1623 add training #3145

hiker · 2025-09-21T13:21:44Z

Adds the PSyclone training material (based on #3131)

… now inlines.

…i atm).

…lass

…_training

arporter

I'm enjoying going through (in outline) your game-of-life example - it's great to see that you can recover performance :-)
Since most of the example transformation scripts need their module and apply-method docstrings updating, I'm going to pause here to let you do that.
As you suggested, I'm basically ignoring the actual text of the readme's so I am by no means reviewing the actual tutorials.

tutorial/training/Makefile

tutorial/training/README.rst

tutorial/training/gocean/2.8-GameOfLife-openmp/fuse_loops.py

tutorial/training/gocean/2.8-GameOfLife-openmp/fuse_loops_last.py

tutorial/training/gocean/2.8-GameOfLife-openmp/openmp.py

tutorial/training/gocean/2.4-GameOfLife-psyclone/solution/read_config_mod.f90

…ith a symlink.

hiker · 2026-01-08T04:55:34Z

Wow, that was an impressive review. I think it took me over an hour just to add 'done' to your comments. You rightly pointed out some repeated errors (generat function, missing docstrings, inlining not mentioned). I think I've fixed them up everywhere.

I've triggered the full IT as well.

hiker · 2026-01-08T06:19:26Z

@arporter I have a problem - turns out the OpenMP offloading does not work (anymore??). Due to a bug in the makefile, it actually hasn't applied the offloading script at all in my recent tests (I am pretty certain it did originally, but I did some refactoring/ simplifications of the Makefile, which I believe introduced this bug recently). Now I've fixed the Makefile, it aborts with:

make -C 2.16-GameOfLife-omp-offload/solution test_run
...
make --no-print-directory  run | tail -n 12 | diff -b - /archive/psyclone-tests/action-runner-software/actions-runner/_work/PSyclone-mirror/PSyclone-mirror/tutorial/training/gocean/gol-lib/glider.correct

libgomp: pointer target not mapped for attach

(see https://github.com/stfc/PSyclone-mirror/actions/runs/20805411529/job/59758702708)

Googling indicates that there might be a mapping missing (which we don't support afaik), or additional compiler options(?). I don't know what to do. I don't have (easy) access to hardware that would support managed memory, and I am not even certain what the right solution would be in this case :( Can you (or Sergi?) help? I assume you can log onto the machine and try running it there?

…ran would compile openmp offloading, but that doesn't work then).

hiker · 2026-01-09T06:43:51Z

As discussed, I have added a USE_GPU=yes/no variable. If this is set to no, the GPU examples will be compiled, but not executed (even without openacc/opnmp support they can be compiled, which is at least a bit better than just checking if the transformations are done).

I only triggered the compilation IT (to speed things up), since I expect there will be another round of changes, and no code here should affect anything else outside of training.

arporter

Made it all the way through! I like your various examples - they are impressive.
Mostly it's just a bit more tidying of docstrings.
I have to admit that I'm not that keen on the use of the .x90 extension on files that are just standard .f90. This could cause confusion with the the Met Office approach where only files written using the DSL have the .x90 extension. I can see that it perhaps makes it easier to automatically run PSyclone on the necessary files but you could just name them in the Makefile.

arporter · 2026-01-09T09:24:00Z

.github/workflows/compilation.yml

        # Enable OpenMP on CPU (for some run tests)
-        F90=mpif90 F90FLAGS="-mp -Minfo=all" make -C tutorial/training test_run
+        USE_GPU=no F90=mpif90 F90FLAGS="-mp -Minfo=all" make -C tutorial/training test_run
        # Delete any previous foles, and compile the openacc example with openacc:


arporter · 2026-01-09T09:29:05Z

tutorial/training/README.rst

+The trainings material is split into four parts, and the files
+and documentation here are only the hands-on section. The full
+training is accompanied by a presentations
+(a slightly outdated version of the slides are available on the `wiki


Ah, I thought the slides were separate material to the training - I'd forgotten you created them. I like your suggestion since the training READMEs etc. are available in the repo anyway.

tutorial/training/README.rst

tutorial/training/gocean/2.10-GameOfLife-mpi/openmp.py

tutorial/training/gocean/2.16-GameOfLife-omp-offload/solution/output_field_mod.f90

arporter · 2026-01-09T16:49:14Z

tutorial/training/transformation/3.6-sympy/solution/dataflow.py

+# Create the PSyIR, and get the variable access information:
+reader = FortranReader()
+psyir = reader.psyir_from_source(code)
+routine = psyir.find_routine_psyir("foo")


You don't seem to use routine but perhaps you should s/psyir.children[0]/routine/ below?

arporter · 2026-01-09T20:46:07Z

tutorial/training/transformation/3.6-sympy/solution/dataflow.py

+print("digraph {")
+
+# Handle each variable
+for var in varinfo:


Also nice. Made me remember that I want to move Habbakuk (https://github.com/arporter/habakkuk) over to PSyIR and this is pretty much it.

