Code

Vectorisation sprint#654

Code

Closed

sv2518 wants to merge 107 commits intomasterOP2/PyOP2:masterfrom vectorisation-sprintOP2/PyOP2:vectorisation-sprintCopy head branch name to clipboard

+528-31

Commits

Commits on Jul 22, 2020

• codegen: Implement SIMD vectorisation

  Show description for 2a8d17c
  

  tj-sun

  authored and

  sv2518

  committed
  2a8d17c

  Copy full SHA for 2a8d17c

• add omp simd vectorization mode

  

  tj-sun

  authored and

  sv2518

  committed
  fbc6e4a

  Copy full SHA for fbc6e4a

• add openmp flag and by pass workaround flag

  

  tj-sun

  authored and

  sv2518

  committed
  5ae780d

  Copy full SHA for 5ae780d

• DROP BEFORE MERGE: test with correct loopy branch

  

  wence-

  authored and

  sv2518

  committed
  ba693dc

  Copy full SHA for ba693dc

• Turn of tree vectorize for certain gcc compilers. We might not need the tree vectorisation flag for our vectorisation anyways.

  sv2518

  committed
  4ec0769

  Copy full SHA for 4ec0769

• Add simd compiler flags.

  sv2518

  committed
  f9e60fd

  Copy full SHA for f9e60fd

• Remove time configuration.

  sv2518

  committed
  00e073d

  Copy full SHA for 00e073d

• Default SIMD width.

  sv2518

  committed
  1cf7698

  Copy full SHA for 1cf7698

• Generate CVec Target with batch size infomation and move typedef into loopy codebase.

  sv2518

  committed
  3e66946

  Copy full SHA for 3e66946

• Move zero declaration to loopy code base to be more robust in naming the variable.

  sv2518

  committed
  1238ce8

  Copy full SHA for 1238ce8

• Added conditionals when to vectorise:

  Show description for 1d54777
  sv2518

  committed
  1d54777

  Copy full SHA for 1d54777

• Drop omp vectorisation.

  sv2518

  committed
  b369213

  Copy full SHA for b369213

• Add -march=native everywhere.

  sv2518

  committed
  1c6346e

  Copy full SHA for 1c6346e

• Silence warnings.

  sv2518

  committed
  856b6aa

  Copy full SHA for 856b6aa

Commits on Aug 24, 2020

• Change vector tag.

  sv2518

  committed
  5e52ce1

  Copy full SHA for 5e52ce1

Commits on Sep 1, 2020

• Give more control over vectorisation to PyOP2.

  sv2518

  committed
  537c14c

  Copy full SHA for 537c14c

• Naming adaption.

  sv2518

  committed
  9317654

  Copy full SHA for 9317654

• Realize ilp first.

  sv2518

  committed
  6723b6a

  Copy full SHA for 6723b6a

• Jenkins.

  sv2518

  committed
  38ebc8a

  Copy full SHA for 38ebc8a

Commits on Feb 28, 2022

• Merge branch 'master' into vectorisation-restructure-checks

  sv2518

  committed
  32b2910

  Copy full SHA for 32b2910

Commits on Mar 1, 2022

• DBM: run against new loopy branch

  sv2518

  committed
  944c6cf

  Copy full SHA for 944c6cf

• Lint

  sv2518

  committed
  3a1eb24

  Copy full SHA for 3a1eb24

• More adapations to new PyOP2

  sv2518

  committed
  681e315

  Copy full SHA for 681e315

• More adapations to new PyOP2

  sv2518

  committed
  48d6142

  Copy full SHA for 48d6142

Commits on Mar 3, 2022

• DBM take the correct branch

  sv2518

  committed
  792c8f0

  Copy full SHA for 792c8f0

• Adapt to new PyOP2 and vectorisation

  sv2518

  committed
  2469870

  Copy full SHA for 2469870

• Adapt to new PyOP2 and vectorisation

  sv2518

  committed
  4bbcde5

  Copy full SHA for 4bbcde5

• Fix return wrapper with kernel not kernel

  sv2518

  committed
  a5c0455

  Copy full SHA for a5c0455

• We do need to inline bc Implementing transforms that apply cleanly across caller-callee is a bit involved and loopy can't deal with it yet.

  sv2518

  committed
  c374031

  Copy full SHA for c374031

• First split then tag because loopy does not support retaggin of inames by default

  sv2518

  committed
  e7d31eb

  Copy full SHA for e7d31eb

• tag_array_axes requires us to specify the tags for each dimension of the temporary. I.e. if the temporary is 1D: 'vec', if the temporary is 2D: 'vec, c', if the temporary is 3D: 'vec,c,c'

  sv2518

  committed
  56a8dde

  Copy full SHA for 56a8dde

• Fix

  sv2518

  committed
  d1171b3

  Copy full SHA for d1171b3

• fix

  sv2518

  committed
  0641c75

  Copy full SHA for 0641c75

• improve comments

  sv2518

  committed
  644842e

  Copy full SHA for 644842e

• tag only non-constant arrays with vec axes

  kaushikcfd

  committed
  9e58b22

  Copy full SHA for 9e58b22

Commits on Mar 4, 2022

• Only vectorise when local kernel is a loopy thing.

