api: add mul interp mode by mloubout · Pull Request #2788 · devitocodes/devito · GitHub

Add a mode for symmetric interpolation (sometime needed for better stability)
Added notebook on interp/staggered
Added lots of tests for it

Codecov Report

❌ Patch coverage is 96.77419% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.44%. Comparing base (347d2c7) to head (cffdd95).

Files with missing lines	Patch %	Lines
devito/core/operator.py	61.53%	3 Missing and 2 partials ⚠️
devito/finite_differences/interpolation.py	93.10%	1 Missing and 1 partial ⚠️
devito/operator/operator.py	71.42%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2788      +/-   ##
==========================================
+ Coverage   83.38%   83.44%   +0.06%     
==========================================
  Files         248      249       +1     
  Lines       51960    52203     +243     
  Branches     4480     4496      +16     
==========================================
+ Hits        43326    43560     +234     
- Misses       7879     7883       +4     
- Partials      755      760       +5

Flag	Coverage Δ
pytest-gpu-aomp-amdgpuX	68.58% <57.60%> (-0.06%)	⬇️
pytest-gpu-gcc-	78.18% <96.05%> (+0.08%)	⬆️
pytest-gpu-icx-	78.10% <96.05%> (+0.08%)	⬆️
pytest-gpu-nvc-nvidiaX	69.13% <57.60%> (-0.05%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.

nitpicking, this should stay in a separate section being something related to the numerics, rather than performance (which is the case of all other opt-options).
As in, later on, perhaps around line 103, with a suitable comment

I'm also not sure if we really want this to be an Operator opt-option, as opposed to a Function parameter for example or something else

it also won't play well with the tuner this way, unless we specialise it somehow

imho this could/should rather be normalized at __new__

Well not easily no, because you can't have a wide variety of things, like 2D functions in 3D and so on that you can't really normalize.

would it be worth moving all of the interpolation-related stuff into a separate module finite_differences/interpolations.py ?

Seconded - it would make it easier to find things and limit the size of files

what's func here (and below) ?

this sentence "Use this ..." doesn't seem to be the complement of the other sentence you used for the "direct" mode above i.e. "the mode to pick unless ..."

Nitpick: might as well call it 08_staggered_interpolation for minimum ambiguity. Looks super good though. Is this a Claude job? Less typos than usual 😅

This is a side-quest PR, so yeah definitely claude tutorial :D

Seconded - it would make it easier to find things and limit the size of files

func.indices_ref.getters isn't equivalent to func.dimensions right? could u remind me why?

Actually no it should be the same, just an oversight

ok maybe add a memo for either this or next PR

mloubout added the API api (symbolics, types, ...) label Nov 10, 2025

mloubout force-pushed the fd-eval-add branch 6 times, most recently from 80c1334 to 01d3d1a Compare November 14, 2025 12:57

mloubout force-pushed the fd-eval-add branch 8 times, most recently from 0823c5f to 7821fc1 Compare November 21, 2025 14:48

mloubout force-pushed the fd-eval-add branch 4 times, most recently from f91130f to 136f771 Compare December 1, 2025 12:08

mloubout force-pushed the fd-eval-add branch 3 times, most recently from edd10f2 to 26490d4 Compare December 4, 2025 20:24

Base automatically changed from fd-eval-add to main December 8, 2025 17:37

mloubout force-pushed the strict-mul-eval-at branch 4 times, most recently from 233c8fb to 99fd21a Compare December 11, 2025 19:36

mloubout force-pushed the strict-mul-eval-at branch from 99fd21a to 198aedc Compare January 7, 2026 03:53

mloubout force-pushed the strict-mul-eval-at branch from 198aedc to 8cc5c13 Compare May 11, 2026 23:18

FabioLuporini reviewed May 12, 2026

View reviewed changes

EdCaunt reviewed May 12, 2026

View reviewed changes

mloubout force-pushed the strict-mul-eval-at branch 5 times, most recently from 64ef414 to 883d31e Compare May 12, 2026 16:46

FabioLuporini reviewed May 13, 2026

View reviewed changes

FabioLuporini approved these changes May 19, 2026

View reviewed changes

mloubout force-pushed the strict-mul-eval-at branch 11 times, most recently from 95f3c90 to 74fb990 Compare May 22, 2026 14:05

mloubout added 5 commits May 29, 2026 14:14

api: prevent evaluated derivatives to be re-evaluted

a43e619

api: add mul interp mode

40eecab

tweak mul mode

ab94a02

tweak mul mode

7f6572a

api: fix symmetric interp mode and add lots of tests

41fe102

mloubout force-pushed the strict-mul-eval-at branch from 74fb990 to 5019189 Compare May 29, 2026 18:14

api: move symbolic config to its own options

cffdd95

mloubout force-pushed the strict-mul-eval-at branch from 5019189 to cffdd95 Compare May 30, 2026 13:06

mloubout merged commit 870a160 into main Jun 2, 2026
62 of 63 checks passed

mloubout deleted the strict-mul-eval-at branch June 2, 2026 15:13

+                    `func`'s, the symmetric form `I * (a * I^T * b)` is built -- all
+                    factors are gathered at the highest-priority "block" location via
+                    `I^T`, multiplied there, and the product is interpolated to `func`
+                    via `I`. Use this for operators whose continuous form decomposes as

+                  """
+                  ref = {}
+                  for dim in deriv.dimensions:
+                      if dim in deriv.dims and dim in func.indices_ref.getters:

                       # Code generation options for derivatives
                       o['expand'] = oo.pop('expand', cls.EXPAND)
+                      o['interp-mode'] = oo.pop('interp-mode', cls.INTERP_MODE)

                       return not self or all(s == 0 for s in self)
+                  def __eq__(self, other):
+                      # Two empty-or-all-zero Staggerings are equivalent regardless of arity

		evalf_table[Pow] = evalf_table[sympy.Pow]


		def _interp_mapper(source, target, dims):

+                  Interpolation mode used by `Mul._eval_at` when projecting a multi-factor
+                  expression onto a target staggered location:
+                  * `'direct'` (default): each factor is shifted to `func`'s location

api: add mul interp mode by mloubout · Pull Request #2788 · devitocodes/devito · GitHub

Uh oh!

api: add mul interp mode - #2788

api: add mul interp mode#2788mloubout merged 6 commits intomaindevitocodes/devito:mainfrom strict-mul-eval-atdevitocodes/devito:strict-mul-eval-atCopy head branch name to clipboard

Conversation

mloubout commented Nov 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Dec 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

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!

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!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

api: add mul interp mode#2788
mloubout merged 6 commits into
mainfrom
strict-mul-eval-at

mloubout commented Nov 10, 2025 •
edited
Loading

codecov Bot commented Dec 11, 2025 •
edited
Loading