Muphys bug fix: drivers run_graupel_only and run_full_muphys #877
Conversation
…merge' into muphys_bug_fix
vectorflux
force-pushed
the
muphys_bug_fix
branch
from
35d058f to
3752592
Compare
...s/muphys/src/icon4py/model/atmosphere/subgrid_scale_physics/muphys/driver/run_full_muphys.py
Outdated
Show resolved
Hide resolved
Clean up the graupel_only driver, and create an integration test to run through pytest. Co-authored-by: Will Sawyer <wsawyer@cscs.ch>
Co-authored-by: Edoardo Paone <edoardo.paone@cscs.ch>
| mask = (maximum(q.c, maximum(q.g, maximum(q.i, maximum(q.r, q.s)))) > g_ct.qmin) | ( | ||
| (t < g_ct.tfrz_het2) & (q.v > _qsat_ice_rho(t, rho)) | ||
| ) | ||
| is_sig_present = maximum(q.g, maximum(q.i, q.s)) > g_ct.qmin |
There was a problem hiding this comment.
I would be interested in the performance of
| mask = (maximum(q.c, maximum(q.g, maximum(q.i, maximum(q.r, q.s)))) > g_ct.qmin) | ( | |
| (t < g_ct.tfrz_het2) & (q.v > _qsat_ice_rho(t, rho)) | |
| ) | |
| is_sig_present = maximum(q.g, maximum(q.i, q.s)) > g_ct.qmin | |
| mask = True | |
| is_sig_present = True |
Is there a constant folding pass in our dace pipeline that would completely eliminate the conditionals on these variables?
There was a problem hiding this comment.
Wouldn't such a pass already be on IR level, by passing the constant values as static args? Anyway, there is also a way to do it on the SDFG level in the dace backend, but we need to pass the constant values into the backend, see https://github.com/GridTools/gt4py/blob/22d9ab9988eab06c2fa7918fc817577f72c8be5a/src/gt4py/next/program_processors/runners/dace/workflow/translation.py#L392
There was a problem hiding this comment.
I don't understand the question.
There was a problem hiding this comment.
We can ignore my question, now I understand better your original question. The dace backend runs apply_fieldview_transforms, which calls dead_code_elimination. The latter uses the constant folding pass. So, to answer your question: yes, there is a constant folding pass in our dace pipeline.
`dz` was computed in the precision of the input file, therefore graupel did not verify to f64 precision for f32 input files.
... and enable it for the graupel_only test and driver.
This extracts icon4py infrastructure changes from #877. Changes: - extract `download_test_data` function - allow stencil tests to allocate fields for named collections
Co-authored-by: Edoardo Paone <edoardo.paone@cscs.ch>
|
cscs-ci run dace |
|
cscs-ci run default |
|
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:
To run test levels ignored by the default test suite (mostly simple datatest for static fields computations) you can use:
For more detailed information please look at CI in the EXCLAIM universe. |
|
cscs-ci run default |
|
cscs-ci run dace |
This PR addresses minor deficiencies in the graupel and saturation_adjustment GT4Py programs, which might influence performance, but more importantly in the run_graupel_only.py and run_full_muphys.py drivers which call them. These were restructured to improve performance, in particular by performing a set-up of both programs in the initialization step.
TODOs: