ICLDisco/parsec

Issues

• Provide API to map real device ID to parsec ID and back

  #660 opened 7 months ago by devreal
  5

• Linker and CC flags are incompatible on OSX

  #662 opened 7 months ago by abouteiller
  0

• Direct GPU-GPU inter-node data transfer

  #661 opened 7 months ago by abouteiller
  0

• The CI configuration should request and use slurm tags to control what tests run and compile

  #659 opened 8 months ago by abouteiller
  0

• No data movement to GPU when data from desc with type keyword

  #643 opened 10 months ago by QingleiCao
  7

• stress:gpu crashes when compiled with CUDA but run without a device

  #641 opened 8 months ago by abouteiller
  4

• Release prep: v4.0 supports TTG and Hicma

  #638 opened 8 months ago by abouteiller
  3

• Overflow when creating data if tile size is large

  #645 opened 8 months ago by QingleiCao
  1

• jdf2c issue when dealing with certain case

  #647 opened 10 months ago by QingleiCao
  1

• MPI_AND_HWLOC_COMPATIBLE needs some optional deactivation

  #511 opened 9 months ago by therault
  3

• Increment on volatile variables in deprecated

  #588 opened a year ago by devreal
  4

• Check dplasma ctest pass

  #619 opened a year ago by abouteiller
  1

• zone_malloc should handle 0-sized allocations gracefully

  #649 opened 9 months ago by devreal
  1

• Memory leaks with the DTD interface

  #640 opened 10 months ago by BrieucNicolas
  3

• freeing a newly created dtd taskpool causes parsec to segfault

  #568 opened 10 months ago by BrieucNicolas
  0

• Broadcast incorrect under certain circumstances, causing segfaults

  #624 opened a year ago by omor1
  1

• Provide CMake target for accelerator support

  #571 opened a year ago by devreal
  1

• remote_multiple_outs_same_preds can deadlock

  #628 opened a year ago by abouteiller
  1

• GPU memory allocation far exceeds requested size

  #630 opened a year ago by abouteiller
  3

• When GPU cannot initialize (OOM) per-stream-info need cleanup

  #636 opened a year ago by abouteiller
  0

• Termination detection fault with dtd

  #634 opened a year ago by abouteiller
  2

• Support for oversubscription broken

  #601 opened a year ago by devreal
  1

• MCA parameters that cannot be understood should cause warnings/errors

  #625 opened a year ago by abouteiller
  3

• DTD GPU enabled crashes when GPU are runtime disabled

  #627 opened a year ago by abouteiller
  0

• Update Copyrights/authors before v4.0

  #618 opened a year ago by abouteiller
  0

• CI testers get_best_device, stage, etc fail

  #552 opened a year ago by abouteiller
  1

• Missing header file inclusion directive in parsec/utils/process_name.c for Apple system.

  #610 opened a year ago by bimalgaudel
  0

• Lifo ctest fails

  #586 opened a year ago by abouteiller
  6

• Parsec context shared between multiple libraries

  #589 opened a year ago by therault
  1

• MAX_DEP_IN_COUNT and MAX_DEP_OUT_COUNT

  #579 opened a year ago by QingleiCao
  0

• Issue of MPI_AND_HWLOC_COMPATIBLE on some systems

  #580 opened a year ago by QingleiCao
  0

• parsec_get_best_device, parsec_gpu_data_stage_in and NEW tiles requires a fix

  #564 opened 2 years ago by therault
  0

• Support for short data lifetime in device code

  #560 opened 2 years ago by devreal
  0

• CMake 3.26 fails due to missing CUDA_ARCHITECTURES

  #555 opened 2 years ago by devreal
  4

• Recursive + GPU seems broken

  #548 opened 2 years ago by therault
  4

• tests mislabeled again

  #547 opened 2 years ago by abouteiller
  1

• Recursive tasks are scheduled on one queue

  #551 opened 2 years ago by omor1
  1

• Does PaRSEC build a DAG and then schedule task or schedule tasks right after tasks are inserted?

  #516 opened 2 years ago by cheng-hsiang-chiu
  13

• Trivial parsec test fails

  #545 opened 2 years ago by therault
  5

• Dynamic-length active message request

  #543 opened 2 years ago by therault
  0

• Recursive Termdet is broken

  #541 opened 2 years ago by abouteiller
  1

• Number of recv-send requests should be runtime configurable (MCA)

  #539 opened 2 years ago by abouteiller
  0

• `profile=off` tasks are traced anyway, sometimes causing segfault

  #529 opened 2 years ago by DSMishler
  3

• Deadlock on GPU when using NEW and matrix size is large

  #527 opened 2 years ago by QingleiCao
  1

• GPU Task Profiling (in PTG)

  #520 opened 2 years ago by therault
  3

• Improve data reuse on GPU

  #514 opened 2 years ago by QingleiCao
  0

• datatype conversion on the node level

  #513 opened 2 years ago by QingleiCao
  0

• datatype conversion on the runtime level

  #512 opened 2 years ago by QingleiCao
  0

• GPU Statistics is wrong

  #505 opened 2 years ago by QingleiCao
  1

• profile2h5 out of memory for large traces

  #507 opened 2 years ago by DSMishler
  3