Code

Gpu code#58

Code

Open

abouzied-nasar wants to merge 170 commits intoSWIFTSIM:masterSWIFTSIM/SWIFT:masterfrom abouzied-nasar:gpu_codeabouzied-nasar/SWIFT:gpu_codeCopy head branch name to clipboard

+19,460-78

Commits

Commits on Oct 21, 2024

• Added required changes to configure.ac to allow comp w cuda & hip. Added cuda and hip Makefile.ac

  abouzied-nasar

  committed
  b16a6ff

  Copy full SHA for b16a6ff

• Added cuda and hip linking directives to Makefile.am

  abouzied-nasar

  committed
  d9ca273

  Copy full SHA for d9ca273

• Added AC_PROG_CXX to configure.ac

  abouzied-nasar

  committed
  ecef19e

  Copy full SHA for ecef19e

• Added first GPU files: src/runner_gpu_pack_functionc.c and src/cuda/part_gpu.h

  abouzied-nasar

  committed
  5cb3705

  Copy full SHA for 5cb3705

• ACTUALLY added first GPU files: src/runner_gpu_pack_functionc.c and src/cuda/part_gpu.h

  abouzied-nasar

  committed
  4edf58e

  Copy full SHA for 4edf58e

Commits on Oct 22, 2024

• Added more files for GPU code. Seems to work fine aside from config.h not being found by the cuda library

  abouzied-nasar

  committed
  0c90ab5

  Copy full SHA for 0c90ab5

• Added ifdefs to a few files to a) Stop can't find config.h errors and b to stop an error related to prototype overloading in dummy.c in src/cuda

  abouzied-nasar

  committed
  f19c35a

  Copy full SHA for f19c35a

• Added dummy.c in src/cuda

  abouzied-nasar

  committed
  6334924

  Copy full SHA for 6334924

Commits on Oct 23, 2024

• Added cudalt.py dummy.C src/cuda/tester.cu

  abouzied-nasar

  committed
  8b64cbb

  Copy full SHA for 8b64cbb

Commits on Oct 24, 2024

• Added code to cell.c and h cell_hydro.c cell_unskip.c engine_config.c task.c and h queue.h and removed unused vars from runner_doiact_functions_hydro_gpu.h

  abouzied-nasar

  committed
  a822b0b

  Copy full SHA for a822b0b

• Added code to engine.c

  abouzied-nasar

  committed
  b0193cb

  Copy full SHA for b0193cb

• Removed bug from engine.c and added some code to scheduler.c

  abouzied-nasar

  committed
  391d2bf

  Copy full SHA for 391d2bf

• added code to scheduler.* engine_unskip.c

  abouzied-nasar

  committed
  9bd433d

  Copy full SHA for 9bd433d

• I had made a mistake by putting runner_doiact_functions_hydro_gpu.h in main dir instead of src

  abouzied-nasar

  committed
  8d0d437

  Copy full SHA for 8d0d437

• Sorted most of the code out. Compiles and runs fine with gpu offload off because I have not made gpu tasks yet

  abouzied-nasar

  committed
  caa0852

  Copy full SHA for caa0852

Commits on Oct 25, 2024

• Added code to engine_marktasks.c

  abouzied-nasar

  committed
  294f7ae

  Copy full SHA for 294f7ae

• All coded up but there seems to be a problem with duplicate unlocks. Not sure where the issue is

  abouzied-nasar

  committed
  61c83a1

  Copy full SHA for 61c83a1

Commits on Oct 28, 2024

• Made some changes here and there to try and get deps right for unpack tasks. Code now hangs so there must be some issue with task activation. problem for tomorrow!

