Code

Feature/mrhs misc#1515

Code

Merged

maddyscientist merged 32 commits intodeveloplattice/quda:developfrom feature/mrhs-misclattice/quda:feature/mrhs-miscCopy head branch name to clipboardDec 5, 2024

+378-220

Commits

Commits on Oct 11, 2024

• Initial support for register for staggered dslash kernel

  maddyscientist

  committed
  8292b17

  Copy full SHA for 8292b17

• Sanity check for QUDA_MAX_MULTI_RHS_TILE

  maddyscientist

  committed
  e816fb1

  Copy full SHA for e816fb1

• Possible WAR for ROCm performance issue with MRHS staggered kernels

  maddyscientist

  committed
  eeea3b0

  Copy full SHA for eeea3b0

• FieldTmp now supports creating temporaries using a T::param_type

  maddyscientist

  committed
  467a493

  Copy full SHA for 467a493

• DslashCoarse now uses getFieldTmp for it mma temporaries to avoid allocation overheads when memory pool is disabled

  maddyscientist

  committed
  36fcf3d

  Copy full SHA for 36fcf3d

Commits on Oct 14, 2024

• Small cleanup of staggered dslash

  maddyscientist

  committed
  db865e4

  Copy full SHA for db865e4

Commits on Oct 15, 2024

• Add default initialization to array::data

  maddyscientist

  committed
  8f72560

  Copy full SHA for 8f72560

• Fix multi-gpu bug

  maddyscientist

  committed
  b2cebb0

  Copy full SHA for b2cebb0

• Fixed a logic bug in the MR convergence check

  

  weinbe2

  authored and

  maddyscientist

  committed
  704f6b2

  Copy full SHA for 704f6b2

• Revert "Add default initialization to array::data"

  Show description for 57c631c
  maddyscientist

  committed
  57c631c

  Copy full SHA for 57c631c

Commits on Oct 16, 2024

• n_src should be a member of DslashArg

  maddyscientist

  committed
  9ddd25b

  Copy full SHA for 9ddd25b

Commits on Oct 17, 2024

• Fix fused pack + dslash kernels with n_src_tile > 1

  maddyscientist

  committed
  01725e6

  Copy full SHA for 01725e6

Commits on Nov 18, 2024

• Merge branch 'feature/trlm_3d' of github.com:lattice/quda into feature/mrhs-misc

  maddyscientist

  committed
  bbd07a7

  Copy full SHA for bbd07a7

• Fix bug in dslash_test_utils.h

  maddyscientist

  committed
  77b6b59

  Copy full SHA for 77b6b59

• nvc++ no longer needs to use constant memory args for dslash

  maddyscientist

  committed
  27339e4

  Copy full SHA for 27339e4

• Fix for nvc++ and remove unneeded target specific thread_array.h files

  maddyscientist

  committed
  936c2c4

  Copy full SHA for 936c2c4

Commits on Nov 19, 2024

• Remove WAR for nvc++ in reduce_helper which is no longer needed

  maddyscientist

  committed
  fd00d08

  Copy full SHA for fd00d08

Commits on Nov 20, 2024

• Add versioned CPM files to .gitignore

  maddyscientist

  committed
  fc5ac97

  Copy full SHA for fc5ac97

• Fix complex_quda.h to be C++20 compliant

  maddyscientist

  committed
  a10d6f3

  Copy full SHA for a10d6f3

• Add new variant of heterogeneous reductions: with some compilers, optimization can optimize out infinities rendering the use of heterogenous atomics with an inifinity sentinal. As an alternative we…

  Show description for c47e1f0
  maddyscientist

  committed
  c47e1f0

  Copy full SHA for c47e1f0

• Update to use latest release of Eigen 3.4: this fixes some bugs with nvc++ allowing for the removal of the WARs deployed previously

  maddyscientist

  committed
  8b61db2

  Copy full SHA for 8b61db2

• Apply ROCm perf WAR to Laplace operator

  maddyscientist

  committed
  490930b

  Copy full SHA for 490930b

• Fix nvc++ compiler warning

  maddyscientist

  committed
  26c3e59

  Copy full SHA for 26c3e59

Commits on Nov 22, 2024

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

  maddyscientist

  committed
  1af8c52

  Copy full SHA for 1af8c52

Commits on Nov 27, 2024

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

  maddyscientist

  committed
  c2547fc

  Copy full SHA for c2547fc

• Fix compiler warning introduced with 3debb29

  maddyscientist

  committed
  2b81309

  Copy full SHA for 2b81309

• Fix process divergence issues (could hang when autotuning) in genericPrintMatrix routine. Apply same patch to genericPrintVector for future proofing

  maddyscientist

  committed
  0a3f608

  Copy full SHA for 0a3f608

Commits on Dec 3, 2024

• Add pragma unroll

  maddyscientist

  committed
  22393c5

  Copy full SHA for 22393c5

Commits on Dec 4, 2024

• Do not run BiCGStab(l) with staggered half precision since it is unstable

  maddyscientist

  committed
  8ffcdd5

  Copy full SHA for 8ffcdd5

Commits on Dec 5, 2024

• sentinal -> sentinel

  maddyscientist

  committed
  a860b1c

  Copy full SHA for a860b1c

• Apply review comments

  maddyscientist

  committed
  c3fbf50

  Copy full SHA for c3fbf50

• Apply clang format

  maddyscientist

  committed
  b6f5edb

  Copy full SHA for b6f5edb