Add strong scaling GPU actions table

experiments/eval/data/actions-normalized.csv

@@ -0,0 +1,4 @@
+label,BC:cells,BC:dynamicecm,Sweep:Cells,Sweep:DynamicECM,Sweep:DynamicECMDampers,Filling,Other
+CPU,0.04629667717328299,3.88243023434036,0.002164069189058585,82.18255346583861,13.842310640806883,0.005797982839735358,0.03844692981207934
+GPU,0.034897486509106485,41.227398706349675,13.180664898871365,22.432483220415783,19.749714111977486,1.9939606413394697,1.3808809345371023
+Booster,0.09128030041871667,42.736382370072874,12.260298369798207,22.36492489089858,16.070196185259338,4.364304866307413,2.1126130172448816

experiments/eval/data/actions.csv

@@ -0,0 +1,4 @@
+label,BC:cells,BC:dynamicecm,Sweep:Cells,Sweep:DynamicECM,Sweep:DynamicECMDampers,Other
+CPU,0.9047934333333334,75.87579921250001,0.04229322083333333,1606.12465625,270.5255006333333,0.8646950250000001
+GPU,0.11372260000000001,134.35027680000002,42.952648800000006,73.102122,64.3596162,10.9978052
+Booster,0.155221,72.67267939999999,20.848482800000003,38.0312728,27.327165999999995,11.013917200000002

experiments/eval/data/strong-gpu-actions.csv

@@ -0,0 +1,9 @@
+label,BC:cells,BC:dynamicecm,Sweep:Cells,Sweep:DynamicECM,Sweep:DynamicECMDampers,Filling,Other
+1,0.034897486509106485,41.227398706349675,13.180664898871365,22.432483220415783,19.749714111977486,1.9939606413394697,1.3808809345371023
+2,0.7137918070198439,50.7137377526309,8.923981074147767,19.35772833011394,17.571497161475467,1.320924803923136,1.3983390706889622
+4,0.6647403410134337,58.291944402330074,7.087118055335309,16.67352093599457,15.236952119238136,1.092645293437499,0.953078852650984
+8,0.6788278363464717,65.05900206274784,5.2089254465999275,14.249993130938188,13.176763096049008,0.7851584656500297,0.8413299616685318
+16,1.441574901760373,61.1245906198054,4.832237599492141,16.029377636605524,14.996293924006913,0.716990077435801,0.858935240893841
+32,3.3736940900007966,47.94830950907802,5.622259685491874,21.238716402604393,19.97281149065964,0.8196124452570666,1.0245963769081996
+64,6.284545018581091,45.28808866195176,4.502683326147901,21.66574860147326,20.522789117514915,0.6460791296539419,1.0900661446771265
+128,11.788574344478201,40.04977497065291,3.2677482351392375,22.1995263401806,21.209134936528987,0.4598839494487746,1.0253572235712836

experiments/eval/data/weak-booster.tsv

@@ -0,0 +1,9 @@
+nodes	tasks	mean_time	std_time	efficiency	efficiency_error
+1	1	41.98463099999999	1.4067920862965122	1.0	0.023219381860496473
+1	2	49.85189340000001	1.7669363566083227	0.8421872899214694	0.020685165974491628
+1	4	59.9694977	2.209151778903949	0.7000997608822725	0.01787171173813342
+2	8	72.810012625	0.5538313058789631	0.5766326565034021	0.003039461920431096
+4	16	84.718001525	5.20969191162485	0.4955809892140865	0.021118450727088397
+8	32	87.22569569999999	1.7018797711775082	0.4813332890390463	0.0065079080866405825
+16	64	85.36740952500001	1.0197522041876155	0.49181099946232654	0.004071100890302765
+32	128	87.5995021046875	0.8707125543332932	0.47927933368645675	0.003301208459097849

experiments/eval/data/weak-cpu.tsv

@@ -0,0 +1,10 @@
+nodes	tasks	mean_time	std_time	efficiency	efficiency_error
+1	48	234.09125610833334	4.9756238110995765	1.0	0.013886639526655382
+2	96	234.09269640000002	1.9886350551995884	0.9999938473447108	0.005550081636855162
+4	192	239.06532950520833	1.8931447841474947	0.9791936647310181	0.005066063671757703
+8	384	241.32357499479167	0.977728668984766	0.9700306160033705	0.0025676679349660694
+16	768	260.8008324059896	12.802824235961744	0.8975863073317291	0.028787758154736998
+32	1536	266.59983470546877	13.970338204797093	0.8780622702446538	0.03006125980502309
+64	3072	281.61153985800786	25.95560834859075	0.8312559074332151	0.05005532970709546
+128	6144	293.4177162058594	10.8076489888669	0.7978088683101088	0.019198998510810975
+256	12288	310.9775802693196	14.459893325743863	0.7527592693518308	0.022867934404056486

