Merge pull request #12 from jeromekelleher/pythran

Pythran

data/tree-performance.csv

@@ -1,73 +1,99 @@
 ,order,implementation,sample_size,time_mean,time_var
-0,msprime,py_numba,10,2.8540134662762286e-05,4.0138743538487945e-12
-1,msprime,py_tskit,10,1.4611827791668475e-05,5.454016455301305e-12
-2,msprime,c_lib,10,4.32e-07,0.0
-3,msprime,c_recursive,10,4.11e-07,0.0
-4,msprime,c_struct_contiguous,10,4.34e-07,0.0
-5,msprime,c_struct_fragmented,10,3.89e-07,0.0
-6,preorder,py_numba,10,2.8566209017299115e-05,2.459800961376813e-11
-7,preorder,py_tskit,10,1.3768869219347835e-05,3.178959404084067e-12
-8,preorder,c_lib,10,4.28e-07,0.0
-9,preorder,c_recursive,10,3.97e-07,0.0
-10,preorder,c_struct_contiguous,10,4.2e-07,0.0
-11,preorder,c_struct_fragmented,10,3.99e-07,0.0
-12,msprime,py_numba,100,7.674175605643541e-05,3.482266007416305e-10
-13,msprime,py_tskit,100,1.6620315029285847e-05,7.567011308450897e-12
-14,msprime,c_lib,100,2.726e-06,0.0
-15,msprime,c_recursive,100,2.945e-06,0.0
-16,msprime,c_struct_contiguous,100,2.865e-06,0.0
-17,msprime,c_struct_fragmented,100,2.74e-06,0.0
-18,preorder,py_numba,100,7.531476335134357e-05,7.18849128501038e-11
-19,preorder,py_tskit,100,1.646423095371574e-05,4.734347490920951e-12
-20,preorder,c_lib,100,2.744e-06,0.0
-21,preorder,c_recursive,100,2.885e-06,0.0
-22,preorder,c_struct_contiguous,100,2.808e-06,0.0
-23,preorder,c_struct_fragmented,100,2.777e-06,0.0
-24,msprime,py_numba,1000,0.000539161802851595,7.013503718161112e-10
-25,msprime,py_tskit,1000,4.834113118704409e-05,4.6100322678013504e-11
-26,msprime,c_lib,1000,3.1756e-05,0.0
-27,msprime,c_recursive,1000,3.3263e-05,0.0
-28,msprime,c_struct_contiguous,1000,3.2627e-05,0.0
-29,msprime,c_struct_fragmented,1000,3.164e-05,0.0
-30,preorder,py_numba,1000,0.0005330153827089817,5.343880201340534e-10
-31,preorder,py_tskit,1000,4.7103321296162906e-05,3.558357414956268e-11
-32,preorder,c_lib,1000,3.1155e-05,0.0
-33,preorder,c_recursive,1000,3.3384e-05,0.0
-34,preorder,c_struct_contiguous,1000,3.1308e-05,0.0
-35,preorder,c_struct_fragmented,1000,3.1941e-05,0.0
-36,msprime,py_numba,10000,0.00532087329775095,2.950362855263974e-08
-37,msprime,py_tskit,10000,0.000503669191733934,3.260384184696477e-10
-38,msprime,c_lib,10000,0.000505534,0.0
-39,msprime,c_recursive,10000,0.000420235,0.0
-40,msprime,c_struct_contiguous,10000,0.000432408,0.0
-41,msprime,c_struct_fragmented,10000,0.000338762,0.0
-42,preorder,py_numba,10000,0.005214149096980691,3.3317304310811266e-08
-43,preorder,py_tskit,10000,0.00033959061093628406,1.7320717704898776e-10
-44,preorder,c_lib,10000,0.000306742,0.0
-45,preorder,c_recursive,10000,0.000337701,0.0
-46,preorder,c_struct_contiguous,10000,0.000327341,0.0
-47,preorder,c_struct_fragmented,10000,0.000337227,0.0
-48,msprime,py_numba,100000,0.05427524673100561,2.4139536645272405e-06
-49,msprime,py_tskit,100000,0.007691314306342975,5.548524184113716e-07
-50,msprime,c_lib,100000,0.00780866,0.0
-51,msprime,c_recursive,100000,0.00716185,0.0
-52,msprime,c_struct_contiguous,100000,0.0067306,0.0
-53,msprime,c_struct_fragmented,100000,0.00429481,0.0
-54,preorder,py_numba,100000,0.051102137439884246,6.899616658855656e-07
-55,preorder,py_tskit,100000,0.003263211432960816,1.3465280043040993e-08