  abouzied-nasar

  committed
  21ed5cd

  Copy full SHA for 21ed5cd

Commits on Oct 29, 2024

• Commented out GPU code from engine_marktasks.c to see if that could have been source of hanging bug. Not the case :(

  abouzied-nasar

  committed
  aa3eeab

  Copy full SHA for aa3eeab

• Removed duplicate engine_addlink for g and f pack tasks. And re-wired deps for unpack tasks. CPU version works perfectly but GPU code now hangs for some reason

  abouzied-nasar

  committed
  9c1b494

  Copy full SHA for 9c1b494

• Minor changes here and there

  abouzied-nasar

  committed
  5e08222

  Copy full SHA for 5e08222

Commits on Oct 30, 2024

• Found a bug in task.c -> Wasn't unlocking gradient pack task

  abouzied-nasar

  committed
  af0d256

  Copy full SHA for af0d256

• Found a bug in task.c -> Wasn't unlocking gradient pack task

  abouzied-nasar

  committed
  6551027

  Copy full SHA for 6551027

• Code still hanging. Will try starting from scratch with runner_main_clean.cu

  abouzied-nasar

  committed
  1dc8451

  Copy full SHA for 1dc8451

• Copied over both runner_main_clean and runner_doiact_functions_hydro_gpu and re-worked runner_main_clean.cu but code still hangs

  abouzied-nasar

  committed
  8103149

  Copy full SHA for 8103149

• Copied over both runner_main_clean and runner_doiact_functions_hydro_gpu and re-worked runner_main_clean.cu but code still hangs

  abouzied-nasar

  committed
  badc9d4

  Copy full SHA for badc9d4

• Issue was not with #ifdefs in runner_main_clean.cu or runner_doiact... Removed them all and code still hangs. Will copy back previous files innext commit

  abouzied-nasar

  committed
  45ea651

  Copy full SHA for 45ea651

• Issue is probably with how I am locking and unlocking tasks or something to with not unskipping tasks properly. IFDEFs are totally correct and code does not hang when GPU code commented out

  abouzied-nasar

  committed
  a9f81dd

  Copy full SHA for a9f81dd

Commits on Oct 31, 2024

• signalling sleeping runners just after packing seems to prevent hanging. Code is doing something but I don't think it's actually progressing through time steps

  abouzied-nasar

  committed
  2cc07c5

  Copy full SHA for 2cc07c5

Commits on Nov 1, 2024

• Testing to see if code still hangs when making deps on pack tasks instead of unpack tasks

  abouzied-nasar

  committed
  4d6fe3d

  Copy full SHA for 4d6fe3d

• Added scheduler_done to runner_doiact_functions_hydro_gpu.h. Also commented out code for packing in runner_main_clean.cu so we only offload seld density tasks for testing

  abouzied-nasar

  committed
  57d77a2

  Copy full SHA for 57d77a2

• Checked engine_maketasks.c and things seem reasonable with nothing missing. Issue maybe in cell_unskip.c

  abouzied-nasar

  committed
  d952965

  Copy full SHA for d952965

• Fixed a bug in scheduler_enqueue() where I had a break before where it should've been. Replaced with if statement

  abouzied-nasar

  committed
  13912ce

  Copy full SHA for 13912ce

Commits on Nov 4, 2024

• Removed all GPU tasks aside from density self pack tasks. COde still hangs

  abouzied-nasar

  committed
  3632ba2

  Copy full SHA for 3632ba2

• Found bug in how we set n_tasks_left* in scheduler_rewait it should be in scheduler_enqueue

  abouzied-nasar

  committed
  d77dd2f

  Copy full SHA for d77dd2f

• Fix is in for force and gradient pack tasks but needs de-bugging as code is not giving correct results

  abouzied-nasar

  committed
  db43727

  Copy full SHA for db43727

Commits on Nov 5, 2024

• Missing brackets. FIX!

  abouzied-nasar

  committed
  abecded

  Copy full SHA for abecded

• Fix bracketting in MPI hydro recv construction

  MatthieuSchaller

  committed
  e8aa0ae

  Copy full SHA for e8aa0ae

• Fixed missing closing curly

  MatthieuSchaller

  committed
  6aa324d

  Copy full SHA for 6aa324d

• Applied code formatting script blindly

  MatthieuSchaller

  committed
  5e4eafe

  Copy full SHA for 5e4eafe

• Added new ifdef-controls to offload only the density/gradient/force hydro tasks to the GPU

  MatthieuSchaller

  committed
  780553d

  Copy full SHA for 780553d

• Added new ifdef-controls to offload only the density/gradient/force hydro tasks to the GPU