experiments/eval/data/weak-gpu.tsv

@@ -0,0 +1,9 @@
+nodes	tasks	mean_time	std_time	efficiency	efficiency_error
+1	1	60.658062599999994	1.8711952545625232	1.0	0.021376748626269895
+1	2	80.6903592	1.1554429061827625	0.751738661240214	0.007459415624087278
+1	4	89.01234124999999	1.3979557323025515	0.6814567704677693	0.00741639040919915
+2	8	101.357940575	4.784240687736861	0.5984539766286584	0.01957478845779047
+4	16	106.2530252	1.5044057675724636	0.5708831582519497	0.005601211794070455
+8	32	113.70197445	2.069642901549501	0.5334829310873062	0.006729131548114211
+16	64	119.22770070624999	0.7600051865549382	0.5087581345667959	0.00224730403456563
+32	128	121.8124414140625	0.5170666166834136	0.4979627852118346	0.0014647476739746731

experiments/eval/scripts/generate-weak-configs.py

@@ -129,6 +129,42 @@ def make_cpu_ex(x: int, y: int, z: int) -> Experiment:
     )
 
 
+def make_gpu_ex(x: int, y: int, z: int) -> Experiment:
+    num_blocks = x * y * z
+    gpus_per_node = num_blocks if num_blocks <= 4 else 4
+    num_nodes = 1 if num_blocks <= 4 else num_blocks // 4
+
+    return Experiment(
+        job_name=f"weak-gpu-{x:02}-{y:02}-{z:02}",
+        account="cellsinsilico",
+        partition="gpus",
+        extra_sbatch_line=f"#SBATCH --gres=gpu:{gpus_per_node}",
+        nastja_binary_path="/p/project/cellsinsilico/paulslustigebude/nastja/build-cuda/nastja",
+        nodes=num_nodes,
+        tasks=num_blocks,
+        num_blocks=(x, y, z),
+        domain_scale=(x, y, z),
+    )
+
+
+def make_booster_ex(x: int, y: int, z: int) -> Experiment:
+    num_blocks = x * y * z
+    gpus_per_node = num_blocks if num_blocks <= 4 else 4
+    num_nodes = 1 if num_blocks <= 4 else num_blocks // 4
+
+    return Experiment(
+        job_name=f"weak-booster-{x:02}-{y:02}-{z:02}",
+        account="hkf6",
+        partition="booster",
+        extra_sbatch_line=f"#SBATCH --gres=gpu:{gpus_per_node}",
+        nastja_binary_path="/p/project/cellsinsilico/paulslustigebude/nastja/build-cuda/nastja",
+        nodes=num_nodes,
+        tasks=num_blocks,
+        num_blocks=(x, y, z),
+        domain_scale=(x, y, z),
+    )
+
+
 experiments = [
     make_cpu_ex(4, 4, 3),
     make_cpu_ex(4, 4, 6),
@@ -138,6 +174,25 @@ experiments = [
     make_cpu_ex(8, 8, 24),
     make_cpu_ex(8, 16, 24),
     make_cpu_ex(16, 16, 24),
+    make_cpu_ex(16, 16, 48),
+
+    make_gpu_ex(1, 1, 1),
+    make_gpu_ex(1, 1, 2),
+    make_gpu_ex(1, 2, 2),
+    make_gpu_ex(2, 2, 2),
+    make_gpu_ex(2, 2, 4),
+    make_gpu_ex(2, 4, 4),
+    make_gpu_ex(4, 4, 4),
+    make_gpu_ex(4, 4, 8),
+
+    make_booster_ex(1, 1, 1),
+    make_booster_ex(1, 1, 2),
+    make_booster_ex(1, 2, 2),
+    make_booster_ex(2, 2, 2),
+    make_booster_ex(2, 2, 4),
+    make_booster_ex(2, 4, 4),
+    make_booster_ex(4, 4, 4),
+    make_booster_ex(4, 4, 8),
 ]
 
 if __name__ == "__main__":

experiments/eval/scripts/make-latex-table.py

@@ -12,9 +12,13 @@ def show_seconds(s: float) -> str:
 if __name__ == '__main__':
     p = argparse.ArgumentParser(description="Make a latex table from a timings tsv")
     p.add_argument("timingfile")
+    p.add_argument("--weak", action="store_true")
     args = p.parse_args()
 
     df = pandas.read_csv(args.timingfile, sep="\t")
 
     for i in range(len(df)):
+        if not args.weak:
             print(f"{df['nodes'][i]} & {df['tasks'][i]} & {show_seconds(df['mean_time'][i])} & {show_seconds(df['std_time'][i])} & {df['speedup'][i]:.02f} & {df['speedup_error'][i]:.02f} \\\\")
+        else:
+            print(f"{df['nodes'][i]} & {df['tasks'][i]} & {show_seconds(df['mean_time'][i])} & {show_seconds(df['std_time'][i])} & {df['efficiency'][i]:.02f} & {df['efficiency_error'][i]:.02f} \\\\")

experiments/eval/scripts/timing-actions.py

@@ -0,0 +1,61 @@
+#!/usr/bin/env python
+
+
+import argparse
+import sys
+
+
+import timing
+
+
+ignore = ["TimeStep"]
+
+
+if __name__ == "__main__":
+    p = argparse.ArgumentParser()
+    p.add_argument("jobs", nargs="+")
+    p.add_argument("--normalize", action="store_true")
+    p.add_argument("--extra-columns", nargs="*")
+    args = p.parse_args()
+
+    columns = [
+        "BC:cells",
+        "BC:dynamicecm",
+        "Sweep:Cells",
+        "Sweep:DynamicECM",
+        "Sweep:DynamicECMDampers",
+    ] + (args.extra_columns or [])
+
+    dfs = dict()
+    labels = []
+    for label, jobid in [jobarg.split(":") for jobarg in args.jobs]:
+        jobs, excluded_array_indices = timing.get_jobs(jobid)
+        df = timing.load_array_mean_timings(jobid, excluded_array_indices).mean()
+        dfs[label] = df
+        labels.extend(df.index)
+
+    labels = set(labels)
+    print(",".join(["label"] + columns + ["Other"]))
+
+    values_by_label = dict()
+    for label, df in dfs.items():
+        values = {"Other": 0}
+        for c in df.index:
+            if c in ignore:
+                continue
+            elif c not in columns:
+                values["Other"] += df[c]
+                print(f"Others+= {c}={df[c]}", file=sys.stderr)
+            else:
+                values[c] = df[c]
+        values_by_label[label] = values
+
+    if args.normalize:
+        print("Normalizing data to 100%...", file=sys.stderr)
+        for values in values_by_label.values():
+            row_length = sum(values.values())
+            for c in values.keys():
+                values[c] *= 100 / row_length
+
+    for label, values in values_by_label.items():
+        print(label + "," + ",".join(f"{values[c]}" for c in columns + ["Other"]))

experiments/eval/scripts/timing.py

@@ -72,18 +72,8 @@ def get_accounting_data(jobid: str):
     return json.loads(sacct_results.stdout.decode("utf8"))
 
 
-if __name__ == "__main__":
-    p = argparse.ArgumentParser(description="Load and analzye data from nastja timing files")
-    p.add_argument("jobid", nargs="+")
-    p.add_argument("--prettify", action="store_true")
-    p.add_argument("--dump-timings", action="store_true")
-    args = p.parse_args()
-
-    results = []
-    for i, jobid in enumerate(args.jobid, 1):
-        print(f"({i:2}/{len(args.jobid):2}) Loading accounting data for {jobid}", file=sys.stderr)
+def get_jobs(jobid: str):
     accounting_data = get_accounting_data(jobid)
-
     jobs = []
     excluded_array_indices = []
     for array_job in accounting_data["jobs"]:
@@ -99,6 +89,20 @@ if __name__ == "__main__":
             continue
         jobs.append(last_step)
 
+    return jobs, excluded_array_indices
+
+
+if __name__ == "__main__":
+    p = argparse.ArgumentParser(description="Load and analzye data from nastja timing files")
+    p.add_argument("jobid", nargs="+")
+    p.add_argument("--prettify", action="store_true")
+    p.add_argument("--dump-timings", action="store_true")
+    args = p.parse_args()
+
+    results = []
+    for i, jobid in enumerate(args.jobid, 1):
+        print(f"({i:2}/{len(args.jobid):2}) Loading accounting data for {jobid}", file=sys.stderr)
+        jobs, excluded_array_indices = get_jobs(jobid)
         array_mean_timings = load_array_mean_timings(jobid, excluded_array_indices)
         if args.dump_timings:
             print(array_mean_timings, file=sys.stderr)