-56,preorder,c_lib,100000,0.0034373,0.0
-57,preorder,c_recursive,100000,0.00361534,0.0
-58,preorder,c_struct_contiguous,100000,0.00380468,0.0
-59,preorder,c_struct_fragmented,100000,0.00390367,0.0
-60,msprime,py_numba,1000000,0.7595824899966829,0.00021048295748450066
-61,msprime,py_tskit,1000000,0.27509022200247274,6.702407744002218e-06
-62,msprime,c_lib,1000000,0.270316,0.0
-63,msprime,c_recursive,1000000,0.207154,0.0
-64,msprime,c_struct_contiguous,1000000,0.193365,0.0
-65,msprime,c_struct_fragmented,1000000,0.0448324,0.0
-66,preorder,py_numba,1000000,0.5094039363902994,9.985812567576443e-07
-67,preorder,py_tskit,1000000,0.04093705428531393,3.841080667539008e-07
-68,preorder,c_lib,1000000,0.0383174,0.0
-69,preorder,c_recursive,1000000,0.0388503,0.0
-70,preorder,c_struct_contiguous,1000000,0.0410454,0.0
-71,preorder,c_struct_fragmented,1000000,0.0397279,0.0
+0,msprime,py_pythran,10,2.5583696900866925e-05,7.809983926760048e-12
+1,msprime,py_numba,10,2.836359350476414e-05,1.7359376706916727e-11
+2,msprime,py_tskit,10,1.3589535839855671e-05,9.184828871558387e-12
+3,msprime,c_lib,10,4.43e-07,0.0
+4,msprime,c_recursive,10,4.11e-07,0.0
+5,msprime,c_struct_contiguous,10,4.18e-07,0.0
+6,msprime,c_struct_fragmented,10,3.91e-07,0.0
+7,preorder,py_pythran,10,2.590082585811615e-05,3.030955428906024e-11
+8,preorder,py_numba,10,2.825020393356681e-05,1.539864884413577e-11
+9,preorder,py_tskit,10,1.3512573204934596e-05,7.730593907037545e-12
+10,preorder,c_lib,10,4.49e-07,0.0
+11,preorder,c_recursive,10,4.59e-07,0.0
+12,preorder,c_struct_contiguous,10,4.34e-07,0.0
+13,preorder,c_struct_fragmented,10,4.03e-07,0.0
+14,msprime,py_pythran,100,5.167797859758139e-05,6.429221542121828e-11
+15,msprime,py_numba,100,7.471199857536703e-05,4.285108452093363e-11
+16,msprime,py_tskit,100,1.6066455864347517e-05,5.76997232335103e-12
+17,msprime,c_lib,100,2.705e-06,0.0
+18,msprime,c_recursive,100,2.99e-06,0.0
+19,msprime,c_struct_contiguous,100,2.877e-06,0.0
+20,msprime,c_struct_fragmented,100,2.808e-06,0.0
+21,preorder,py_pythran,100,5.117504834197462e-05,4.769600366560993e-11
+22,preorder,py_numba,100,7.487921428401023e-05,7.214357494266344e-11
+23,preorder,py_tskit,100,1.6021172399632633e-05,9.526227517346357e-12
+24,preorder,c_lib,100,2.715e-06,0.0
+25,preorder,c_recursive,100,2.926e-06,0.0
+26,preorder,c_struct_contiguous,100,2.876e-06,0.0
+27,preorder,c_struct_fragmented,100,2.76e-06,0.0
+28,msprime,py_pythran,1000,0.0003081394217442721,3.145327963317433e-10
+29,msprime,py_numba,1000,0.0005472428781213239,5.127235068371507e-10
+30,msprime,py_tskit,1000,4.7597346594557164e-05,5.6991689453188005e-11
+31,msprime,c_lib,1000,3.2224e-05,0.0
+32,msprime,c_recursive,1000,3.3759e-05,0.0
+33,msprime,c_struct_contiguous,1000,3.2924e-05,0.0
+34,msprime,c_struct_fragmented,1000,3.2055e-05,0.0
+35,preorder,py_pythran,1000,0.0003074059736682102,3.0694768360113097e-10
+36,preorder,py_numba,1000,0.0005346568721579387,2.1227533897813347e-10
+37,preorder,py_tskit,1000,4.589068912900984e-05,1.2172149457959373e-11
+38,preorder,c_lib,1000,3.0893e-05,0.0
+39,preorder,c_recursive,1000,3.3279e-05,0.0
+40,preorder,c_struct_contiguous,1000,3.1057e-05,0.0
+41,preorder,c_struct_fragmented,1000,3.1633e-05,0.0
+42,msprime,py_pythran,10000,0.0029379673992516475,6.2762750953491905e-09
