compiler: Change PETScArray type to PetscScalar #64

ZoeLeibowitz · 2025-03-20T16:03:42Z

Change the _C_ctype of PetscArray objects to PetscScalar
Refactored _restrict_keyword to be a property of Basic objects, instead of a string defined in the CGen class. By overriding this property for PetscArray objects, we can prevent specific warnings generated by PETSc.

TODO:

Add test that checks for single/double precision and ensures the multi-dim casting is correct

…sc_functions

… PETScArray

…he else statement

compiler: Rework /petsc/iet and PETScArrays

Rework /petsc/iet and PETScArrays

- Ensure that PetscInitialize and PetscFinalize are called only once per script, rather than for each Operator constructed. - Support PETSc command line args

* compiler/dsl: Add machinery to support coupled solvers

mloubout · 2025-03-20T17:28:54Z

devito/ir/iet/visitors.py

            if isinstance(obj, (AbstractFunction, IndexedData)) and mode >= 1:
                if not obj._mem_stack:
-                    strtype = '%s%s' % (strtype, self._restrict_keyword)
+                    strtype = '%s%s' % (strtype, obj._restrict_keyword)


f-string, we try to update all string to f".. {}" whenever we touch one

I don't think _restrict_keyword should be an object property that seems at the wrong place.

Do you have any other suggestions on where it could go if I need to drop it for certain objects like PETScArrays?

Could it be attached to the class?

mloubout · 2025-03-20T17:30:49Z

tests/test_petsc.py


-    f = Function(name='f', grid=grid, space_order=2)
-    g = Function(name='g', grid=grid, space_order=2)
+    f = Function(name='f', grid=grid, space_order=2, dtype=np.float64)


you don't need to repeat the dtype of it's already the one from the grid.

mloubout · 2025-03-20T17:31:12Z

tests/test_petsc.py

-    grid = Grid(shape=(11, 11))
-    f = Function(name='f', grid=grid, space_order=2)
+    grid = Grid(shape=(11, 11), dtype=np.float64)
+    f = Function(name='f', grid=grid, space_order=2, dtype=np.float64)


dtype not needed

mloubout · 2025-03-20T17:31:16Z

tests/test_petsc.py

-    f1 = Function(name='f1', grid=grid1, space_order=2)
-    f2 = Function(name='f2', grid=grid2, space_order=4)
-    f3 = Function(name='f3', grid=grid3, space_order=6)
+    f1 = Function(name='f1', grid=grid1, space_order=2, dtype=np.float64)


mloubout · 2025-03-20T17:31:24Z

tests/test_petsc.py


-    f = Function(name='f', grid=grid, space_order=2)
-    g = Function(name='g', grid=grid, space_order=2)
+    f = Function(name='f', grid=grid, space_order=2, dtype=np.float64)


mloubout · 2025-03-20T17:31:31Z

tests/test_petsc.py


-    f1 = Function(name='f1', grid=grid, space_order=2)
-    g1 = Function(name='g1', grid=grid, space_order=2)
+    f1 = Function(name='f1', grid=grid, space_order=2, dtype=np.float64)


EdCaunt · 2025-03-21T09:21:14Z

devito/petsc/types/types.py

                 arrays=None, **kwargs):
        self._target = kwargs.get('target', target)
+
+        petsc_precision = mapper[get_petsc_precision()]


Nitpick: This dict should probably have a more descriptive name

EdCaunt · 2025-03-21T09:26:20Z

devito/types/basic.py

        """
        return False

+    @property


I would make this an attribute of the class rather than the object. I.e. have _restrict_keyword = 'restrict' immediately under the docstring.

EdCaunt · 2025-03-21T09:31:59Z

tests/test_petsc.py

+        op = Operator(petsc)
+
+    # Check the solve function returns the correct output
+    op.apply()


Nitpick: I think it's probably better if the op.apply() is also within the switchconfig

JDBetteridge · 2025-03-21T11:15:15Z

devito/petsc/utils.py

+    """
+    petsc_dir = get_petsc_dir()
+    petsc_arch = get_petsc_arch()
+    variables_path = os.path.join(


Suggestion, use pathlib instead of chaining os.path commands

ZoeLeibowitz and others added 30 commits February 18, 2025 10:10

compiler: Add FuncPtrCall

a9801e7

misc: Update docstring

3f73add

compiler: Edit names, move position of Callback and other small edits

ccd10b9

misc: Update docstring for Callback

f34ae6d

misc: Clean up blank lines

ae28dac

misc: Split Dockerfile.cpu into .cpu and .intel

20fd2d1

try petscfunc from abstractfunc

2a6b923

latest.

