Add single precision option to the dycore (intermediate state) - #970

Open

starkphi wants to merge 182 commits into

mainfrom

add_single_precision_dycore_part1

Commits on Dec 3, 2025

add single to wp/vp definition
starkphi
committed
5ecfe67
vp/wp not needed to define dtype correspondance
starkphi
committed
f7c326b

Commits on Dec 4, 2025

cast to vp/wp, import directly (no ta.wpfloat), use WP_EPS instead of DBL_EPS
starkphi
committed
a75f6e6
add single_precision_marker to pytests
starkphi
committed
050bbb5
also use the convenient IconSavepoint.dtime()
starkphi
committed
98476ac

Commits on Dec 5, 2025

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
ea4a0a3
test_apply_diffusion_to_vn made single precision ready
starkphi
committed
adcf8fb
small formatting change by ruff
starkphi
committed
d031445

Commits on Dec 8, 2025

Commits on Dec 19, 2025

Apply suggestions from code review

starkphi
and
egparedes
authored
7604fd5

Commits on Dec 30, 2025

fix type error in compiled _calculate_divdamp_fields
starkphi
committed
20b7f8e
directly use WP_EPS/VP_EPS from constants and capitalize constant variable names
starkphi
committed
af5777e
remove obsolete lines
starkphi
committed
b65791a
reduce to one 'global' statement
starkphi
committed
b1fccb3
keep --enable_output a boolean flag
starkphi
committed
53b2c44
Merge branch 'add_single_precision_dycore_part1' of github.com:C2SM/icon4py into add_single_precision_dycore_part1
starkphi
committed
ab3bce6
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
173badc

Commits on Feb 17, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
37f59c6

Commits on May 20, 2026

Merge branch 'main' from 15.04.2026 (incl. #1185) into add_single_precision_dycore_part1
starkphi
committed
8a1cada

Commits on May 21, 2026

Merge branch 'main-temp_260508' into add_single_precision_dycore_part1
starkphi
committed
04ee97b
Remove logic error
starkphi
committed
43ad1d2
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
7eb7d3f

Commits on May 26, 2026

typos and small merge errors
starkphi
committed
0311e85

Commits on Jun 2, 2026

more casting to wpfloat
starkphi
committed
9409567
concise way to cast elements of config classes to wpfloat
starkphi
committed
279ef41
keep RBF calculations in double-precision
starkphi
committed
9f674d7

Commits on Jun 3, 2026

add casts for advection in single
starkphi
committed
944a39e

Commits on Jun 9, 2026

make field factories strictly double-precision
starkphi
committed
102a2dc

Commits on Jun 11, 2026

Commits on Jun 12, 2026

Setup jabw fields in double, cast for init
starkphi
committed
b288285
use gtx.sqrt directly
starkphi
committed
55d35bf
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
5aed3aa

Commits on Jun 18, 2026

Commits on Jun 19, 2026

Commits on Jun 22, 2026

fix standalone driver datatests for single
starkphi
committed
e00e698
make muphys test single precision ready
starkphi
committed
1c0a480

Commits on Jun 23, 2026

Commits on Jun 24, 2026

revert to atol=0.0 as default
starkphi
committed
e6011a9

Commits on Jun 26, 2026

use direct call to wpfloat and make SaturationAdjustmentConfig more robust
starkphi
committed
d49256b
use math from gtx
starkphi
committed
5235eda

Commits on Jun 29, 2026

[tmp] cast math functions on scalars to wpfloat (with gtx.astype)
starkphi
committed
be0d204
small casting fix
starkphi
committed
057741b
pre-commit changes (ruff)
starkphi
committed
6b31915

Commits on Jun 30, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
6d3c982
make sure type is a float
starkphi
committed
5f8bf39
have to use np.astype directly here (no gtx.field)
starkphi
committed
5a16537
fix merge error
starkphi
committed
32ee9e6
more wpfloat casts
starkphi
committed
0e4eaa7
more casts and single precision specific tolerance
starkphi
committed
4d3f8c0

Commits on Jul 22, 2026

Commits on Jul 23, 2026

adjust tols for single
starkphi
committed
1791ccd

Commits on Jul 24, 2026

adjust tols for single
starkphi
committed
0c0db75
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
d1f40d1

Commits on Jul 27, 2026

precision adjustment for changes that were merged in from main
starkphi
committed
52eb451
directly calculate vct_a/_b in wp
starkphi
committed
cd837c9
adjust tols for single
starkphi
committed
7e03aa9
readd ruff hint (merge error)
starkphi
committed
a84b03a
rm comment
starkphi
committed
98f387a
mention single-precision testing
starkphi
committed
db0ca62
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
413e445
add single tests to CI
starkphi
committed
a9fce47
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
0f7a728
double and single to default?
starkphi
committed
bc34a21

Commits on Jul 28, 2026

fix __post_init__duplicate (merge error)
starkphi
committed
1d1ab92
Revert "[ok?] add break condition for faster runs in single"
starkphi
committed
d1871b1
only cast if not None
starkphi
committed
24fd9e9
pre-commit formatting changes
starkphi
committed
ae13076
Merge branch 'add_single_precision_dycore_part1' of github.com:C2SM/icon4py into add_single_precision_dycore_part1
starkphi
committed
eb6957a
changes to make pre-commit happy
starkphi
committed
5bc2bec
fix reverting error
starkphi
committed
a0ed564
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
bacc214
more pre-commit oriented changes
starkphi
committed
aede9c4
more pre-commit
starkphi
committed
bbc1985

Commits on Jul 30, 2026

fix errors I introduced earlier
starkphi
committed
180d8e1

Commits on Jul 31, 2026

muphys and microphysics integration tests single_precision_ready
starkphi
committed
af35539

Commits on Aug 3, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
c5c7572

Commits on Aug 4, 2026

use hash of job name for build
starkphi
committed
ffcf033

Commits on Aug 12, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
2bc8d72

Commits on Aug 24, 2026

Rename precision env var
starkphi
committed
22c5a3b

Commits on Aug 25, 2026

readding mdformat-myst
starkphi
committed
d52c59f
change tols and fix type errors
starkphi
committed
64ee7ce
pre-commit
starkphi
committed
5e6f41b

Commits on Aug 26, 2026

Keep the ta.{wp,vp}float spelling where main uses it

starkphi
and
claude
committed
f4b0466

Commits on Aug 27, 2026

add return type and fix a type warning
starkphi
committed
df8ec66
Wire PRECISION_VARIANTS through the CI pipeline generator

starkphi
and
claude
committed
646612b
Use GTXFieldType as return type for FieldSource.export_field

starkphi
and
claude
committed
d064583

Commits on Aug 28, 2026

Update ci/default.yml

starkphi
and
msimberg
authored
0f527a2

Commits on Aug 31, 2026

Commits on Sep 1, 2026

Make {w,v}pfloat use a little more consistent
starkphi
committed
3094ed3

Commits on Sep 2, 2026

rm comment
starkphi
committed
d14f1f4
remove cast to int
starkphi
committed
611f658
simplify TypeVar T, inline state_utils.to_data_array
starkphi
committed
8ec043a
ci: normalize float precision to a matrix dimension

starkphi
and
claude
committed
eee65c3
fix max_nudging_coeffient use
starkphi
committed
6be2df2
pre-commit changes
starkphi
committed
4f42e9d

Commits on Sep 4, 2026

mypy: remove broad valid-type ignore
starkphi
committed
9ff0fb7

Commits on Sep 8, 2026

Update .cscs-ci/merge.yml

starkphi
and
msimberg
authored
48097c1
Revert old change
starkphi
authored
e60f3a8
clean up
starkphi
committed
72b8f6c
Apply suggestion from @msimberg

starkphi
and
msimberg
authored
d10d30b

Commits on Sep 9, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
33f288e
uniformly use _set_constant_.. style
starkphi
committed
a908017

Commits on Sep 10, 2026

Commits on Sep 11, 2026

Revert "remove cast to int"
Philipp Stark
committed
1ecce6c
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
4f8447d

