DaCe Orchestration for the Diffusion Granule #514
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
cscs-ci run dace |
|
cscs-ci run default |
|
launch jenkins spack |
|
cscs-ci run dace |
|
cscs-ci run default |
egparedes
reviewed
Oct 7, 2024
model/atmosphere/diffusion/src/icon4py/model/atmosphere/diffusion/diffusion.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
model/common/src/icon4py/model/common/orchestration/decorator.py
Outdated
Show resolved
Hide resolved
DropD
reviewed
Oct 10, 2024
DropD
reviewed
Oct 10, 2024
|
cscs-ci run dace |
|
cscs-ci run default |
|
launch jenkins spack |
|
cscs-ci run dace |
|
cscs-ci run default |
|
launch jenkins spack |
|
cscs-ci run dace |
|
Mandatory Tests Please make sure you run these tests via comment before you merge!
Optional Tests To run benchmarks you can use:
To run tests and benchmarks with the DaCe backend you can use:
In case your change might affect downstream icon-exclaim, please consider running
For more detailed information please look at CI in the EXCLAIM universe. |
|
cscs-ci run default |
|
launch jenkins spack |
|
cscs-ci run dace |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR concerns only the DaCe backend.
Currently, the Diffusion granule calls the various stencils one after the other (in
_do_diffusion_step). This means that for every stencil, there is a separate SDFG, and consequently one stencil is not aware of the others (this behavior limits the analyzability of the full granule). This PR introduces a decorator that fuses all these SDFGs under one compilation unit, allowing DaCe for further analysis and optimizations. Placing a GT4Py program inside a DaCe program region, and extracting the underlying SDFG, is possible due to this GT4Py PR.The halo exchanges are also taken care from the DaCe orchestrator. A follow-up PR will introduce automated halo exchanges. Currently, the halo exchange class implements the SDFGConvertible interface like GT4Py Program.
The orchestrator is activated either through an env var
ICON4PY_DACE_ORCHESTRATION=AnyValueor through this pytest option--dace-orchestration=AnyValue.The orchestrator suppports ahead of time compilation, however given that DaCe does not support nested Structures -like self in the Diffusion class-, some of the arguments need to be provided at compile time, through
dace.compiletime. This annotation means that the correspoding argument will be considered from the closure of the function.The orchestrator provides full caching support and takes into consideration when a dace.compiletime argument changes, with a subsequent re-compilation of the fused SDFG.