compiler: add rudimentary support for multi-cond buffering by mloubout · Pull Request #2838 · devitocodes/devito

mloubout · 2026-01-28T18:16:45Z

API up to debate, but need something within these line to allow for the MFE (see test)

codecov · 2026-01-28T18:37:15Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.97%. Comparing base (1c65e48) to head (8304fdb).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2838      +/-   ##
==========================================
+ Coverage   78.95%   78.97%   +0.01%     
==========================================
  Files         248      248              
  Lines       50877    50901      +24     
  Branches     4394     4395       +1     
==========================================
+ Hits        40170    40197      +27     
+ Misses       9908     9905       -3     
  Partials      799      799

☔ 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.

review-notebook-app · 2026-01-29T00:15:29Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

examples/userapi/05_conditional_dimension.ipynb

EdCaunt

What's the reasoning behind this change?

tests/test_buffering.py

EdCaunt · 2026-01-29T10:25:56Z

devito/ir/clusters/algorithms.py

                    exprs[i] = e.func(*e.args, conditionals=conditionals)

-            guards = {d: sympy.And(*v, evaluate=False) for d, v in guards.items()}
+            # Combination mode is And by default and Or if all conditions are


"Combination mode is And by default. If all conditions are Or then Or combination mode is used." may be less ambiguous

(nitpicking, with backtips ` around mode)

EdCaunt · 2026-01-29T10:28:22Z

devito/types/dimension.py

        index into arrays. A typical use case is when arrays are accessed
        indirectly via the ``condition`` expression.
+    relation: Or/And, default=And
+        How this ConditionalDimension will be combined with other ones.


Seems slightly unintuitive to specify this here? Surely it would make more sense to specify how ConditionalDimensions are combined at the point of combination?

Perhaps with an API like cdim.and(other) and cdim.or(other) which returns a new ConditionalDimension which combines the conditions of its constituent ConditionalDimensions?

Surely it would make more sense to specify how ConditionalDimensions are combined at the point of combination?

You cannot so that's not relvant. A Function a defined dimension. It will be combined with any indices and implicit_dims at lowering. Combinining it by hand won't do anything as it will be combined yet again at lowering with And making it useless

Ah ok, so this is to do with how it's handled during lowering and the combination is not exposed at the API level? It might be worth adding a note to the notebook explaining when you would want to change relation?

mloubout · 2026-01-29T13:01:57Z

What's the reasoning behind this change?

The test/MFE

FabioLuporini · 2026-01-29T15:19:19Z

tests/test_buffering.py

+    eqs.append(Eq(f.forward, T+1, implicit_dims=ctend))
+
+    # run operator with buffering
+    op = Operator(eqs, opt=('streaming', 'buffering'))


you don't need or want "streaming" here

FabioLuporini · 2026-02-03T13:07:05Z

devito/ir/clusters/algorithms.py

                    exprs[i] = e.func(*e.args, conditionals=conditionals)

-            guards = {d: sympy.And(*v, evaluate=False) for d, v in guards.items()}
+            # Combination `mode` is And by default and Or if all conditions are


ultra-nitpick, can be rephrased/adjusted/fixed, but can go in a different PR

FabioLuporini · 2026-02-03T13:07:26Z

devito/types/dimension.py

+    relation: Or/And, default=And
+        How this ConditionalDimension will be combined with other ones during
+        lowering for example combining Function's ConditionalDimension with
+        and Equation's implicit_dim. All dimensions within an equation


ultra-nitpick: Dimensions (not dimensions)

FabioLuporini · 2026-02-03T13:08:51Z

devito/types/dimension.py


    __rkwargs__ = DerivedDimension.__rkwargs__ + \
-        ('factor', 'condition', 'indirect')
+        ('factor', 'condition', 'indirect', 'relation')


not a big fan of the name "relation" as it's a bit vague

mloubout added compiler API api (symbolics, types, ...) labels Jan 28, 2026

mloubout force-pushed the multi-cond-buffering branch from c4cd3ac to cb9731e Compare January 28, 2026 18:20

mloubout force-pushed the multi-cond-buffering branch from cb9731e to e1a2dfb Compare January 29, 2026 00:15

mloubout force-pushed the multi-cond-buffering branch from e1a2dfb to 51a7df1 Compare January 29, 2026 02:05

EdCaunt reviewed Jan 29, 2026

View reviewed changes

examples/userapi/05_conditional_dimension.ipynb Show resolved Hide resolved

examples/userapi/05_conditional_dimension.ipynb Show resolved Hide resolved

EdCaunt reviewed Jan 29, 2026

View reviewed changes

FabioLuporini reviewed Jan 29, 2026

View reviewed changes

mloubout force-pushed the multi-cond-buffering branch 5 times, most recently from ec47415 to 7707aba Compare February 2, 2026 17:18

compiler: add rudimentary support for multi-cond buffering

a3c2426

mloubout force-pushed the multi-cond-buffering branch from 7707aba to 28fcf9d Compare February 3, 2026 11:35

FabioLuporini reviewed Feb 3, 2026

View reviewed changes

FabioLuporini approved these changes Feb 3, 2026

View reviewed changes

mloubout force-pushed the multi-cond-buffering branch 2 times, most recently from 5ebeaf8 to f680b19 Compare February 3, 2026 14:09

CI: misc docker deploy fixes

8304fdb

mloubout force-pushed the multi-cond-buffering branch from e6428da to 8304fdb Compare February 3, 2026 21:13

mloubout merged commit 013b2dc into main Feb 4, 2026
42 checks passed

mloubout deleted the multi-cond-buffering branch February 4, 2026 03:46

Conversation

mloubout commented Jan 28, 2026

Uh oh!

codecov bot commented Jan 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

review-notebook-app bot commented Jan 29, 2026

Uh oh!

Uh oh!

Uh oh!

EdCaunt left a comment

Choose a reason for hiding this comment

Uh oh!

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!

mloubout commented Jan 29, 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!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov bot commented Jan 28, 2026 •

edited

Loading