2c0d33e

remove data allignment

12ab5bd

new petsc types

2e940ad

add const func in petscfunction

0815c70

dsl: Simplify PETScFunction class and add diff dtypes inside test_pet…

cc38a90

…sc_functions

dsl: Inherit from DiscreteFunction for PETScFunction instead?'

1dc5695

compiler: Switch PETScFunction to inherit from Array

6152445

compiler: Edit PETScFunction to inherit from ArrayBasic

cbedc60

misc: Fix pytest for test_petsc.py

469e308

compiler: Edit _C_ctype property of PETScFunction to utilise CustomDtype

9129223

compiler: Edit _C_ctype property of PETScFunction

acff687

misc: Update file due to incorrect rebase

3362ad8

compiler: Change name to PETScArray from PETScFunction

5b15b4e

compiler: Add intermediate class between PETScArray and ArrayBasic

7f3920a

compiler: Remove PETSc specific checks in place_definitions

15e7400

compiler: Remove PETSc specific check from place_definitions and edit…

ea9be50

… PETScArray

compiler: Remove is_PETScArray

905a7e2

compiler: Remove elif is_ObjectArray and use alloc_object_.. inside t…

415ad20

…he else statement

misc: Delete empty file

adc5428

misc: Move dtype_to_petsctype to petsc.py and clean up blank lines

0a571c7

misc: Clean up blank line

ce3cfb3

Update nodes.py

245242a

Update nodes.py

b45b17c

ZoeLeibowitz and others added 16 commits February 18, 2025 16:15

Address some comments

2c4a568

compiler: Move uxreplace_efuncs into Builder

b83ec92

misc: Clean up

4d75af8

misc: Add timeloop abstraction docs

5f25b0f

misc: Clean up and docs

37b62b3

misc: Spellings

10af16f

flake8

ef7ee34

Merge pull request #38 from ZoeLeibowitz/cleanup2

3b80b8b

compiler: Rework /petsc/iet and PETScArrays

Clean up

b989d06

Merge pull request #36 from ZoeLeibowitz/FieldFromPointer

be99236

Rework /petsc/iet and PETScArrays

compiler: Initialize and Finalize PETSc once and command line args

15f78ea

Merge pull request #43 from ZoeLeibowitz/init_finalize

cdc75f0

- Ensure that PetscInitialize and PetscFinalize are called only once per script, rather than for each Operator constructed. - Support PETSc command line args

Coupled (#41)

c68dd4f

* compiler/dsl: Add machinery to support coupled solvers

compiler: Error thrown for petsc precision

61e2ad6

flake8

c773e7f

misc: Flake8 tests

969c8d2

This was linked to issues Mar 20, 2025

Convert to PetscScalar instead of float/double #47

Open

Drop restrict keyword when using PETSc #46

Open

ZoeLeibowitz marked this pull request as ready for review March 20, 2025 17:26

ZoeLeibowitz requested review from EdCaunt, FabioLuporini, JDBetteridge and mloubout March 20, 2025 17:27

mloubout requested changes Mar 20, 2025

View reviewed changes

EdCaunt reviewed Mar 21, 2025

View reviewed changes

JDBetteridge reviewed Mar 21, 2025

View reviewed changes

ZoeLeibowitz force-pushed the master branch from c68dd4f to 4eb3171 Compare March 21, 2025 17:04

misc: Switch to Pathlib

d5904fb

ZoeLeibowitz force-pushed the master branch 2 times, most recently from 794ac3e to cccb988 Compare March 24, 2025 13:48

compiler: Change PETScArray type to PetscScalar #64

Are you sure you want to change the base?

compiler: Change PETScArray type to PetscScalar #64

Uh oh!

Conversation

ZoeLeibowitz commented Mar 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

ZoeLeibowitz commented Mar 20, 2025 •

edited

Loading