Skip to content

Commit a9a8713

Browse files
szy21szy21
authored
Merge pull request #3917 from CliMA/zs/common_config
add common numerics configs
2 parents 0aabab3 + defb6da commit a9a8713

File tree

84 files changed

+345
-446
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

84 files changed

+345
-446
lines changed

‎.buildkite/gpu_pipeline/pipeline.yml

Lines changed: 19 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -11,6 +11,7 @@ env:
1111
OMPI_MCA_opal_warn_on_missing_libcuda: 0
1212
SLURM_KILL_BAD_EXIT: 1
1313
SLURM_GRES_FLAGS: "allow-task-sharing"
14+
COMMON_CONFIG_PATH: "config/common_configs/"
1415
GPU_CONFIG_PATH: "config/gpu_configs/"
1516
MODEL_CONFIG_PATH: "config/model_configs/"
1617
CLIMAATMOS_GC_NSTEPS: 10
@@ -44,6 +45,7 @@ steps:
4445
- >
4546
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=baroclinic_wave_helem30/output_active/report
4647
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
48+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze43.yml
4749
--config_file ${GPU_CONFIG_PATH}baroclinic_wave_helem30.yml
4850
--job_id baroclinic_wave_helem30
4951
artifact_paths: "baroclinic_wave_helem30/output_active/*"
@@ -62,6 +64,7 @@ steps:
6264
srun --cpu-bind=threads --cpus-per-task=4
6365
nsys profile --delay 100 --trace=osrt,nvtx,cuda,mpi,ucx --output=baroclinic_wave_helem30_4process/output_active/report-%q{PMI_RANK}
6466
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
67+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze43.yml
6568
--config_file ${GPU_CONFIG_PATH}baroclinic_wave_helem30.yml
6669
--job_id baroclinic_wave_helem30_4process
6770
artifact_paths: "baroclinic_wave_helem30_4process/output_active/*"
@@ -81,6 +84,7 @@ steps:
8184
- >
8285
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=held_suarez_equil_helem30/output_active/report
8386
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
87+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
8488
--config_file ${GPU_CONFIG_PATH}held_suarez_equil_helem30.yml
8589
--job_id held_suarez_equil_helem30
8690
artifact_paths: "held_suarez_equil_helem30/output_active/*"
@@ -99,6 +103,7 @@ steps:
99103
srun --cpu-bind=threads --cpus-per-task=4
100104
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=held_suarez_equil_helem30_4process/output_active/report-%q{PMI_RANK}
101105
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
106+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
102107
--config_file ${GPU_CONFIG_PATH}held_suarez_equil_helem30.yml
103108
--job_id held_suarez_equil_helem30_4process
104109
artifact_paths: "held_suarez_equil_helem30_4process/output_active/*"
@@ -120,6 +125,7 @@ steps:
120125
- >
121126
srun --cpu-bind=threads --cpus-per-task=4
122127
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=gpu_aquaplanet_dyamond_diag_1process/output_active/report julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
128+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
123129
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_diag_1process.yml
124130
--job_id gpu_aquaplanet_dyamond_diag_1process
125131
artifact_paths: "gpu_aquaplanet_dyamond_diag_1process/output_active/*"
@@ -141,6 +147,7 @@ steps:
141147
srun --cpu-bind=threads --cpus-per-task=4
142148
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=gpu_aquaplanet_dyamond_ss_1process/output_active/report
143149
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
150+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
144151
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ss.yml
145152
--job_id gpu_aquaplanet_dyamond_ss_1process
146153
artifact_paths: "gpu_aquaplanet_dyamond_ss_1process/output_active/*"
@@ -161,6 +168,7 @@ steps:
161168
- >
162169
srun --cpu-bind=threads --cpus-per-task=4
163170
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
171+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
