docs[cartesian]: Update ADRs with recent development#2512
Open
romanc wants to merge 2 commits intoGridTools:mainfrom
Open
docs[cartesian]: Update ADRs with recent development#2512romanc wants to merge 2 commits intoGridTools:mainfrom
romanc wants to merge 2 commits intoGridTools:mainfrom
Conversation
3 tasks
We have been lagging behind on updating the ADRs. This PR updates a bunch of ADRs to reflect recent development in `gt4py.cartesian`.
romanc
force-pushed
the
romanc/adr-update
branch
from
f93ce73 to
250d6c4
Compare
twicki
reviewed
Mar 12, 2026
|
|
||
| ## Context | ||
|
|
||
| During performance optimization work, we discovered that there's currently no way to stream memory directly (i.e. without copy) from Fortran into a DaCe-based backend. For hybrid Fortran/GT4Py applications, this can be a bottleneck. |
Contributor
There was a problem hiding this comment.
Suggested change
| During performance optimization work, we discovered that there's currently no way to stream memory directly (i.e. without copy) from Fortran into a DaCe-based backend. For hybrid Fortran/GT4Py applications, this can be a bottleneck. | |
| During performance optimization work, we discovered that there's currently no way to stream memory directly (i.e. without copy) from Fortran into a DaCe-based backend. For hybrid Fortran/GT4Py applications where we have no control over the memory layout of the data, this can be a bottleneck. |
|
|
||
| ## Decision | ||
|
|
||
| We decided to add a backend that allows directly streaming memory from Fortran into GT4Py stencils (and back) without any copies. To communicate the expected device and memory layout, we named the backend `dace:cpu_KJI`. |
Contributor
There was a problem hiding this comment.
Suggested change
| We decided to add a backend that allows directly streaming memory from Fortran into GT4Py stencils (and back) without any copies. To communicate the expected device and memory layout, we named the backend `dace:cpu_KJI`. | |
| We decided to add a backend that allows directly streaming memory from Fortran into GT4Py stencils (and back) without any copies. To communicate the expected device and memory layout, we named the backend `dace:cpu_KJI`. We chose to keep `cpu` as part of the name because we do not see the potential use-case of using such a poorly-fitting layout to do computation on GPU ever. |
| Power users of GT4Py have now a way to stream memory directly from (and back to) Fortran without a copy. | ||
|
|
||
| To help with the growing number of DaCe-based backends, we introduce an automatic match for DaCe-based backends between layout and schedule. This makes the default scheduling always cache-optimal by construction and supersedes a previous rubber band fix introduced with the `dace:cpu_kfirst` backend. | ||
|
|
Contributor
There was a problem hiding this comment.
do we need to talk about the consequence of having another backend and the lack of extra overhead for maintainig this? And can we say that with the new way of describing backends this also does not add significant cognitive load to have to keep more in mind?
Contributor
There was a problem hiding this comment.
just in general, GT has a OpenMP requirement: https://github.com/GridTools/gridtools/blob/c5a85af55fd4df8629b8f64c88673261ebcb7624/docs/_sources/introduction/introduction.rst.txt?plain=1#L61
This is not tackled with a slightly more general name GT4PY_CARTESIAN_ENABLE_OPENMP compared to the use that is just to target DaCe backends.
I think all the choices made are very valid, but it still strives us further away from being able to use the GT-backends with coverage. we might need to think about how to handle this drift
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Description
We have been lagging behind on updating the ADRs. This PR updates a bunch of ADRs to reflect recent development in
gt4py.cartesian.Requirements
N/A