(Closes #3251) Implementation for the PSyDirective. #3255

LonelyCat124 · 2025-12-09T15:20:13Z

Example and doc TODO

Basic implementation of PSyDirective including reading and writing.

At the moment I implemented it as a Standalone Directive (as it doesn't make sense to me to be a RegionDirective) but that could be more difficult to use in some ways, but I'll put together an example and see.

codecov · 2025-12-09T15:32:44Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.95%. Comparing base (2823f9d) to head (c541e7f).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3255   +/-   ##
=======================================
  Coverage   99.95%   99.95%           
=======================================
  Files         376      377    +1     
  Lines       53545    53580   +35     
=======================================
+ Hits        53523    53558   +35     
  Misses         22       22

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

sergisiso · 2025-12-09T15:48:26Z

I still don't think we need psyclone (!$psy) specific directives, I thought the plan was to add a generic Directive node, so that we could do psyir.walk(Directive) as an slight improvement to the current psyir.walk(Codeblock) (which tbh is almost the same as this PR implements, just with a slighly different _directive_handler)

LonelyCat124 · 2025-12-09T16:00:45Z

We could do that instead, but it would mean more false positives when people are looking for something specific to their script, e.g. after psyir.walk(???Directive) (name is not straightforward as we can't make it Directive since thats a necessary abstract class, along with StandaloneDirective and RegionDirective nodes. The other pain point is taking things out of CodeBlocks is it could be misinterpreted as support for them, and having non-RegionDirective nodes having to be StandaloneDirective subclasses.

I would maybe with ok with doing some GenericDirective and excluding all directives that start omp, ompx, omx, acc if you think that would be better?

sergisiso · 2025-12-09T16:19:18Z

name is not straightforward as we can't make it Directive since thats a necessary abstract class, ...
The other pain point is taking things out of CodeBlocks is it could be misinterpreted as support for them
I would maybe with ok with doing some GenericDirective

Good points, what about [Unrecognised/Unsupported/Unknown]Directive (being sibling of Standalone/RegionDirective)?

So it is clear that we don't support them and if we ever try to parse OpenMP, OpenACC into their own Directive the name will still make sense for those that we have not done yet. (my preferd is UnknownDirective, but I am unsure)

but it would mean more false positives when people are looking for something specific to their script

I don't understand this. It will be better than the current Codeblock, and it would allow searching for "ivdep" or other know compiler directive that may exist in the code.

excluding all directives that start omp, ompx, omx, acc

Why excluding those?

LonelyCat124 · 2025-12-10T10:22:58Z

Yeah I agree with most of that, so I'll modify this PR towards that.

My reasoning for ignoring omp/acc/etc. directives is party that keeping them as CodeBlocks means they'd keep any current properties that CodeBlocks have (i.e. disabling transformations etc.) and also make it clear that they're not meant to be used as part of any tree manipulation programming, but are just for keeping elements of the source code that PSyclone doesn't or can't support.

…tives supported

…yclone scripts

LonelyCat124 · 2025-12-15T10:33:46Z

@sergisiso this is ready for review, but no rush to do so.

sergisiso

@LonelyCat124 I like how this is implemented now, and especially having an example to showcase it. There are just minor comments now.

src/psyclone/tests/psyir/frontend/fparser2_directive_test.py

src/psyclone/psyir/nodes/unknown_directive.py

sergisiso · 2025-12-18T15:06:29Z

src/psyclone/psyir/nodes/unknown_directive.py

+    Directive representing PSyclone-specific directives in the tree.
+
+    :param directive_string: The content after the !$psy part of this
+                             node in the tree.


param kwargs is missing

I think thats just up to examining parent classes docs - I don't think I want to document kwargs here right? I think they just result in the base Node class keyword arguments here so are simply provided for internal use wherever the node is constructed if required.

We typically have :param kwargs: additional keyword arguments provided to the PSyIR node., I will add it myself if there is nothing else.

src/psyclone/psyir/nodes/unknown_directive.py

examples/psyir/unknown_directives/README.md

examples/psyir/unknown_directives/omp.py

examples/psyir/unknown_directives/Makefile

examples/psyir/unknown_directives/directive_filtering.F90

examples/psyir/custom_directives/Makefile

LonelyCat124 · 2026-01-09T15:46:53Z

@sergisiso This is probably ready for another look now

sergisiso

@LonelyCat124 Almost there. The Directive part is all good now, but the example Makefile does not recurse to the directories and doesn't test some files that previously were tested and the README needs updating.

sergisiso · 2026-01-13T09:54:50Z

examples/psyir/custom_directives/omp.py

+    for loop in routine.walk(Loop):
+        if loop in loops_to_skip:
+            continue
+        OMPLoopTrans(omp_directive="paralleldo").apply(loop)


This should be indented in (otherwise it just parallelises the last routine)

sergisiso · 2026-01-13T09:55:56Z

examples/psyir/custom_directives/identify_custom_directives.py

@@ -0,0 +1,56 @@
+# -----------------------------------------------------------------------------


Also change the name of this file to something more descriptive like: identify_custom_directives.py

sergisiso · 2026-01-13T09:57:57Z

examples/psyir/create_and_modify_psyir/Makefile

+# ------------------------------------------------------------------------------
+# Authors: A. R. Porter and S. Siso, STFC Daresbury Laboratory
+
+include ../common.mk


This is broken now (so make transform in this folder does not work)

Also, two of the examples in this folder's README should not be here, and instead should be in a parent readme

Fixed, and moved the README sections into their own examples readme's (not in the parent).