  MatthieuSchaller

  committed
  3b58f55

  Copy full SHA for 3b58f55

• Fix another bracketing issue

  MatthieuSchaller

  committed
  81c2283

  Copy full SHA for 81c2283

• Fix another bracketing issue

  MatthieuSchaller

  committed
  b21d912

  Copy full SHA for b21d912

• Fix logic mistake

  MatthieuSchaller

  committed
  b4f4203

  Copy full SHA for b4f4203

• First attempt at MPI dependencies

  MatthieuSchaller

  committed
  4a90516

  Copy full SHA for 4a90516

• Changed fprints to message in engine_maktasks.c

  abouzied-nasar

  committed
  d212364

  Copy full SHA for d212364

• Fix comm name typo

  MatthieuSchaller

  committed
  7c3728d

  Copy full SHA for 7c3728d

• Only scream if a local cell has no ghost_in

  MatthieuSchaller

  committed
  af8bc7b

  Copy full SHA for af8bc7b

• Add unpack ---> recv dependencies

  MatthieuSchaller

  committed
  732370d

  Copy full SHA for 732370d

• Fix error message in the task construction

  MatthieuSchaller

  committed
  c43b7b5

  Copy full SHA for c43b7b5

• Only create the timing files if we are dumping timings. Assign the devices with the mpi ranks.

  MatthieuSchaller

  committed
  e64e17a

  Copy full SHA for e64e17a

• Put the gpu pack/unpack tasks in the correct category

  MatthieuSchaller

  committed
  9981039

  Copy full SHA for 9981039

• Do not update the total task time in signal_sleeping_runners

  MatthieuSchaller

  committed
  dd041fe

  Copy full SHA for dd041fe

Commits on Nov 6, 2024

• Time the pack and unpack of density pair separately

  MatthieuSchaller

  committed
  e506c79

  Copy full SHA for e506c79

• Fix typos

  MatthieuSchaller

  committed
  3ccbaf0

  Copy full SHA for 3ccbaf0

• Added timers for packing all GPU task types

  abouzied-nasar

  committed
  707ad45

  Copy full SHA for 707ad45

• Added timers for unpacking all GPU task types

  abouzied-nasar

  committed
  425b074

  Copy full SHA for 425b074

• modified scheduler_report_task_times_mapper() to account for different pack and unpack tasks d/f/g. There is a bug in runner_gpu_pack_functions.c when enabling timers

  abouzied-nasar

  committed
  4634905

  Copy full SHA for 4634905

• Reverted back to one if statement only for pack tasks as unpack tasks do nothing

  abouzied-nasar

  committed
  7b8baa6

  Copy full SHA for 7b8baa6

• made pack task types set to task_category_gpu

  abouzied-nasar

  committed
  23cf181

  Copy full SHA for 23cf181

• First attempt at re-instating task stealing

  MatthieuSchaller

  committed
  d7f980f

  Copy full SHA for d7f980f

• Put varaible back in the code

  MatthieuSchaller

  committed
  5b67cd4

  Copy full SHA for 5b67cd4

Commits on Nov 7, 2024

• put in fix for swift_task_debug in runner_main_clean.cu

  abouzied-nasar

  committed
  51c788a

  Copy full SHA for 51c788a

Commits on Nov 8, 2024

• Fix code to allow running with SWIFT_DEBUG_TASKS

  MatthieuSchaller

  committed
  ddc843b

  Copy full SHA for ddc843b

• Fix typo

  MatthieuSchaller

  committed
  0447148

  Copy full SHA for 0447148

• Applied formatting script

  MatthieuSchaller

  committed
  037dbd5

  Copy full SHA for 037dbd5

• Commented out task splitting for hydro and added fix for cell-less tasks in split_tasks for gravity

  abouzied-nasar

  committed
  7eab151

  Copy full SHA for 7eab151

• Call signal_sleeping_runners() in runner_main and not inside the pack task

  MatthieuSchaller

  committed
  55c50b2

  Copy full SHA for 55c50b2

Commits on Nov 9, 2024

• Changed the fix for duplicate unlocks so that we skip checking for duplicate unpack tasks unlocking other tasks e.g. unpack->ghost (not the other way around)

