Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

compiler: Fix placement of ConditionalDimension in subdomain #2050

Merged
merged 3 commits into from
Apr 30, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion devito/ir/clusters/algorithms.py
Original file line number Diff line number Diff line change
Expand Up @@ -238,7 +238,7 @@ def guard(clusters):
if cd._factor is not None:
k = d
else:
dims = pull_dims(cd.condition)
dims = pull_dims(cd.condition, flag=False)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this should be flag=cd.indirect for the case where the conditional is used as the indexing dimension

k = max(dims, default=d, key=lambda i: c.ispace.index(i))

# Pull `cd` from any expr
Expand Down
100 changes: 99 additions & 1 deletion tests/test_subdomains.py
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@

from conftest import opts_tiling, assert_structure
from devito import (ConditionalDimension, Constant, Grid, Function, TimeFunction,
Eq, solve, Operator, SubDomain, SubDomainSet)
Eq, solve, Operator, SubDomain, SubDomainSet, Lt)
from devito.ir import FindNodes, Expression, Iteration
from devito.tools import timed_region

Expand Down Expand Up @@ -693,3 +693,101 @@ class Dummy(SubDomainSet):
# Switch the thickness symbols between MultiSubDimensions with the rebuild
remixed = [d._rebuild(thickness=t) for d, t in zip(sdims, tkns[::-1])]
assert [d.thickness for d in remixed] == tkns[::-1]


class TestSubDomain_w_condition(object):

def test_condition_w_subdomain_v0(self):

shape = (10, )
grid = Grid(shape=shape)
x, = grid.dimensions

class Middle(SubDomain):
name = 'middle'

def define(self, dimensions):
return {x: ('middle', 2, 4)}

mid = Middle()
my_grid = Grid(shape=shape, subdomains=(mid, ))

f = Function(name='f', grid=my_grid)

sdf = Function(name='sdf', grid=my_grid)
sdf.data[5:] = 1

condition = Lt(sdf[mid.dimensions[0]], 1)

ci = ConditionalDimension(name='ci', condition=condition,
parent=mid.dimensions[0])
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

mid.dimensions[0] is a root dimension already, do we actually need this test?

IOW, would this test actually fail in current master?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.


op = Operator(Eq(f, f + 10, implicit_dims=ci,
subdomain=my_grid.subdomains['middle']))
op.apply()

assert_structure(op, ['x'], 'x')

def test_condition_w_subdomain_v1(self):

shape = (10, 10)
grid = Grid(shape=shape)
x, y = grid.dimensions

class Middle(SubDomain):
name = 'middle'

def define(self, dimensions):
return {x: x, y: ('middle', 2, 4)}

mid = Middle()
my_grid = Grid(shape=shape, subdomains=(mid, ))

sdf = Function(name='sdf', grid=grid)
sdf.data[:, 5:] = 1
sdf.data[2:6, 3:5] = 1

x1, y1 = mid.dimensions

condition = Lt(sdf[x1, y1], 1)
ci = ConditionalDimension(name='ci', condition=condition, parent=y1)

f = Function(name='f', grid=my_grid)
op = Operator(Eq(f, f + 10, implicit_dims=ci,
subdomain=my_grid.subdomains['middle']))

op.apply()

assert_structure(op, ['xy'], 'xy')

def test_condition_w_subdomain_v2(self):
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mloubout could you help sketching a test about what you mention as the conditional being the indexing dimension?
or at least show me an example with some pseducode?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok sorry after bit of digging, forgot this algorithm section is only for non indirect (see line 223 in algorithms.py)

GTG

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah right, this is what happens if I have long time to see a PR


shape = (10, 10)
grid = Grid(shape=shape)
x, y = grid.dimensions

class Middle(SubDomain):
name = 'middle'

def define(self, dimensions):
return {x: ('middle', 2, 4), y: ('middle', 2, 4)}

mid = Middle()
my_grid = Grid(shape=shape, subdomains=(mid, ))

sdf = Function(name='sdf', grid=my_grid)
sdf.data[2:4, 5:] = 1
sdf.data[2:6, 3:5] = 1

x1, y1 = mid.dimensions

condition = Lt(sdf[x1, y1], 1)
ci = ConditionalDimension(name='ci', condition=condition, parent=y1)

f = Function(name='f', grid=my_grid)
op = Operator(Eq(f, f + 10, implicit_dims=ci,
subdomain=my_grid.subdomains['middle']))

op.apply()

assert_structure(op, ['xy'], 'xy')
Loading