Code

Debugging PI crash#88

Code

Draft

manodeep wants to merge 52 commits intomainACCESS-NRI/ACCESS-ESM1.6:mainfrom

update_to_oneapiACCESS-NRI/ACCESS-ESM1.6:update_to_oneapiCopy head branch name to clipboard

Commits

Commits on Mar 24, 2025

• Updating to oneAPI + re-organised the yaml file (draft)

  manodeep

  committed
  6923525

  Copy full SHA for 6923525

• Request gcc-runtime and glibc with %gcc

  manodeep

  committed
  2a9bb04

  Copy full SHA for 2a9bb04

Commits on Mar 25, 2025

• Added the compiler flags to target the SPR arch (but produce exes that will also run on cascadelake, broadwell)

  manodeep

  committed
  083ad75

  Copy full SHA for 083ad75

• Hmm cice4 does not like march=sapphirerapids - swapping to cascadelake

  manodeep

  committed
  86a0187

  Copy full SHA for 86a0187

• Add flto to UM

  manodeep

  committed
  57e2386

  Copy full SHA for 57e2386

• Add mtune=sapphirerapids and remove CORE-AVX2 and cascadelake

  manodeep

  committed
  869ae24

  Copy full SHA for 869ae24

• added qopt-zmm-usage=high

  manodeep

  committed
  8d416f2

  Copy full SHA for 8d416f2

• Add opt-prefetch=3

  manodeep

  committed
  b0b745b

  Copy full SHA for b0b745b

• Add fma

  manodeep

  committed
  cb43b46

  Copy full SHA for cb43b46

Commits on Mar 26, 2025

• Removed fma (to rebuild the failed deployment)

  manodeep

  committed
  ae1c3c9

  Copy full SHA for ae1c3c9

Commits on Mar 28, 2025

• Changing to fp-model fast but with high precision (and adding protect-parens to limit re-ordering)

  manodeep

  committed
  b5f9f7a

  Copy full SHA for b5f9f7a

Commits on Mar 31, 2025

• Added nan-compares and ieee-compares flags

  manodeep

  committed
  5ae1455

  Copy full SHA for 5ae1455

• Remove the nan+ieee compares options from cice4 (otherwise the deploy fails)

  manodeep

  committed
  63189b9

  Copy full SHA for 63189b9

• Silly me - those flags only apply to fortran

  manodeep

  committed
  72889d2

  Copy full SHA for 72889d2

Commits on Apr 7, 2025

• Remove compiler options that don't seem to speedup runtime

  manodeep

  committed
  53f3fac

  Copy full SHA for 53f3fac

• Added some more flags to better track div-by-zero

  manodeep

  committed
  62eb39a

  Copy full SHA for 62eb39a

• Build with my mom5 patch to avoid div-by-zero

  manodeep

  committed
  4f3ed08

  Copy full SHA for 4f3ed08

Commits on Apr 9, 2025

• Go back to the version I was using for performance testing and update the dependency versions

  manodeep

  committed
  baa0707

  Copy full SHA for baa0707

Commits on Apr 12, 2025

• Remove nan initialisers, add unroll and zmm-usage

  manodeep

  committed
  4fcda11

  Copy full SHA for 4fcda11

• Always vectorize (regardless of profitability)

  manodeep

  committed
  26d4cd8

  Copy full SHA for 26d4cd8

Commits on May 1, 2025

• Remove check all and fpe0. Aligned the code options to make it easier to read

  manodeep

  committed
  dd14cc1

  Copy full SHA for dd14cc1

• Added recommended flags for profiling

  manodeep

  committed
  3812f3d

  Copy full SHA for 3812f3d

• Hmm may be sibling calls optimisation is the issue

  manodeep

  committed
  88a4a82

  Copy full SHA for 88a4a82

Commits on May 6, 2025

• g3 might be helpful for the stack traces

  manodeep

  committed
  7b7571e

  Copy full SHA for 7b7571e

Commits on May 7, 2025

• valgrind has a SIGILL. Drop down to AVX2 (instead of sapphirerapids, ie AVX512)