  abouzied-nasar

  committed
  3800f53

  Copy full SHA for 3800f53

• Modified signal_sleeping_runners to only signal once per pack. Also moved enqueue_deps to be executed as each pack task is done. Seems to work fine. Need to test if dead time is reduced by this

  abouzied-nasar

  committed
  40b67ee

  Copy full SHA for 40b67ee

Commits on Nov 11, 2024

• Setup and tested for optimal pack_size. GPU code is about 20% fatser than previously with excessive calls to signal_sleeping_runners but dead time is roughly the same

  

  Abouzied Nasar

  committed
  339539d

  Copy full SHA for 339539d

• Changed part_gpu.h to compile on Bede

  

  Abouzied Nasar

  committed
  642b4de

  Copy full SHA for 642b4de

• Made duplicate CPU tasks implicit and switched back to enqueueing dependencies ONLY after they have been actually done on the GPU

  abouzied-nasar

  committed
  9664fa8

  Copy full SHA for 9664fa8

Commits on Nov 12, 2024

• Commented out debug code in cell_unskip.c, come back and put code in SWIFT_DEBUG_CHECKS ifdefs. Also removed un-necessary check in runner_doiact_functions_hydro_gpu.h

  abouzied-nasar

  committed
  03d15b1

  Copy full SHA for 03d15b1

• Modified cell_unskip.c so GPU debug bits only active when SWIFT_DEBUG_CHECKS is on

  abouzied-nasar

  committed
  66883bb

  Copy full SHA for 66883bb

• Added some files for HIP compilations

  abouzied-nasar

  committed
  fb531da

  Copy full SHA for fb531da

Commits on Nov 19, 2024

• Quick stab at splitting GPU tasks. May have to revert as done in haste

  abouzied-nasar

  committed
  69d7d47

  Copy full SHA for 69d7d47

• Fixed a few bugs with GPU task splitting. Code hangs after a few steps when using low thread count and almost immediately if using many threads

  abouzied-nasar

  committed
  72fa9d3

  Copy full SHA for 72fa9d3

• Fixed a bug in if statement in runner_main2(). Added some code to make GPU tasks implicit in engine_maketasks(). CPU code seems to run fine without hanging. Need to figure out what issue is with th…

  Show description for 83972f3
  abouzied-nasar

  committed
  83972f3

  Copy full SHA for 83972f3

• Found another bug. I was not accounting for split tasks when allowing task stealing

  abouzied-nasar

  committed
  d250edf

  Copy full SHA for d250edf

Commits on Nov 20, 2024

• Converted sub_selfs and sub_pairs to selfs and pairs in maketasks. Code still hangs when using more than 4 threads I think issue is in engine_make_extra_hydroloop_tasks_mapper()

  abouzied-nasar

  committed
  cb7abcf

  Copy full SHA for cb7abcf

• Removed sub tasks from runner_main and moved making density sub tasks into self and pairs to after creation of sub tasks for gradient and force

  abouzied-nasar

  committed
  2ee5742

  Copy full SHA for 2ee5742

• commented out sub tasks in runner_main. Edited how tasks are created

  abouzied-nasar

  committed
  4d726d1

  Copy full SHA for 4d726d1

• Made sub tasks implicit and fixed bug with atomic_dec in runner_main

  abouzied-nasar

  committed
  dcbad17

  Copy full SHA for dcbad17

• reverted to converting subs

  abouzied-nasar

  committed
  2beb67e

  Copy full SHA for 2beb67e

• Too tired to carry on. There might be something to playing with space_subsize_pair/self_hydro_default

  abouzied-nasar

  committed
  2a0d137

  Copy full SHA for 2a0d137

Commits on Nov 21, 2024

• A few edits to test on Bede GHs

  abouzied-nasar

  committed
  4bfa575

  Copy full SHA for 4bfa575

• Changed a few parameters

  abouzied-nasar

  committed
  50c6a74

  Copy full SHA for 50c6a74

• Removed redundnat commented out code from engine_maketasks()