  sv2518

  committed
  3f133fd

  Copy full SHA for 3f133fd

• shift iel-loop to have lbound of 0

  kaushikcfd

  committed
  dcd0b69

  Copy full SHA for dcd0b69

Commits on Mar 6, 2022

• Fix import

  sv2518

  committed
  907fe58

  Copy full SHA for 907fe58

• Debug: try with newer python version

  sv2518

  committed
  ca2aaaf

  Copy full SHA for ca2aaaf

• Debug: try with newer python version

  sv2518

  committed
  0440f66

  Copy full SHA for 0440f66

Commits on Mar 7, 2022

• change target before inlining

  kaushikcfd

  committed
  4bcb592

  Copy full SHA for 4bcb592

• ignore loopy vectorization fallback warnings

  kaushikcfd

  committed
  d42e7e8

  Copy full SHA for d42e7e8

Commits on Mar 11, 2022

• Revert "Debug: try with newer python version"

  Show description for 7e37e02
  sv2518

  committed
  7e37e02

  Copy full SHA for 7e37e02

• Make complex check tighter

  sv2518

  committed
  b541dbd

  Copy full SHA for b541dbd

• extend the set of variables that cannot be vecotrized

  kaushikcfd

  committed
  caa567a

  Copy full SHA for caa567a

Commits on Mar 14, 2022

• Attempt to fix Slate by inlining of all subkernels

  sv2518

  committed
  c3a96fa

  Copy full SHA for c3a96fa

• Add comment

  sv2518

  committed
  dc996de

  Copy full SHA for dc996de

Commits on Mar 15, 2022

• placate flake8

  kaushikcfd

  committed
  fa343e1

  Copy full SHA for fa343e1

Commits on Apr 1, 2022

• blas callables: do not accept vectorized dtypes

  kaushikcfd

  committed
  aa7bc0c

  Copy full SHA for aa7bc0c

Commits on May 5, 2022

• allow inverse.c::inverse() to take in vector dtypes

  kaushikcfd

  committed
  8302d52

  Copy full SHA for 8302d52

• Merge remote-tracking branch 'origin/master' into vectorisation-sprint

  kaushikcfd

  committed
  a767fe2

  Copy full SHA for a767fe2

• do not invoke the vectorization pass if one of the arguments is a MixedMatKernelArg

  kaushikcfd

  committed
  85de156

  Copy full SHA for 85de156

• makes freeing logic accurate

  kaushikcfd

  committed
  30f8ecb

  Copy full SHA for 30f8ecb

Commits on May 6, 2022

• rewrite solve to accept strided inputs

  kaushikcfd

  committed
  0d5023d

  Copy full SHA for 0d5023d

• blas-helpers: corrects the freeing logic

  kaushikcfd

  committed
  d25545b

  Copy full SHA for d25545b

• Don't vectorise the kernel which generates the coordinates for the extrusion

  sv2518

  committed
  0ade829

  Copy full SHA for 0ade829

• PyOP2 compilation: add a pathway to compile with gcc on Mac.

  sv2518

  committed
  a4bab8e

  Copy full SHA for a4bab8e

Commits on May 8, 2022

• do not vectorize the entire kernel if some instruction are surrounded by conditional depending on vectorizing iname

  kaushikcfd

  committed
  175eb14

  Copy full SHA for 175eb14

• loop being split starts from '0' => do not peel at the head

  kaushikcfd

  committed
  8256bd2

  Copy full SHA for 8256bd2

Commits on May 9, 2022

• Merge branch 'vectorisation-sprint' of github.com:OP2/PyOP2 into vectorisation-sprint

  sv2518

  committed
  6585dbb

  Copy full SHA for 6585dbb

• Add comment

  sv2518

  committed
  4c0ca6e

  Copy full SHA for 4c0ca6e

Commits on May 10, 2022

• Fix complex check?

  sv2518

  committed
  e744092

  Copy full SHA for e744092

• Fix complex check?

  sv2518

  committed
  5fc4264

  Copy full SHA for 5fc4264

• Fix complex check?

  sv2518

  committed
  31f0c39

  Copy full SHA for 31f0c39

• Fix complex check?

  sv2518

  committed
  7e8a86a

  Copy full SHA for 7e8a86a

Commits on May 11, 2022

• clarifies vectorization strategy

  kaushikcfd

  committed
  63f1e52

  Copy full SHA for 63f1e52

• Updates to transform startegy

  Show description for 8b19370
  kaushikcfd

  committed
  8b19370

  Copy full SHA for 8b19370

Commits on May 19, 2022

• Time configuration is not used anywhere and add doc

  sv2518

  committed
  7a2cbd6

  Copy full SHA for 7a2cbd6

• Move conditional

  sv2518

  committed
  69d4921

  Copy full SHA for 69d4921

• sun2020study -> cross-element

  sv2518

  committed
  43960e6

  Copy full SHA for 43960e6

• Make default_simd_width more readable

  sv2518

  committed
  b4c9926

  Copy full SHA for b4c9926

• cleanup

  sv2518

  committed
  c603f3f

  Copy full SHA for c603f3f

• Lint

  sv2518

  committed
  1cee3d7

  Copy full SHA for 1cee3d7

Commits on May 20, 2022

• corrects the condition to not vectorize temps passed to BLAS calls

  kaushikcfd

  committed
  a671b6c

  Copy full SHA for a671b6c

Commits on May 24, 2022

• Add vectorisation config to cache keys

  sv2518

  committed
  4aa86e1

  Copy full SHA for 4aa86e1

• Tests: add a vectorisation test

  sv2518

  committed
  60b4b3e

  Copy full SHA for 60b4b3e

• Cleanup

  sv2518

  committed
  1b3c29e

  Copy full SHA for 1b3c29e

• Cleanup

  sv2518

  committed
  0a54a34

  Copy full SHA for 0a54a34

• Use reconfigure not init for changing the vectorisation strategy in the test

  sv2518

  committed
  9b23200

  Copy full SHA for 9b23200

• Cleanup

  sv2518

  committed
  acb9c89

  Copy full SHA for acb9c89

• Test: improve the vectorisation test.

  sv2518

  committed
  49e2779

  Copy full SHA for 49e2779

• Put vectorisation strategy only in cache key of the global kernel.

  sv2518

  committed
  e5fe4d2

  Copy full SHA for e5fe4d2

Commits on May 25, 2022

• lint

  sv2518

  committed
  0eff9d6

  Copy full SHA for 0eff9d6

• Fix docs

  sv2518

  committed
  22ce06e

  Copy full SHA for 22ce06e

• Fix config error

  sv2518

  committed
  bdefbfa

  Copy full SHA for bdefbfa

• Fix config error

  sv2518

  committed
  2a459e5

  Copy full SHA for 2a459e5

Commits on May 27, 2022

• Don't add py-cpuinfo

  

  Jack Betteridge

  authored and

  Jack Betteridge

  committed
  56c65da

  Copy full SHA for 56c65da

Commits on Jun 22, 2022

• Add nbytes property

  Show description for ca5c51b
  connorjward

  committed
  ca5c51b

  Copy full SHA for ca5c51b

• Drop unused args

  sv2518

  committed
  dc5f3bc

  Copy full SHA for dc5f3bc

• Time->extra_info

  sv2518

  committed
  ac36708

  Copy full SHA for ac36708

• Merge branch 'vectorisation-sprint' into connorjward/add-nbytes

  sv2518

  authored
  89c9dec

  Copy full SHA for 89c9dec

• Merge pull request #666 from OP2/connorjward/add-nbytes

  Show description for e2af4c7
  sv2518

  authored
  e2af4c7

  Copy full SHA for e2af4c7

• Merge branch 'vectorisation-sprint' into JDBetteridge/vectorisation-sprint

  sv2518

  authored
  4de6f06

  Copy full SHA for 4de6f06

• Merge pull request #665 from OP2/JDBetteridge/vectorisation-sprint

  Show description for 2840f28
  sv2518

  authored
  2840f28

  Copy full SHA for 2840f28

Commits on Jun 24, 2022

• Fix bandwidth calculation

  

  Sophia Vorderwuelbecke

  committed
  89feb72

  Copy full SHA for 89feb72

• Add simd compiler flag also to LinuxGNU compiler

  

  Sophia Vorderwuelbecke

  committed
  0857145

  Copy full SHA for 0857145

Commits on Jun 27, 2022

• Add vectorisation flag to linux clang compiler too

  

  Sophia Vorderwuelbecke

  committed
  662241e

  Copy full SHA for 662241e

Commits on Jul 6, 2022

• account for changed in loopy's vectorization syntax

  kaushikcfd

  committed
  203223c

  Copy full SHA for 203223c

• run CI with py3.8

  Show description for fae323f
  kaushikcfd

  committed
  fae323f

  Copy full SHA for fae323f

Commits on Jul 7, 2022

• Fallback for stopping criterium

  sv2518

  committed
  030cae5

  Copy full SHA for 030cae5

• Fallback for stopping criterium

  sv2518

  committed
  ece0e62

  Copy full SHA for ece0e62

• Reduce inames to untag

  sv2518

  committed
  934e147

  Copy full SHA for 934e147

• Reduce inames to untag

  sv2518

  committed
  bd95ba3

  Copy full SHA for bd95ba3

• Fallback for stopping criterium

  sv2518

  committed
  fd6650d

  Copy full SHA for fd6650d

Commits on Jul 11, 2022

• unroll (not vectorize) loops surrounding CInstructions

  kaushikcfd

  committed
  f69755d

  Copy full SHA for f69755d

• get rid of noop insns

  kaushikcfd

  committed
  e72f316

  Copy full SHA for e72f316

Commits on Oct 4, 2022

• Fix merge leftovers for vectorisation in chapter 3

  sv2518

  committed
  09bf629

  Copy full SHA for 09bf629