diff --git a/.gitignore b/.gitignore index a89285e..3695513 100644 --- a/.gitignore +++ b/.gitignore @@ -1 +1,2 @@ *.img +*.png diff --git a/analysis/ampere_mini-em_avg.json b/analysis/ampere_mini-em_avg.json new file mode 100644 index 0000000..ccafafd --- /dev/null +++ b/analysis/ampere_mini-em_avg.json @@ -0,0 +1,3377 @@ +[ + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 31.67, + "page faults": 2757, + "cycles": 34506129, + "instructions": 75341056, + "branch mispredictions": 315890, + "branches": 16795763, + "ITLB accesses": 22839688, + "ITLB misses": 5094, + "DTLB misses": 12448, + "DTLB accesses": 30552656, + "L1I cache accesses": 26241679, + "L1I cache misses": 230488, + "L1D cache misses": 376311, + "L1D cache accesses": 27346779, + "LL cache misses": 470446, + "LL cache accesses": 497357, + "L2D TLB accesses": 133794, + "L2D TLB misses": 17138, + "L2D cache misses": 260141, + "L2D cache accesses": 1449409, + "\u0394 watt": 7.638666666666673, + "branch miss rate": 0.018807719542124998, + "ITLB miss rate": 0.00022303281901223869, + "DTLB miss rate": 0.0004074277535805725, + "L1I cache miss rate": 0.008783279454031886, + "L1D cache miss rate": 0.01376070651684427, + "L2D cache miss rate": 0.17948074008095713, + "LL cache miss rate": 0.9458919850328839 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 31.99, + "page faults": 2748, + "cycles": 30819840, + "instructions": 74713950, + "branch mispredictions": 314555, + "branches": 17082499, + "ITLB accesses": 22662021, + "ITLB misses": 5013, + "DTLB misses": 12452, + "DTLB accesses": 30478553, + "L1I cache accesses": 26662876, + "L1I cache misses": 236659, + "L1D cache misses": 385640, + "L1D cache accesses": 27887366, + "LL cache misses": 475900, + "LL cache accesses": 504259, + "L2D TLB accesses": 134409, + "L2D TLB misses": 17441, + "L2D cache misses": 263745, + "L2D cache accesses": 1481533, + "\u0394 watt": 3.9360000000000035, + "branch miss rate": 0.018413874925442702, + "ITLB miss rate": 0.00022120710240273805, + "DTLB miss rate": 0.00040854957910895574, + "L1I cache miss rate": 0.008875974219735336, + "L1D cache miss rate": 0.01382848419603343, + "L2D cache miss rate": 0.1780216842959286, + "LL cache miss rate": 0.9437610434320458 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell36", + "maxwell size": 36, + "matrix columns": 147852, + "task clock (msec)": 35.04, + "page faults": 2789, + "cycles": 36247748, + "instructions": 75696892, + "branch mispredictions": 317778, + "branches": 16939023, + "ITLB accesses": 22586635, + "ITLB misses": 5111, + "DTLB misses": 12387, + "DTLB accesses": 30378289, + "L1I cache accesses": 26656102, + "L1I cache misses": 231037, + "L1D cache misses": 379530, + "L1D cache accesses": 28042717, + "LL cache misses": 467202, + "LL cache accesses": 500467, + "L2D TLB accesses": 137989, + "L2D TLB misses": 18528, + "L2D cache misses": 262804, + "L2D cache accesses": 1482821, + "\u0394 watt": 15.126666666666672, + "branch miss rate": 0.01876011385072209, + "ITLB miss rate": 0.00022628426058153418, + "DTLB miss rate": 0.0004077583171323441, + "L1I cache miss rate": 0.008667321276006522, + "L1D cache miss rate": 0.013533995297245983, + "L2D cache miss rate": 0.17723245084875383, + "LL cache miss rate": 0.9335320810363121 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.15, + "page faults": 2747, + "cycles": 45356846, + "instructions": 77730625, + "branches": 16888960, + "branch mispredictions": 290099, + "ITLB accesses": 7996, + "ITLB misses": 14530, + "DTLB accesses": 17640131, + "DTLB misses": 14505, + "L1D cache accesses": 17593501, + "L1D cache misses": 574293, + "LL cache accesses": 62141, + "LL cache misses": 10948, + "branch miss rate": 0.017176842150138316, + "ITLB miss rate": 1.8171585792896447, + "DTLB miss rate": 0.0008222728051169234, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03264233764502017, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17617997779243977 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 31.09, + "page faults": 3226, + "cycles": 50299542, + "instructions": 77878301, + "branches": 16919929, + "branch mispredictions": 292094, + "ITLB accesses": 7694, + "ITLB misses": 13886, + "DTLB accesses": 17764513, + "DTLB misses": 12428, + "L1D cache