  abouzied-nasar

  committed
  d808695

  Copy full SHA for d808695

• before adding time

  abouzied-nasar

  committed
  18513c9

  Copy full SHA for 18513c9

• Chnaged task_unlocks to cell_unlocktree. Modified timer function for hanging in runner_main

  abouzied-nasar

  committed
  306d32e

  Copy full SHA for 306d32e

• Remved unnecessary counter from engine_maketasks

  abouzied-nasar

  committed
  3e0c7b8

  Copy full SHA for 3e0c7b8

• Changed how s->nr_self_pack_tasks is incremented in scheduler.c

  abouzied-nasar

  committed
  fe5d945

  Copy full SHA for fe5d945

• Commented out timing code in runner_main. The code ran for a full simulation so edit from prev commit seems to have done the trick.

  abouzied-nasar

  committed
  744c469

  Copy full SHA for 744c469

Commits on Dec 14, 2024

• Made changes to task dump for debugging to account for (skip) cell-less unpack tasks in some loop. Allowed stealing (broken and code hangs). Changed if statements in scheduler steal function

  abouzied-nasar

  committed
  5b54b0f

  Copy full SHA for 5b54b0f

• Added comments to remin myself to threadpool some bits in GPU task mgmt in engine_maketasks.c

  abouzied-nasar

  committed
  7a168d5

  Copy full SHA for 7a168d5

Commits on Dec 15, 2024

• Added code to pass eta_neighbours (h/dx) to GPU code for calculating ideal N parts per task

  abouzied-nasar

  committed
  f7bf486

  Copy full SHA for f7bf486

Commits on Dec 16, 2024

• Removed conversion of sub-pairs to pairs, etc, in engine_maketasks for GPU tasks. Added some debug checks here and there. For some reason I get cells with way more particles than anticipated even t…

  Show description for ff7087c
  abouzied-nasar

  committed
  ff7087c

  Copy full SHA for ff7087c

Commits on Dec 17, 2024

• Added more debug checks to figure out why we got cells with 8x expected particles. Also added a modified scheduler_splittasks() function to split GPU tasks but still no joy. Could it be that some c…

  Show description for ca9deaf
  abouzied-nasar

  committed
  ca9deaf

  Copy full SHA for ca9deaf

• Hard-coded space_splitsize_default to 100 with no avail. Something fishy is going on...

  abouzied-nasar

  committed
  408b0e4

  Copy full SHA for 408b0e4

• Hard-coded space_grid_split_threshold_default to 100 and space_subsize_pair_hydro_default with no avail. Something fishy is going on...

  abouzied-nasar

  committed
  fa1e7c4

  Copy full SHA for fa1e7c4

Commits on Dec 18, 2024

• I was barking up the wrong tree. The issue was that some particles will have a slightly larger smoothing length that should not be in the ideal sized box. Setting count_max_parts_tmp to 2x bigger s…

  Show description for 26585ec
  abouzied-nasar

  committed
  26585ec

  Copy full SHA for 26585ec

• Added some comments

  abouzied-nasar

  committed
  4063e37

  Copy full SHA for 4063e37

• Okay. Now back to stealing. Code hangs when stealing enabled

  abouzied-nasar

  committed
  fd849fc

  Copy full SHA for fd849fc

• Testing to see what happens when only self-dens tasks are done on GPU. Code works fine and does not hang with stealing enabled. Issue is with pair tasks...

  abouzied-nasar

  committed
  fd9cfa0

  Copy full SHA for fd9cfa0

Commits on Dec 19, 2024

• Reverted back to offloading only density tasks (self and pair)

  abouzied-nasar

  committed
  e9cb4dd

  Copy full SHA for e9cb4dd

• Weirdly code does not hang when using bigger test case and only density pairs and selfs, in comparison to same setup but with smaller test case

  abouzied-nasar

  committed
  e94fe6e

  Copy full SHA for e94fe6e

• Reverted to doing all tasks on GPU. Need to test on GHopper

  abouzied-nasar

  committed
  0e3196d

  Copy full SHA for 0e3196d

• Deleted obsolete debug code

  abouzied-nasar

  committed
  3dfb81c

  Copy full SHA for 3dfb81c

• Deleted obsolete debug code

  abouzied-nasar

  committed
  57cb5d9

  Copy full SHA for 57cb5d9

• Put in a weird fix in scheduler_gettask but I think it is wrong and should be replaced with a continue; if the task in unpack

  abouzied-nasar

  committed
  02e18e4

  Copy full SHA for 02e18e4

• Deleted commented out obsolete code

  abouzied-nasar

  committed
  153558e

  Copy full SHA for 153558e

• Changed condition to launch_leftovers to if leftover tasks <=1 instead of ==0 for when stealing is active

  abouzied-nasar

  committed
  f71201e

  Copy full SHA for f71201e

• Changed condition to launch_leftovers to if leftover tasks <=1 instead of ==0 for when stealing is active for selfs too. Code still hangs but after 60-70 steps. Not perfect but good enough for the …

  Show description for 707928b
  abouzied-nasar

  committed
  707928b

  Copy full SHA for 707928b

• Extended <2 condition to other GPU task subtypes but the code now hangs. Need to double check if I put a bug in somewhere or if it is just not good enough to let me run a few steps for the paper

  abouzied-nasar

  committed
  c5716b6

  Copy full SHA for c5716b6

Commits on Dec 23, 2024

• Minor changes for testing. Commented out debug code for checking if tasks not split to target level for GPU

  abouzied-nasar

  committed
  b78ba84

  Copy full SHA for b78ba84

• Reverted condition for launch leftovers to n_packs_left < 1 instead of < 2. Removed some commented out code in runner_main. Edited scheduler_gettask so stealing code is a bit cleaner when stealing …

  Show description for b533d16
  abouzied-nasar

  committed
  b533d16

  Copy full SHA for b533d16

Commits on Dec 30, 2024

• Allowed stealing engine_config. Changed debug if statement to error rather than exit(0). Removed if statment querying for n tasks left in scheduler from within runner_main as it was causing kernel …

  Show description for ceb4d6f
  abouzied-nasar

  committed
  ceb4d6f

  Copy full SHA for ceb4d6f

• Added simple debug code to scream when we have way more particles than anticipated. The code works fine for h=1.9dx and lower but we end up with 8x particles per cell if we set h=1.99dx due to some…

  Show description for 4929cf8
  abouzied-nasar

  committed
  4929cf8

  Copy full SHA for 4929cf8

• Added code to monitor how many GPU tasks a thread steals. Code still hangs with s->waiting greater than zero and the dumper thread shows that some unpacks are waiting (for packs) and some end_force…

  Show description for 392e2ce
  abouzied-nasar

  committed
  392e2ce

  Copy full SHA for 392e2ce

Commits on Jan 6, 2025

• Very weird behaviour found. Added some debug code set to crash run if launch_leftovers = 1 whilst the number of tasks left is equal or greater to 1 and the code sets launc_leftovers to 1 even when …

  Show description for 29e074b
  abouzied-nasar

  committed
  29e074b

  Copy full SHA for 29e074b

Commits on Jan 7, 2025

• Moving tasks_left counter incrementation from scheduler.c to queue_insert() causes counters to go negative. Code hangs when changing condition to launch_leftovers to if (tasks_left == 0)

  abouzied-nasar

  committed
  949fd47

  Copy full SHA for 949fd47

• Changed if statements to atomic_CAS. Code hangs but the CAS does not seem to be working as intended. Probably incorrect usage of the CAS

  abouzied-nasar

  committed
  5abe90b

  Copy full SHA for 5abe90b

• Changed comp val to int in CAS instead of int *. Removed debug code. CAS seems to work

  abouzied-nasar

  committed
  5082ca9

  Copy full SHA for 5082ca9

• Removed debug code which throws error if launching leftovers but still have GPU tasks in the queue

  abouzied-nasar

  committed
  38953c4

  Copy full SHA for 38953c4

• Combined atomic dec and CAS calls into one operation. Code seems to run for longer before hanging. Need to investigate why!

  abouzied-nasar

  committed
  c7e6a6c

  Copy full SHA for c7e6a6c

Commits on Jan 8, 2025

• Removed un-necesseary mods from swift.c

  abouzied-nasar

  committed
  f3e3138

  Copy full SHA for f3e3138

• Added comment to explain changes in task.c

  abouzied-nasar

  committed
  a58d968

  Copy full SHA for a58d968

• Removed un-necessary code from space.h

  abouzied-nasar

  committed
  4513dca

  Copy full SHA for 4513dca

• Reverted swift.c and space.h as eta_neighbours was indeed required for GPU memory allocation. Removed packs_stolen counter update from scheduler_gettask(). Cleaned up scheduler.c

  abouzied-nasar

  committed
  cf16ae4

  Copy full SHA for cf16ae4

• Reverted swift.c and space.h as eta_neighbours was indeed required for GPU memory allocation. Removed packs_stolen counter update from scheduler_gettask(). Cleaned up scheduler.c

  abouzied-nasar

  committed
  7871c95

  Copy full SHA for 7871c95

• Cleaned runner_main a bit whilst checking for source of new code hang issue

  abouzied-nasar

  committed
  3bf586e

  Copy full SHA for 3bf586e

• Cleaned up un-necessary code from queue.c

  abouzied-nasar

  committed
  091819d

  Copy full SHA for 091819d

Commits on Jan 9, 2025

• Reverted back to signalling runners individually for each task which fixed code hanging without task stealing. Tried out a few ideas to try and prevent code from hanging with stealing ON to no avail

  abouzied-nasar

  committed
  ea17e54

  Copy full SHA for ea17e54

• Added additional switch in runner_main to double check whether to launch_leftovers. Seems to run for slightly longer but still hangs. Need to test on GH to see if I can get away with this code for …

  Show description for 4ab67c2
  abouzied-nasar

  committed
  4ab67c2

  Copy full SHA for 4ab67c2

Commits on Jan 10, 2025

• Code hanging. Unsure why. Made significant changes. Changed counters so that they are an array attached to scheduler struct where the array indices represent the scheduler's queues. The code hangs …

  Show description for f8463e8
  abouzied-nasar

  committed
  f8463e8

  Copy full SHA for f8463e8

Commits on Jan 27, 2025

• Reverted back to decrementing counters attached to queues. Code still hangs but at least runs for a few steps

  abouzied-nasar

  committed
  0573903

  Copy full SHA for 0573903

• Make queue debug variables volatile. More uniform naming convention. Replace atomic operations on queue counters by queue locking operations

  

  Abouzied Nasar

  committed
  8dc1b50

  Copy full SHA for 8dc1b50

• Changed dec and inc ops when stealing to be non-atomic. Also replaced counters ending with sched->nr_pair_pack_tasks to sched->nr_pair_pack_tasks_d in engine_maketasks.c

  abouzied-nasar

  committed
  73fd85f

  Copy full SHA for 73fd85f

• Repeated changing atomic_incs to regular increment within queue lock regions in scheduler enqueue function

  abouzied-nasar

  committed
  3e08973

  Copy full SHA for 3e08973

• Repeated the same for atomic_decs in runner_doiact_functions_hydro_gpu

  abouzied-nasar

  committed
  3a71e3f

  Copy full SHA for 3a71e3f

• Applied code formatting script

  MatthieuSchaller

  committed
  b4c57bf

  Copy full SHA for b4c57bf

• Rename the density pack and unpack tasks such that they have a _d suffix to match the gradient and force ones

  MatthieuSchaller

  committed
  532e06f

  Copy full SHA for 532e06f

• Fix compiler warnings

  MatthieuSchaller

  committed
  214e172

  Copy full SHA for 214e172

Commits on Jan 28, 2025

• Fix missing changes to task names in runner_main()