Commits on Sep 14, 2026

Commits on Sep 15, 2026

Fix test command for single-precision execution

starkphi
and
Copilot
authored
dbbb3ee

Commits on Sep 16, 2026

Rename export_field -> get (#1472)

authored
16c8dad

Commits on Sep 17, 2026

Update model/atmosphere/diffusion/src/icon4py/model/atmosphere/diffusion/diffusion.py

starkphi
and
msimberg
authored
65072a0

Commits on Sep 18, 2026

Merge branch 'add_single_precision_dycore_part1' of github.com:C2SM/icon4py into add_single_precision_dycore_part1
starkphi
committed
5ddf51a
add missing cast
starkphi
committed
9f00fd9
move double precision variable one level up for benchmarks' yml
starkphi
committed
3857c92
let <no dtype> be wpfloat
starkphi
committed
56c217c
add missing casts
starkphi
committed
c172087
change tols for single
starkphi
committed
9c6023b
ruff formatting
starkphi
committed
19f7f87
remove unnecessary line
starkphi
committed
cd86826
agent updated agent file
starkphi
committed
53ac11a
use 'get' -> for double precision it is no different than get_full_precision
starkphi
committed
b42acb1

Commits on Sep 19, 2026

remove weird exchange "or"
starkphi
committed
25643c4

Commits on Sep 21, 2026

Commits on Sep 22, 2026

Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
f169bfe
use dtime fct in tests
starkphi
committed
562a634
fix typo
starkphi
committed
8be974b
make mypy happy
starkphi
committed
e388163
fix dtypes in diffusion
starkphi
committed
36e0e53
more single precision tests in diffusion
starkphi
committed
567696a

Commits on Sep 24, 2026

remove get_wpfloat and get_int32
Philipp Stark
committed
54fa65c
use astype only if needed
Philipp Stark
committed
e787b2d
fix typo
Philipp Stark
committed
f6bb879
use -HUGE instead of eps as fill value for cold pool enhancement

Philipp Stark
and
claude
committed
5f464ed
fix timedelta with float32 dt in muphys component datatest

Philipp Stark
and
claude
committed
9706af2
restore strict dependency/param validation in NumpyDataProvider

Philipp Stark
and
claude
committed
9cd61df

Commits on Sep 25, 2026

more relyable attributes selection for dataclass_float_to_wp
Philipp Stark
committed
aab150d
keep vertical grid computations in double precision

Philipp Stark
and
claude
committed
25d0de3
store vct_a/vct_b as plain float64 fields of VerticalGrid

Philipp Stark
and
claude
committed
4da177f
add data-free tests for the dtype cast in FieldSource.get/get_scalar

Philipp Stark
and
claude
committed
2c8485d
pass working precision vct_a to en_smag_fac_for_zero_nshift in tests

Philipp Stark
and
claude
committed
c7fac46

Commits on Sep 28, 2026

mainly use atols for single
starkphi
committed
99d3f6c
compute literal-only constant expressions in f64 and round once

starkphi
and
claude
committed
523c120
format test_solve_nonhydro.py

starkphi
and
claude
committed
f92acb2
bump tight single-precision atols to ~2x the measured error

starkphi
and
claude
committed
96f326d

Commits on Sep 29, 2026

Commits on Sep 30, 2026

fix some type errors
starkphi
committed
149a156
register cattrs structure hook for wpfloat (np.float32 in single precision)

starkphi
and
claude
committed
1740754
cast max_nudging_coefficient to wpfloat
starkphi
committed
ab04309

Commits on Oct 1, 2026

raise TypeError instead of assert
starkphi
committed
591261b
register cattrs hooks for np.float32 and np.float64 independent of precision

starkphi
and
claude
committed
68317a3
write config floats with float(v), exposing the float32 representation

starkphi
and
claude
committed
bb9e99b
add _wp variants for tmx helpers and use them in tmx diagnostics

starkphi
and
claude
committed
aef4b6a
tmx: cast scalar parameters to wpfloat and mark diagnostics datatest single_precision_ready

starkphi
and
claude
committed
32cb3a6
patch tmx tols
starkphi
committed
546406d
Merge branch 'main' into add_single_precision_dycore_part1
starkphi
committed
d5d6b74
ignore gt4py mypy plugin float blurring in numpy-floats config test

starkphi
and
claude
committed
dd603ab
pre-commit fixes
starkphi
committed
b1c613c