compiler: Enhance detect_accesses and patch symbolic padding by FabioLuporini · Pull Request #2886 · devitocodes/devito · GitHub

Test in PRO in the PR/branch with same name as this one

Codecov Report

❌ Patch coverage is 91.57895% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.47%. Comparing base (bd39274) to head (fc4c548).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
devito/types/parallel.py	66.66%	7 Missing ⚠️
devito/symbolics/extended_sympy.py	95.83%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2886      +/-   ##
==========================================
+ Coverage   83.43%   83.47%   +0.03%     
==========================================
  Files         248      248              
  Lines       51381    51471      +90     
  Branches     4431     4433       +2     
==========================================
+ Hits        42871    42964      +93     
+ Misses       7757     7756       -1     
+ Partials      753      751       -2

Flag	Coverage Δ
pytest-gpu-aomp-amdgpuX	68.80% <69.35%> (-0.01%)	⬇️
pytest-gpu-gcc-	78.08% <91.57%> (+0.03%)	⬆️
pytest-gpu-icx-	77.69% <91.57%> (+0.02%)	⬆️
pytest-gpu-nvc-nvidiaX	69.32% <69.35%> (-0.01%)	⬇️

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.

This look very specific and seems like that's something retrieve_index should catch

the idea is that other objects can end up there, in the future, maybe...

as for retrieve_indexed catching it: disagree, it's not an implicit Indexed, it's rarther a logical representation of the base address of the TensorMove -- as an Indexed, for homogeneity

Would if step < 1 be safer? Prevents a negative number being supplied

fair, yes, I will add that. Will also ensure it's an integer

Maybe implies that extended_sympy should be moved somewhere else?

yes, it's a long standing issue, documented somewhere

Can the round up factor also be symbolic?

not sure, I cooked up something simple for my needs after days of frustration

see comment above -- now ensuring it's an integer number

'bdw' is oddly specific

stupid AI, will patch in the next PR

I would check it's an integer before checking it's greater than zero, but whatever

FabioLuporini requested a review from mloubout April 15, 2026 09:19

FabioLuporini added the compiler label Apr 15, 2026

mloubout reviewed Apr 15, 2026

View reviewed changes

FabioLuporini changed the title compiler: Enhance detect_accesses compiler: Enhance detect_accesses and patch padding Apr 16, 2026

FabioLuporini changed the title compiler: Enhance detect_accesses and patch padding compiler: Enhance detect_accesses and patch symbolic padding Apr 16, 2026

compiler: Enhance detect_accesses

5582088

FabioLuporini force-pushed the tma-write branch from 63b23b5 to c975ff5 Compare April 16, 2026 12:07

EdCaunt reviewed Apr 16, 2026

View reviewed changes

compiler: Patch Array padding

fc4c548

FabioLuporini force-pushed the tma-write branch from c975ff5 to fc4c548 Compare April 16, 2026 13:21

mloubout reviewed Apr 16, 2026

View reviewed changes

mloubout approved these changes Apr 16, 2026

View reviewed changes

EdCaunt approved these changes Apr 16, 2026

View reviewed changes

FabioLuporini merged commit cc4fe0d into main Apr 16, 2026
42 checks passed

FabioLuporini deleted the tma-write branch April 16, 2026 15:48

                           # Already a multiple of `mmts`, no need to pad
                           return nopadding
+                      else:
+                          from devito.symbolics import RoundUp  # noqa

+                  # Search among the Indexeds (Most accesses typically stem from Indexeds)
+                  plain_indexeds = retrieve_indexed(exprs, deep=True)
+                  # Search among higher order objects, which still represent meaningful accesses

+                      value = sympify(value)
+                      step = sympify(step)
+                      if step == 0:

+                  u = Function(name='u', grid=grid)
+                  a = dSymbol('a', dtype=np.int32)
+                  expr = RoundUp(a, 16)

+                      if step < 1:
+                          raise ValueError("Cannot round up with negative `step`")
+                      if not is_integer(step):

compiler: Enhance detect_accesses and patch symbolic padding by FabioLuporini · Pull Request #2886 · devitocodes/devito · GitHub

Uh oh!

compiler: Enhance detect_accesses and patch symbolic padding - #2886

compiler: Enhance detect_accesses and patch symbolic padding#2886FabioLuporini merged 2 commits intomaindevitocodes/devito:mainfrom tma-writedevitocodes/devito:tma-writeCopy head branch name to clipboard

Conversation

FabioLuporini commented Apr 15, 2026

Uh oh!

codecov Bot commented Apr 15, 2026 • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FabioLuporini Apr 16, 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!

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

compiler: Enhance detect_accesses and patch symbolic padding#2886
FabioLuporini merged 2 commits into
mainfrom
tma-write

codecov Bot commented Apr 15, 2026 •
edited
Loading

FabioLuporini Apr 16, 2026 •
edited
Loading