arporter · 2026-01-09T20:50:27Z

tutorial/training/transformation/3.8-code-creation/combine_mod.x90

+
+        do j=ystart, ystop
+            do i=xstart, xstop
+                current(i, j) = current(i, j) - die(i, j) + born(i, j)


Should be a .f90?

arporter · 2026-01-09T20:52:51Z

tutorial/training/transformation/3.8-code-creation/solution/combine_mod.x90

+
+    subroutine combine(current, die, born)
+
+        implicit none


arporter · 2026-01-09T20:56:22Z

tutorial/training/transformation/common.mk

+# arguments passed to the first invocation of Make. The last entry
+# in this list is the current file.
+this_file := $(abspath $(lastword $(MAKEFILE_LIST)))
+# PSyclone directory is up two from this file


hiker added 30 commits February 21, 2024 12:23

#1623 More updates to handson.

e8deeec

#1623 Display schedule before the error happens.

0a8df3b

#1623 Made comparisons in README clearer.

533fcbc

1623 Remove example 7, which is too difficult.

b416af9

Merge remote-tracking branch 'origin/master' into 1623_add_training

fe5dcd8

#2049 Moved write_driver into base class.

81fd81c

#2049 Moved get_driver_as_string to base class, meaning gocean driver…

287419f

… now inlines.

#2049 Add a compilation test of the gocean driver (works only with mp…

b64217a

…i atm).

#2049 Updated gocean extract example to work with inlining.

cb863c6

#2049 Moved flatten_signature to base class.

bbdba64

#2049 Moved some tests.

be81aec

#2049 Moved import_modules into base class.

2ea3ae9

#2049 Start to move add_all_kernel_symbols to the base class.

5df947d

#2049 Fixed tests and error in logic.

488ae3d

#2049 Moved rest of add_all_kernel_symbols to base class.

92722a6

#2049 Import go_wp for gocean driver.

44d4d2b

Merge branch '1992_mpi_support_for_extraction' into 2049_driver_basec…

ff148a0

…lass

#2049 Fix failing test.

23696d9

Merge branch '1992_mpi_support_for_extraction' into 2049_driver_basec…

767425d

…lass

Merge branch '1992_mpi_support_for_extraction' into 2049_driver_basec…

304a570

…lass

Merge branch '1992_mpi_support_for_extraction' into 2049_driver_basec…

b376483

…lass

Merge branch '1992_mpi_support_for_extraction' into 2049_driver_basec…

6711833

…lass

Merge remote-tracking branch 'origin/master' into 2049_driver_baseclass

fdc038c

Merge remote-tracking branch 'origin/master' into 2049_driver_baseclass

755d1da

Merge remote-tracking branch 'origin/1623_add_training' into 1623_add…

0147782

…_training

Merge branch '257_verify_loop_fusion' into 1623_add_training

08fdf88

Merge branch '2690_value_verification' into 1623_add_training

d8c1e97

#1623 Started to support new value_range_check library.

64456e8

Merge branch '2690_value_verification' into 1623_add_training

346b298

Merge branch '2690_value_verification' into 1623_add_training

fc53fa1

arporter added the under review label Jan 6, 2026

arporter requested changes Jan 6, 2026

View reviewed changes

arporter added reviewed with actions and removed under review labels Jan 6, 2026

hiker added 11 commits January 7, 2026 14:37

#1623 Removed two unnecessary files, and replaced the remaining one w…

b380760

…ith a symlink.

#1623 Started to address reviewers comments.

108fb9a

Merge remote-tracking branch 'origin/master' into 1623_add_training

6b4ce26

#1623 Fix flake8.

8c43ef2

#1623 Addressed more issues raised in review.

c809814

#1623 Added missing docstrings.

4c7c825

#1623 More updated based on code review.

724f873

#1623 Added more doc strings to Fortran files.

3f0438d

#1623 Removed unused compile target from top level Makefiles.

f6a10fe

#1623 Updated README with additional information.

d4735c8

#1623 Renamed psyclone_for_lfric_users to lfric_users.

fa501df

hiker temporarily deployed to integration January 8, 2026 04:21 — with GitHub Actions Inactive

hiker added 2 commits January 9, 2026 14:56

#1623 Add USE_GPU variable to avoid execution of GPU code (e.g. gfort…

baaed9e

…ran would compile openmp offloading, but that doesn't work then).

Merge remote-tracking branch 'origin/master' into 1623_add_training

0f6ec90

hiker deployed to integration January 9, 2026 03:58 — with GitHub Actions Active

hiker added ready for review and removed reviewed with actions labels Jan 9, 2026

hiker requested a review from arporter January 9, 2026 06:44

arporter added under review and removed ready for review labels Jan 9, 2026

arporter requested changes Jan 9, 2026

View reviewed changes

arporter added reviewed with actions and removed under review labels Jan 9, 2026

1623 add training #3145

Are you sure you want to change the base?

1623 add training #3145

Conversation

hiker commented Sep 21, 2025

Uh oh!

arporter left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hiker commented Jan 8, 2026

Uh oh!

hiker commented Jan 8, 2026

Uh oh!

hiker commented Jan 9, 2026

Uh oh!

arporter left a comment

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

arporter Jan 9, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants