Npir backend #300
Merged
Npir backend #300
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
- switched to the existing DebugModuleGenerator - implemented codegen with python for loops - ranges are aalways the full domain - lap codegen working, need to check loop bounds
* split stencil and computation modules * add file info pass * make gtir nodes immutable * update unit tests for stencil object snippet generators * naive python ir and translator from gtir * pnir -> python ast * [wip/exp] stencil module via AST * [wip] code generator wiring & fix linter errors Co-authored-by: Rico Häuselmann <ricoh@cscs.ch>
* complete python code generation path * move `gtc` related modules into `gt4py.gtc` subpackage * move `gtc` related tests into `tests/unittests/test_gtc` Co-authored-by: Rico Häuselmann <ricoh@cscs.ch>
Co-authored-by: Rico Häuselmann <ricoh@cscs.ch>
- Replace horizontal loop by ParAssignStmt (parallel model) - Remove stencil indirection (as unused in current GT4Py) - Add IfStmt, ScalarAccess - introduce dtype propagation - add logical, comparison and better type propagation - scalar vars and keyword args in defir - use SymbolName/SymbolRef Co-authored-by: Johann Dahm <johann.dahm@gmail.com>
Move common nodes to common module and introduce generics for Exprs and Stmts. Unrelated change: - Add expr_kind to each node which is either `FIELD` or `SCALAR` to distinguish between scalar and field `if`s because of their different semantics and limitations in the GTScript parallel model. Co-authored-by: Rico Haeuselmann <r.haeuselmann@gmx.ch>
As discussed, the extra nesting of `Stmt`s in `VerticalInterval` in `VerticalLoop` doesn't provide extra semantic information. Simplifies the IR and makes it compatible with the current Definition IR.
add type inferral pass for field access in GTIR
NPIR and generator: copy and simple arithmetic stencils.
npir using generic nodes
use textwrap.dedent to represent multiline template strings.
GTC: introduce NativeFuncCall
GTC: Add cast node Co-authored-by: Rico Haeuselmann <r.haeuselmann@gmx.ch>
Adds the GridTools C++ 2.x IR and codegenerator.
|
It seems all tests pass except in the regression test for vertical advection dycore for gtc:dace there is a numerical drift introduced, slightly larger than tolerance. I assume it has something to do with the OIR optimizations code reuse, but I can't tell why or even if it may just be acceptable. edit: The problem was an incompatible version of dace. Has been pinned to an older commit by #456. |
jdahm
reviewed
Jul 7, 2021
jdahm
reviewed
Jul 7, 2021
Comment on lines
+52
to
+54
| oir = OirPipeline(gtir_to_oir.GTIRToOIR().visit(gtir)).full( | ||
| skip=[GreedyMerging().visit, NoFieldAccessPruning().visit] | ||
| ) |
jdahm
reviewed
Jul 7, 2021
jdahm
reviewed
Jul 7, 2021
jdahm
reviewed
Jul 7, 2021
jdahm
reviewed
Jul 8, 2021
jdahm
reviewed
Jul 8, 2021
|
bors try |
tryBuild failed: |
jdahm
reviewed
Jul 12, 2021
jdahm
reviewed
Jul 12, 2021
3 tasks
egparedes
pushed a commit
that referenced
this pull request
Sep 4, 2024
`MaskInlining` will be a key feature of the improved gt4py / DaCe bridge as envisioned in GEOS-ESM/NDSL#53, which will propagate more control flow elements from gt4py to DaCe. We thus want to make sure if `MaskInlining` itself is posing any problems. We tracked the git history of this explicit exclude back to PR #300, which not only added the npir backend, but also refactored how pipeline stages are defined. Previous to this PR optimization passes to be specified explicitly and the PR changed this to an exclude list. It looks like the `MaskInlining` pass was added after the list of optimization passes was formed and thus excluded (to keep the list constant). From this history we thus don't expect `MaskInlining` to break anything. Co-authored-by: Roman Cattaneo <>
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.
Description
Numpy backend for the gtc pipeline. Supports all features up to and including lower dimensional fields.