+43,msprime,py_numba,10000,0.0051710225654533135,1.524685222496294e-08
+44,msprime,py_tskit,10000,0.0004982919344911352,8.841957693966572e-10
+45,msprime,c_lib,10000,0.000486727,0.0
+46,msprime,c_recursive,10000,0.000402896,0.0
+47,msprime,c_struct_contiguous,10000,0.000431504,0.0
+48,msprime,c_struct_fragmented,10000,0.000339091,0.0
+49,preorder,py_pythran,10000,0.0028233553573954852,1.3426214560934356e-09
+50,preorder,py_numba,10000,0.005058508399641141,2.422371675065888e-09
+51,preorder,py_tskit,10000,0.0003375611142255366,1.216721460301983e-10
+52,preorder,c_lib,10000,0.000305847,0.0
+53,preorder,c_recursive,10000,0.000336043,0.0
+54,preorder,c_struct_contiguous,10000,0.000325554,0.0
+55,preorder,c_struct_fragmented,10000,0.000335297,0.0
+56,msprime,py_pythran,100000,0.03365420203434769,2.3322594997706654e-05
+57,msprime,py_numba,100000,0.054033959006192164,2.67945655799003e-06
+58,msprime,py_tskit,100000,0.007538950289250351,3.5949492946489047e-07
+59,msprime,c_lib,100000,0.00751888,0.0
+60,msprime,c_recursive,100000,0.00692209,0.0
+61,msprime,c_struct_contiguous,100000,0.00633943,0.0
+62,msprime,c_struct_fragmented,100000,0.00424398,0.0
+63,preorder,py_pythran,100000,0.02893548096390441,4.3292710987952544e-07
+64,preorder,py_numba,100000,0.05132964390935376,5.868534453378092e-07
+65,preorder,py_tskit,100000,0.0034820095631293954,1.0937169630520294e-07
+66,preorder,c_lib,100000,0.00344897,0.0
+67,preorder,c_recursive,100000,0.00363671,0.0
+68,preorder,c_struct_contiguous,100000,0.00387891,0.0
+69,preorder,c_struct_fragmented,100000,0.00391498,0.0
+70,msprime,py_pythran,1000000,0.49991502292687073,5.48326295686209e-05
+71,msprime,py_numba,1000000,0.7700307431048714,0.00010178847386899849
+72,msprime,py_tskit,1000000,0.27924976880894975,1.3112941214562399e-05
+73,msprime,c_lib,1000000,0.279332,0.0
+74,msprime,c_recursive,1000000,0.214285,0.0
+75,msprime,c_struct_contiguous,1000000,0.198957,0.0
+76,msprime,c_struct_fragmented,1000000,0.0459972,0.0
+77,preorder,py_pythran,1000000,0.28591151580912993,3.0705415181927243e-06
+78,preorder,py_numba,1000000,0.5129637230769731,1.7103412821326635e-05
+79,preorder,py_tskit,1000000,0.04187946917954832,6.223032514941855e-07
+80,preorder,c_lib,1000000,0.0405432,0.0
+81,preorder,c_recursive,1000000,0.0396973,0.0
+82,preorder,c_struct_contiguous,1000000,0.0419802,0.0
+83,preorder,c_struct_fragmented,1000000,0.0405991,0.0
+84,msprime,py_pythran,10000000,7.074751639994792,0.0005614327797677364
+85,msprime,py_numba,10000000,10.0817625684198,0.002755264054662895
+86,msprime,py_tskit,10000000,4.04270198950544,0.0014427176057664867
+87,msprime,c_lib,10000000,4.09035,0.0
+88,msprime,c_recursive,10000000,3.66669,0.0
+89,msprime,c_struct_contiguous,10000000,2.93677,0.0
+90,msprime,c_struct_fragmented,10000000,1.15941,0.0
+91,preorder,py_pythran,10000000,2.962819242104888,0.0010798510649896842
+92,preorder,py_numba,10000000,5.11672681128839,0.0012743940016086866
+93,preorder,py_tskit,10000000,0.4149351096013561,6.0432335819378294e-05
+94,preorder,c_lib,10000000,0.395052,0.0
+95,preorder,c_recursive,10000000,0.398717,0.0
+96,preorder,c_struct_contiguous,10000000,0.425057,0.0
+97,preorder,c_struct_fragmented,10000000,0.41109,0.0

