and m_kern_max

  Show description for 0018828
  tkoskela

  committed
  0018828

  Copy full SHA for 0018828

Commits on Mar 1, 2024

• Cleanup + array syntax copies of B and C

  tkoskela

  committed
  c94c274

  Copy full SHA for c94c274

Commits on Mar 14, 2024

• Merge branch 'develop' into tk-optimize-multiply-kernel

  tkoskela

  committed
  6fe131f

  Copy full SHA for 6fe131f

Commits on Mar 18, 2024

• remove extra kernel

  tkoskela

  committed
  c35452e

  Copy full SHA for c35452e

Commits on Mar 21, 2024

• remove experimental kernels

  tkoskela

  committed
  daff6b3

  Copy full SHA for daff6b3

• Clean up code and comments

  tkoskela

  committed
  f534039

  Copy full SHA for f534039

• Merge branch 'develop' into tk-optimize-multiply-kernel

  tkoskela

  committed
  d4e6225

  Copy full SHA for d4e6225

• Merge branch 'tk-optimize-multiply-kernel' of github.com:OrderN/CONQUEST-release into tk-optimize-multiply-kernel

  tkoskela

  committed
  0acdb7f

  Copy full SHA for 0acdb7f

• Revert inefficient vectorized copies, add explaining comment

  tkoskela

  committed
  1bea12a

  Copy full SHA for 1bea12a

• Add missing variable declaration

  tkoskela

  committed
  6cecd58

  Copy full SHA for 6cecd58

• Rename myriad make file to new convention

  tkoskela

  committed
  4fddbce

  Copy full SHA for 4fddbce

Commits on Apr 19, 2024

• Remove barriers in multiply_module

  tkoskela

  committed
  843db36

  Copy full SHA for 843db36

Commits on Apr 23, 2024

• Update to myriad makefile

  tkoskela

  committed
  950f2f2

  Copy full SHA for 950f2f2

Commits on May 3, 2024

• Remove unused targets and commented out pointers

  tkoskela

  committed
  744d635

  Copy full SHA for 744d635

Commits on May 13, 2024

• Merge branch 'develop' into tk-optimize-multiply-kernel

  tkoskela

  committed
  4347802

  Copy full SHA for 4347802

Commits on May 17, 2024

• Revert back to loop-based implementation of c copy

  tkoskela

  committed
  cdd509e

  Copy full SHA for cdd509e