  abouzied-nasar

  committed
  2a4ec93

  Copy full SHA for 2a4ec93

• Make git ignore the new executables

  abouzied-nasar

  committed
  38ab772

  Copy full SHA for 38ab772

• Correct executable name

  abouzied-nasar

  committed
  88c4f2d

  Copy full SHA for 88c4f2d

Commits on Jan 31, 2025

• Corrected one of the silly exit(0)s in src/runner_doiact_functions_hydro_gpu.h

  abouzied-nasar

  committed
  3a279b3

  Copy full SHA for 3a279b3

• Added if statement to stop code launching kernels for cells with no particles

  abouzied-nasar

  committed
  2f60860

  Copy full SHA for 2f60860

• Added the same if statement to self tasks

  abouzied-nasar

  committed
  9cb3d35

  Copy full SHA for 9cb3d35

• Extended (if(numBlocks_x > 0)) to all cuda calls as code was still crashing due to memcpys

  abouzied-nasar

  committed
  cf3b019

  Copy full SHA for cf3b019

• Added if statements in runner_main2() so that if cell hydro count is zero just call scheduler_done()

  abouzied-nasar

  committed
  7ba6d3d

  Copy full SHA for 7ba6d3d

• Fixed bug in implementation of pair skip condition when count == 0. Fixed counter decrementation in scheduler_enqueue for the case when cell->hydro.count == 0

  abouzied-nasar

  committed
  7afbe56

  Copy full SHA for 7afbe56

• Reverted runner_main2 not to skip tasks with zero particles as they are now not activated in cell_unskip

  abouzied-nasar

  committed
  fa013d2

  Copy full SHA for fa013d2

• Converting sub tasks into selfs and pairs in engine_maketasks

  abouzied-nasar

  committed
  a662bb0

  Copy full SHA for a662bb0

• Removed blocking if statements from launch and packing code in runner_doiact_functions_hydro_gpu.h

  abouzied-nasar

  committed
  6ba1c4e

  Copy full SHA for 6ba1c4e

• Skipping task activation for GPU tasks without particles

  abouzied-nasar

  committed
  6be7e2e

  Copy full SHA for 6be7e2e

• Modified .yml files for gresho and islated galaxy. In isolated_galaxy.yml set the max time step to 1e-6 to prevent code crashing when there are only a few particle updates. Need to modify GPU code …

  Show description for 53fc479
  abouzied-nasar

  committed
  53fc479

  Copy full SHA for 53fc479

• Added code to prevent update of inactive particles during unpacking

  abouzied-nasar

  committed
  d7806bd

  Copy full SHA for d7806bd

• Stopped the formation of star particles. Changed cell_unskip so that only gpu tasks are not activated if they have zero particles

  abouzied-nasar

  committed
  037c469

  Copy full SHA for 037c469

Commits on Feb 5, 2025

• Minor change to gresho.yml

  abouzied-nasar

  committed
  243030b

  Copy full SHA for 243030b

• Removed excessive allocation of GPU and Buffer arrays. Back to 2x instead of 20x

  abouzied-nasar

  committed
  00a53d6

  Copy full SHA for 00a53d6

• Reverted space_splitsize_default to 400

  abouzied-nasar

  committed
  f083d6b

  Copy full SHA for f083d6b

• Small changes: Increased count_max_parts_tmp to 8x for now. Edited debug message

  abouzied-nasar

  committed
  1871a08

  Copy full SHA for 1871a08

• Nearly done with fixing splitting. Apparently I only fixed it for density tasks in previous commit. I am now extending the revised splitting strategy to gradient and force subtypes

  abouzied-nasar

  committed
  8ab4d45

  Copy full SHA for 8ab4d45

• Implemented splitting correctly. Code gives correct results and dependency graphs are looking good. Split domain with 8^3 cells into 16^3 cells with no issues

  abouzied-nasar

  committed
  0901a73

  Copy full SHA for 0901a73

• Tested and code works better for case with ~512 parts cell in comparison to with ideal size of ~64 parts per cell and 2 million parts. Will now test on G-H Isamabrd-AI

  abouzied-nasar

  committed
  9b214a1

  Copy full SHA for 9b214a1

Commits on Feb 6, 2025

• Added counters to double check how many cells have target part number and how many have more

  abouzied-nasar

  committed
  ecb9489

  Copy full SHA for ecb9489

• Modified the debugging code in prev. commit to track max number of particles in a task

  abouzied-nasar

  committed
  81361f4

  Copy full SHA for 81361f4