  manodeep

  committed
  138a487

  Copy full SHA for 138a487

• Helps to put a '-' in front of flags

  manodeep

  committed
  5db0f29

  Copy full SHA for 5db0f29

• Change the global arch to x86-64-v3, i.e., AVX2

  manodeep

  committed
  dc85776

  Copy full SHA for dc85776

• Added flags to all compiled dependencies

  manodeep

  committed
  8955165

  Copy full SHA for 8955165

Commits on May 8, 2025

• Added the nan + huge init back

  manodeep

  committed
  429392c

  Copy full SHA for 429392c

• Added the -heap-arrays compiler flag

  manodeep

  committed
  781c4c3

  Copy full SHA for 781c4c3

Commits on May 13, 2025

• Remove the initialisers and add the nominus0 (but may nominus0 is the default?)

  manodeep

  committed
  a7a0ddd

  Copy full SHA for a7a0ddd

• Update the UM commit hash to fix build failure

  manodeep

  committed
  8824a70

  Copy full SHA for 8824a70

Commits on May 14, 2025

• Revert to exe (pr67-2) with div-by-zero + xcore-avx2 for valgrind

  manodeep

  committed
  6bff325

  Copy full SHA for 6bff325

• Still crashed with cpuid-type error. Adding xcore-avx2 to everything

  manodeep

  committed
  b3b3951

  Copy full SHA for b3b3951

Commits on May 15, 2025

• Keep getting wombatlite errors. Updated packages to the ones within the latest released version

  manodeep

  committed
  825a052

  Copy full SHA for 825a052

• Segfaults again (but no wombatlite errors) within valgrind. Going back to compiler-level check

  manodeep

  committed
  72f0e76

  Copy full SHA for 72f0e76

Commits on May 28, 2025

• Add -O0 to UM and remove memory checking from everything

  manodeep

  committed
  425c66d

  Copy full SHA for 425c66d

Commits on May 29, 2025

• Changed UM to undo phenology changes but with oneapi compile failure fixes

  manodeep

  committed
  488e4db

  Copy full SHA for 488e4db

• Explicitly initialising arrays to 0 in mom5

  manodeep

  committed
  4f3f48f

  Copy full SHA for 4f3f48f

• Remove -O0 to build with the previous commit

  manodeep

  committed
  dd65dfa

  Copy full SHA for dd65dfa

• Build with the file-local nooptimise directive

  manodeep

  committed
  4918359

  Copy full SHA for 4918359

Commits on May 30, 2025

• Checking whether dummy aliases flag fixes

  manodeep

  committed
  8618e3c

  Copy full SHA for 8618e3c

• Remove dummy_aliases; compile with the print statement within the offending if condition with the divide-by-zero

  manodeep

  committed
  01b39d1

  Copy full SHA for 01b39d1

• over-riding the default -O3 with -O2

  manodeep

  committed
  b8a18c8

  Copy full SHA for b8a18c8

• Trying to re-use the previous builds with hashes

  manodeep

  committed
  7938736

  Copy full SHA for 7938736

• Change to -O1 for mom5

  manodeep

  committed
  04962a5

  Copy full SHA for 04962a5

• Revert back to the regular flags

  manodeep

  committed
  5159ba8

  Copy full SHA for 5159ba8

Commits on May 31, 2025

• Make sure there is one source for cice4

  manodeep

  committed
  785fb77

  Copy full SHA for 785fb77

Commits on Jun 2, 2025

• Building with the new mom5 that prevents vectorisation in the two offending for loops. Also removed the -O1 flag

  manodeep

  committed
  5e50f56

  Copy full SHA for 5e50f56

• Building with the updated code that protect the divides by adding into the if condition.

  manodeep

  committed
  b41f693

  Copy full SHA for b41f693

Commits on Jun 3, 2025

• Update versions of spack-packages, access-fms and access-generic-tracers

  dougiesquire

  committed
  f387236

  Copy full SHA for f387236

• Check whether the troubles go away with the newer version

  manodeep

  committed
  a9dc871

  Copy full SHA for a9dc871