Skip to content

feat[next-dace]: Enable tasklet fusion in dataflow optimization #2452

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.

Sign up for GitHub

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

Closed
edopao wants to merge 4 commits into from
Closed

feat[next-dace]: Enable tasklet fusion in dataflow optimization #2452

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a7f8c2c
Enable tasklet fusion in dataflow optimization
edopao Jan 21, 2026
d1cee12
edit
edopao Jan 21, 2026
ccdf111
Merge branch 'main' into dace_tasklet_fusion
edopao Jan 21, 2026
19172fd
edit
edopao Jan 21, 2026
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
11 changes: 11 additions & 0 deletions src/gt4py/next/program_processors/runners/dace/transformations/auto_optimize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
import dace
from dace import data as dace_data
from dace.sdfg import nodes as dace_nodes, propagation as dace_propagation, utils as dace_sdutils
from dace.transformation import dataflow as dace_dataflow
from dace.transformation.auto import auto_optimize as dace_aoptimize
from dace.transformation.passes import analysis as dace_analysis

Expand Down Expand Up @@ -674,6 +675,16 @@ def _gt_auto_process_dataflow_inside_maps(
time, so the compiler will fully unroll them anyway.
"""

# The SDFG might contain tasklets with no input connectors, which simply write
# a constant value into a scalar node. If these tasklets were moved into the map
# scope, they would require an empty memlet edge from MapEntry, for synchronization.
# Empty memlets are not properly handled in code generation, so it is better
# to avoid this pattern. Running `TaskletFusion` at this stage helps to inline
# these constant-write tasklets into compute-tasklets.
sdfg.apply_transformations_repeated(
dace_dataflow.TaskletFusion, validate=False, validate_all=validate_all
)
Comment on lines +684 to +686
Copy link
Contributor

@philip-paul-mueller philip-paul-mueller Jan 23, 2026

Choose a reason for hiding this comment

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

The location is not optimial because it essentially destroy LoopBlooking but since that stencil does not depend on it, it should be fine.


# Constants (tasklets are needed to write them into a variable) should not be
# arguments to a kernel but be present inside the body.
sdfg.apply_transformations_once_everywhere(
Expand Down