sergisiso · 2026-01-13T10:02:13Z

examples/psyir/Makefile

-
-run: compile
-	@echo "No run targets for PSyIR creation examples"
+EXAMPLES=custom_directives


You should also list the new create_and_modify_psyir as this was tested before (it was what this Makefile was doing). Also test the matmul and transpose directories (they both work fine for me)

Fixed those and added them too.

sergisiso · 2026-01-13T10:13:02Z

And not your fault but the examples/psyir/matmul|transpose ~~do not remove the output files with make clean~~ (they do, it is the make clean is not recursively propagated) a ideally this output files should also be .gitignored

LonelyCat124 · 2026-01-13T15:03:32Z

@sergisiso I think I added everything and the gitignore should ignore the .F90 files generated in those directories now.

sergisiso · 2026-01-15T10:35:38Z

src/psyclone/psyir/frontend/fparser2.py

+        # We don't turn OpenMP, OpenACC or directives we can't output
+        # correctly into Directive nodes. PSyclone currently always
+        # outputs directives starting with !$
+        dont_match = ["!$omp", "!$acc", "!$ompx", "!dir$"]


@LonelyCat124 I was a bit on the fence on these and wanted to wait for a use case. With what we discussed today: add profiling hooks around openmp in existing directives maybe it sways my opinion that these should be UnknownDirective so they are easier to identify. What do you think?

The !dir$ (and other !xxx$ ?) should still be CodeBlocks.

My main resoning for this is preventing them occurring in parallel loops. I could also add UnknownDirective into excluded_node_types in ParallelLoopTrans if you'd prefer, though this would block compiler directives appearing inside Parallel loops.

I could also add UnknownDirective into excluded_node_types in ParallelLoopTrans if you'd prefer, though this would block compiler directives appearing inside Parallel loops.

I am ok with this.

I'm still disallowing ompx extensions since we don't know at the moment what these might be and thus they're not useful to keep I think.

@sergisiso This is done now.

Implementation for the PSyDirective. Example and doc TODO

91aef63

LonelyCat124 added the in progress label Dec 10, 2025

LonelyCat124 added 8 commits December 10, 2025 15:16

Chance PSyDirective to UnknownDirective and change the types of direc…

d2dbb24

…tives supported

Added a simple example for using directives to filter behaviour of PS…

9219cf3

…yclone scripts

Merge branch 'master' into 3251_psy_directive

b5322a0

Merge branch 'master' into 3251_psy_directive

60bc725

linting

e994474

Missed conflict

bde8e38

Fixed failing test for omp directives still being kept as clodeblocks

73a14ac

Linting updatE

6cb1b14

LonelyCat124 added ready for review and removed in progress labels Dec 15, 2025

LonelyCat124 requested a review from sergisiso December 15, 2025 10:33

LonelyCat124 marked this pull request as ready for review December 15, 2025 10:33

sergisiso added under review and removed ready for review labels Dec 18, 2025

sergisiso requested changes Dec 18, 2025

View reviewed changes

sergisiso added reviewed with actions and removed under review labels Dec 18, 2025

LonelyCat124 added 4 commits January 9, 2026 14:20

changes for review

7d48c64

changes for review

a8a448e

Merge branch 'master' into 3251_psy_directive

2e61afd

Add mising Makefile import

57ca42d

LonelyCat124 removed the reviewed with actions label Jan 9, 2026

LonelyCat124 added the ready for review label Jan 9, 2026

LonelyCat124 requested a review from sergisiso January 9, 2026 15:46

sergisiso added under review and removed ready for review labels Jan 12, 2026

sergisiso requested changes Jan 13, 2026

View reviewed changes

sergisiso added reviewed with actions and removed under review labels Jan 13, 2026

LonelyCat124 added 2 commits January 13, 2026 15:00

Examples fixes

93ef087

gitignore precision

75e67be

LonelyCat124 added 4 commits January 14, 2026 15:00

merged master

66f9115

Fixed issues with matmul

f29b6c5

copyright

109ef5f

fix transpose

0daf279

sergisiso reviewed Jan 15, 2026

View reviewed changes

Allowing omp and acc directives to end up UnknownDirectives

c541e7f

LonelyCat124 requested a review from sergisiso January 15, 2026 13:22

LonelyCat124 added ready for review and removed reviewed with actions labels Jan 15, 2026

		@@ -0,0 +1,56 @@
		# -----------------------------------------------------------------------------

(Closes #3251) Implementation for the PSyDirective. #3255

Are you sure you want to change the base?

(Closes #3251) Implementation for the PSyDirective. #3255

Uh oh!

Conversation

LonelyCat124 commented Dec 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov bot commented Dec 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

sergisiso commented Dec 9, 2025

Uh oh!

LonelyCat124 commented Dec 9, 2025

Uh oh!

sergisiso commented Dec 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

LonelyCat124 commented Dec 10, 2025

Uh oh!

LonelyCat124 commented Dec 15, 2025

Uh oh!

sergisiso left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LonelyCat124 Jan 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

LonelyCat124 commented Jan 9, 2026

Uh oh!

sergisiso left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sergisiso commented Jan 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

LonelyCat124 commented Jan 13, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LonelyCat124 commented Dec 9, 2025 •

edited

Loading

codecov bot commented Dec 9, 2025 •

edited

Loading

sergisiso commented Dec 9, 2025 •

edited

Loading

LonelyCat124 Jan 9, 2026 •

edited

Loading

sergisiso left a comment •

edited

Loading

sergisiso commented Jan 13, 2026 •

edited

Loading