plot.py

@@ -18,16 +18,18 @@ def save(name):
 def tree_performance():
 
     df = pd.read_csv("data/tree-performance.csv")
+    df = df[df.sample_size >= 1000]
     fig, ax = plt.subplots(1, 1)
+    implementations = sorted(set(df.implementation))
     dfo = df[df.order == "msprime"]
     line_map = {}
-    for implementation in set(df.implementation):
+    for implementation in implementations:
         dfi = dfo[dfo.implementation == implementation]
         (line,) = ax.loglog(dfi.sample_size, dfi.time_mean, "-o", label=implementation)
         line_map[implementation] = line
 
     dfo = df[df.order == "preorder"]
-    for implementation in set(df.implementation):
+    for implementation in implementations:
         dfi = dfo[dfo.implementation == implementation]
         (line,) = ax.loglog(
             dfi.sample_size,
@@ -54,13 +56,14 @@ def tree_performance_relative():
     fig, axes = plt.subplots(1, 2, figsize=(8, 4), sharey=True)
 
     dfo = df[df.order == "msprime"]
+    implementations = sorted(set(df.implementation))
     print(dfo)
     norm = np.array(dfo[dfo.implementation == "c_lib"].time_mean)
     print(norm)
     line_map = {}
     ax = axes[0]
     ax.set_title("Node order = msprime")
-    for implementation in set(df.implementation):
+    for implementation in implementations:
         if "py" not in implementation:
             dfi = dfo[dfo.implementation == implementation]
             (line,) = ax.plot(
@@ -72,7 +75,7 @@ def tree_performance_relative():
     ax.set_title("Node order = preorder")
     dfo = df[df.order == "preorder"]
     norm = np.array(dfo[dfo.implementation == "c_lib"].time_mean)
-    for implementation in set(df.implementation):
+    for implementation in implementations:
         if "py" not in implementation:
             dfi = dfo[dfo.implementation == implementation]
             ax.plot(

tree_performance/Makefile

@@ -1,4 +1,5 @@
-TARGETS=c_implementation cpp_implementation
+PYTHRAN_SO=pythran_implementation.cpython-38-x86_64-linux-gnu.so
+TARGETS=c_implementation cpp_implementation ${PYTHRAN_SO}
 TSK_OBJECTS=kastore.o tskit_tables.o tskit_core.o tskit_trees.o \
 	tskit_stats.o tskit_genotypes.o tskit_convert.o
 CFLAGS += -g -Wall -Ikastore/c -I tskit/c -O3 -march=native
@@ -12,6 +13,9 @@ c_implementation: c_implementation.c libtskit.a
 cpp_implementation: cpp_implementation.cc libtskit.a
 	${CXX} ${CXXFLAGS} $^ -o $@ -lm
 
+${PYTHRAN_SO}: pythran_implementation.py
+	pythran pythran_implementation.py
+
 libtskit.a: ${TSK_OBJECTS}
 	${AR} rcs $@ ${TSK_OBJECTS}