164172
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ss.yml
165173
--job_id gpu_aquaplanet_dyamond_ss_2process
166174
artifact_paths: "gpu_aquaplanet_dyamond_ss_2process/output_active/*"
@@ -181,6 +189,7 @@ steps:
181189
- >
182190
srun --cpu-bind=threads --cpus-per-task=4
183191
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
192+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
184193
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ss.yml
185194
--job_id gpu_aquaplanet_dyamond_ss_4process
186195
artifact_paths: "gpu_aquaplanet_dyamond_ss_4process/output_active/*"
@@ -219,6 +228,7 @@ steps:
219228
- >
220229
srun --cpu-bind=threads --cpus-per-task=4
221230
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
231+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
222232
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ws_1process.yml
223233
--job_id gpu_aquaplanet_dyamond_ws_1process
224234
artifact_paths: "gpu_aquaplanet_dyamond_ws_1process/output_active/*"
@@ -239,6 +249,7 @@ steps:
239249
- >
240250
srun --cpu-bind=threads --cpus-per-task=4
241251
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
252+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
242253
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ws_2process.yml
243254
--job_id gpu_aquaplanet_dyamond_ws_2process
244255
artifact_paths: "gpu_aquaplanet_dyamond_ws_2process/output_active/*"
@@ -259,6 +270,7 @@ steps:
259270
- >
260271
srun --cpu-bind=threads --cpus-per-task=4
261272
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
273+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he30ze63.yml
262274
--config_file ${GPU_CONFIG_PATH}gpu_aquaplanet_dyamond_ws_4process.yml
263275
--job_id gpu_aquaplanet_dyamond_ws_4process
264276
artifact_paths: "gpu_aquaplanet_dyamond_ws_4process/output_active/*"
@@ -300,6 +312,7 @@ steps:
300312
- >
301313
nsys profile --delay 200 --trace=nvtx,mpi,cuda,osrt --output=gpu_aquaplanet_diagedmf/output_active/report
302314
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
315+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he16ze63.yml
303316
--config_file ${MODEL_CONFIG_PATH}aquaplanet_diagedmf.yml
304317
--job_id gpu_aquaplanet_diagedmf
305318
artifact_paths: "gpu_aquaplanet_diagedmf/output_active/*"
@@ -316,7 +329,8 @@ steps:
316329
- label: "gpu_aquaplanet_diagedmf_benchmark"
317330
command: >
318331
julia --color=yes --project=.buildkite perf/benchmark.jl
319-
--config_file ${MODEL_CONFIG_PATH}aquaplanet_diagedmf.yml
332+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he16ze63.yml
333+
--config_file ${MODEL_CONFIG_PATH}diagnostic_edmfx_aquaplanet.yml
320334
--job_id gpu_aquaplanet_diagedmf_benchmark
321335
artifact_paths: "gpu_aquaplanet_diagedmf_benchmark/output_active/*"
322336
env:
@@ -334,7 +348,8 @@ steps:
334348
- >
335349
nsys profile --delay 100 --trace=nvtx,mpi,cuda,osrt --output=gpu_aquaplanet_progedmf/output_active/report
336350
julia --threads=3 --color=yes --project=.buildkite .buildkite/ci_driver.jl
337-
--config_file ${MODEL_CONFIG_PATH}aquaplanet_progedmf.yml
351+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he16ze63.yml
352+
--config_file ${MODEL_CONFIG_PATH}prognostic_edmfx_aquaplanet.yml
338353
--job_id gpu_aquaplanet_progedmf
339354
artifact_paths: "gpu_aquaplanet_progedmf/output_active/*"
340355
env:
@@ -350,7 +365,8 @@ steps:
350365
- label: "gpu_aquaplanet_progedmf_benchmark"
351366
command: >
352367
julia --color=yes --project=.buildkite perf/benchmark.jl
353-
--config_file ${MODEL_CONFIG_PATH}aquaplanet_progedmf.yml
368+
--config_file ${COMMON_CONFIG_PATH}numerics_sphere_he16ze63.yml
369+
--config_file ${MODEL_CONFIG_PATH}prognostic_edmfx_aquaplanet.yml
354370
--job_id gpu_aquaplanet_progedmf_benchmark
355371
artifact_paths: "gpu_aquaplanet_progedmf_benchmark/output_active/*"
356372
env:

‎.buildkite/longruns_gpu/pipeline.yml

Lines changed: 73 additions & 32 deletions
Original file line numberDiff line numberDiff line change
@@ -12,6 +12,7 @@ env:
1212
SLURM_KILL_BAD_EXIT: 1
1313
SLURM_GRES_FLAGS: "allow-task-sharing"
1414
CONFIG_PATH: "config/longrun_configs"
15+
COMMON_CONFIG_PATH: "config/common_configs"
1516
CLIMAATMOS_GC_NSTEPS: 10
1617

1718
timeout_in_minutes: 1440
@@ -39,8 +40,11 @@ steps:
3940
steps:
4041

4142
- label: ":computer: hydrostatic balance"
42-
command:
43-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
43+
command: >
44+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
45+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze63.yml
46+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
47+
--job_id $$JOB_NAME
4448
artifact_paths: "$$JOB_NAME/output_active/*"
4549
agents:
4650
slurm_gpus: 1
@@ -50,8 +54,11 @@ steps:
5054
JOB_NAME: "longrun_hydrostatic_balance"
5155

5256
- label: ":computer: dry baroclinic wave"
53-
command:
54-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
57+
command: >
58+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
59+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze43.yml
60+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
61+
--job_id $$JOB_NAME
5562
artifact_paths: "$$JOB_NAME/output_active/*"
5663
agents:
5764
slurm_gpus: 1
@@ -61,8 +68,11 @@ steps:
6168
JOB_NAME: "longrun_dry_baroclinic_wave"
6269

6370
- label: ":computer: dry baroclinic wave high res"
64-
command:
65-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
71+
command: >
72+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
73+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze43.yml
74+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
75+
--job_id $$JOB_NAME
6676
artifact_paths: "$$JOB_NAME/output_active/*"
6777
agents:
6878
slurm_gpus: 1
@@ -72,8 +82,11 @@ steps:
7282
JOB_NAME: "longrun_dry_baroclinic_wave_he60"
7383

7484
- label: ":computer: baroclinic wave equilmoist"
75-
command:
76-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
85+
command: >
86+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
87+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze43.yml
88+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
89+
--job_id $$JOB_NAME
7790
artifact_paths: "$$JOB_NAME/output_active/*"
7891
agents:
7992
slurm_gpus: 1
@@ -83,8 +96,11 @@ steps:
8396
JOB_NAME: "longrun_moist_baroclinic_wave"
8497

8598
- label: ":computer: baroclinic wave equilmoist high res"
86-
command:
87-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
99+
command: >
100+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
101+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze43.yml
102+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
103+
--job_id $$JOB_NAME
88104
artifact_paths: "$$JOB_NAME/output_active/*"
89105
agents:
90106
slurm_gpus: 1
@@ -94,8 +110,11 @@ steps:
94110
JOB_NAME: "longrun_moist_baroclinic_wave_he60"
95111

96112
- label: ":computer: dry held-suarez"
97-
command:
98-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
113+
command: >
114+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
115+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze63.yml
116+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
117+
--job_id $$JOB_NAME
99118
artifact_paths: "$$JOB_NAME/output_active/*"
100119
agents:
101120
slurm_gpus: 1
@@ -105,8 +124,11 @@ steps:
105124
JOB_NAME: "longrun_dry_held_suarez"
106125

107126
- label: ":computer: held-suarez, equilmoist"
108-
command:
109-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
127+
command: >
128+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
129+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze63.yml
130+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
131+
--job_id $$JOB_NAME
110132
artifact_paths: "$$JOB_NAME/output_active/*"
111133
agents:
112134
slurm_gpus: 1
@@ -119,8 +141,10 @@ steps:
119141
steps:
120142

121143
- label: ":computer: aquaplanet equilmoist allsky radiation + 0M microphysics"
122-
command:
123-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
144+
command: >
145+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
146+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
147+
--job_id $$JOB_NAME
124148
artifact_paths: "$$JOB_NAME/output_active/*"
125149
agents:
126150
slurm_gpus: 1
@@ -130,8 +154,10 @@ steps:
130154
JOB_NAME: "longrun_aquaplanet_allsky_0M"
131155

132156
- label: ":computer: aquaplanet equilmoist allsky radiation + diagnostic edmf + 0M microphysics"
133-
command:
134-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
157+
command: >
158+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
159+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
160+
--job_id $$JOB_NAME
135161
artifact_paths: "$$JOB_NAME/output_active/*"
136162
agents:
137163
slurm_gpus: 1
@@ -141,8 +167,10 @@ steps:
141167
JOB_NAME: "longrun_aquaplanet_allsky_diagedmf_0M"
142168

143169
- label: ":computer: aquaplanet equilmoist allsky radiation + prognostic edmf + 0M microphysics"
144-
command:
145-
- srun julia --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
170+
command: >
171+
srun julia --project=.buildkite .buildkite/ci_driver.jl
172+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
173+
--job_id $$JOB_NAME
146174
artifact_paths: "$$JOB_NAME/output_active/*"
147175
agents:
148176
slurm_gpus: 1
@@ -152,8 +180,10 @@ steps:
152180
JOB_NAME: "longrun_aquaplanet_allsky_progedmf_0M"
153181

154182
- label: ":computer: aquaplanet equilmoist allsky radiation + 0M microphysics + earth topography"
155-
command:
156-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
183+
command: >
184+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
185+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
186+
--job_id $$JOB_NAME
157187
artifact_paths: "$$JOB_NAME/output_active/*"
158188
agents:
159189
slurm_gpus: 1
@@ -163,8 +193,10 @@ steps:
163193
JOB_NAME: "longrun_aquaplanet_allsky_0M_earth"
164194

165195
- label: ":umbrella: aquaplanet non-equil moisture allsky radiation + 1M microphysics"
166-
command:
167-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
196+
command: >
197+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
198+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
199+
--job_id $$JOB_NAME
168200
artifact_paths: "$$JOB_NAME/output_active/*"
169201
agents:
170202
slurm_gpus: 1
@@ -174,8 +206,10 @@ steps:
174206
JOB_NAME: "longrun_aquaplanet_allsky_1M"
175207

176208
- label: ":computer: aquaplanet equilmoist allsky radiation + time-varying insolation + 0M microphysics + slab ocean"
177-
command:
178-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
209+
command: >
210+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
211+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
212+
--job_id $$JOB_NAME
179213
artifact_paths: "$$JOB_NAME/output_active/*"
180214
agents:
181215
slurm_gpus: 1
@@ -189,8 +223,11 @@ steps:
189223
steps:
190224

191225
- label: ":computer: aquaplanet dyamond"
192-
command:
193-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
226+
command: >
227+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
228+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he30ze63.yml
229+
--config_file $CONFIG_PATH/$$JOB_NAME.yml
230+
--job_id $$JOB_NAME
194231
artifact_paths: "$$JOB_NAME/output_active/*"
195232
agents:
196233
slurm_gpus: 1
@@ -204,8 +241,10 @@ steps:
204241
steps:
205242

206243
- label: ":computer: amip target diagnostic edmf"
207-
command:
208-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
244+
command: >
245+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
246+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
247+
--job_id $$JOB_NAME
209248
artifact_paths: "$$JOB_NAME/output_active/*"
210249
agents:
211250
slurm_gpus: 1
@@ -215,8 +254,10 @@ steps:
215254
JOB_NAME: "amip_target_diagedmf"
216255

217256
- label: ":computer: amip target edonly edmf"
218-
command:
219-
- srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl --config_file $CONFIG_PATH/$$JOB_NAME.yml --job_id $$JOB_NAME
257+
command: >
258+
srun julia --color=yes --project=.buildkite .buildkite/ci_driver.jl
259+
--config_file $COMMON_CONFIG_PATH/numerics_sphere_he16ze63.yml --config_file $CONFIG_PATH/$$JOB_NAME.yml
260+
--job_id $$JOB_NAME
220261
artifact_paths: "$$JOB_NAME/output_active/*"
221262
agents:
222263
slurm_gpus: 1

0 commit comments

Comments
 (0)