accesses": 17864775, + "L1D cache misses": 574110, + "LL cache accesses": 65570, + "LL cache misses": 25239, + "branch miss rate": 0.01726331121129409, + "ITLB miss rate": 1.8047829477514947, + "DTLB miss rate": 0.0006995969999290158, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03213642489200116, + "L2D cache miss rate": null, + "LL cache miss rate": 0.38491688272075647 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 34.72, + "page faults": 2791, + "cycles": 37907762, + "instructions": 75121705, + "branch mispredictions": 313698, + "branches": 16778291, + "ITLB accesses": 22495586, + "ITLB misses": 4976, + "DTLB misses": 12077, + "DTLB accesses": 30380854, + "L1I cache accesses": 26482288, + "L1I cache misses": 235168, + "L1D cache misses": 385004, + "L1D cache accesses": 27728465, + "LL cache misses": 468613, + "LL cache accesses": 492063, + "L2D TLB accesses": 134110, + "L2D TLB misses": 17486, + "L2D cache misses": 255725, + "L2D cache accesses": 1452539, + "\u0394 watt": 3.1706666666666656, + "branch miss rate": 0.0186966598683978, + "ITLB miss rate": 0.0002211989498739886, + "DTLB miss rate": 0.00039752009604469976, + "L1I cache miss rate": 0.008880199475211507, + "L1D cache miss rate": 0.01388479311783036, + "L2D cache miss rate": 0.17605379270367266, + "LL cache miss rate": 0.9523435007305975 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 31.98, + "page faults": 2783, + "cycles": 30882808, + "instructions": 74791258, + "branch mispredictions": 311433, + "branches": 16883258, + "ITLB accesses": 22674676, + "ITLB misses": 5239, + "DTLB misses": 12341, + "DTLB accesses": 30377113, + "L1I cache accesses": 26540749, + "L1I cache misses": 238727, + "L1D cache misses": 377012, + "L1D cache accesses": 27809507, + "LL cache misses": 468899, + "LL cache accesses": 496246, + "L2D TLB accesses": 134879, + "L2D TLB misses": 17575, + "L2D cache misses": 257854, + "L2D cache accesses": 1458582, + "\u0394 watt": 5.262666666666668, + "branch miss rate": 0.018446261971474937, + "ITLB miss rate": 0.00023105071049306283, + "DTLB miss rate": 0.00040625980487349144, + "L1I cache miss rate": 0.008994734850926776, + "L1D cache miss rate": 0.013556946550688584, + "L2D cache miss rate": 0.1767840272264432, + "LL cache miss rate": 0.9448922510206631 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 36.03, + "page faults": 2754, + "cycles": 31415325, + "instructions": 73820867, + "branch mispredictions": 312352, + "branches": 16726998, + "ITLB accesses": 22571063, + "ITLB misses": 5115, + "DTLB misses": 14900, + "DTLB accesses": 30351769, + "L1I cache accesses": 26977665, + "L1I cache misses": 229090, + "L1D cache misses": 367879, + "L1D cache accesses": 28461927, + "LL cache misses": 471301, + "LL cache accesses": 499109, + "L2D TLB accesses": 139843, + "L2D TLB misses": 17816, + "L2D cache misses": 262194, + "L2D cache accesses": 1474249, + "\u0394 watt": 3.3026666666666706, + "branch miss rate": 0.018673524083640113, + "ITLB miss rate": 0.00022661759439508898, + "DTLB miss rate": 0.0004909104309537938, + "L1I cache miss rate": 0.008491839453117978, + "L1D cache miss rate": 0.012925301930540402, + "L2D cache miss rate": 0.17784919643832214, + "LL cache miss rate": 0.9442847153627765 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 36.85, + "page faults": 2771, + "cycles": 31621903, + "instructions": 74382715, + "branch mispredictions": 314723, + "branches": 16785485, + "ITLB accesses": 22955256, + "ITLB misses": 5739, + "DTLB misses": 12741, + "DTLB accesses": 30795163, + "L1I cache accesses": 26537679, + "L1I cache misses": 236820, + "L1D cache misses": 383053, + "L1D cache accesses": 27799952, + "LL cache misses": 467805, + "LL cache accesses": 492372, + "L2D TLB accesses": 134647, + "L2D TLB misses": 18367, + "L2D cache misses": 255210, + "L2D cache accesses": 1477471, + "\u0394 watt": 2.5333333333333314, + "branch miss rate": 0.0187497114322285, + "ITLB miss rate": 0.00025000810271948176, + "DTLB miss rate": 0.0004137338061824839, + "L1I cache miss rate": 0.008923915313015882, + "L1D cache miss rate": 0.013778908683007797, + "L2D cache miss rate": 0.17273435485366548, + "LL cache miss rate": 0.9501047988106553 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 30.97, + "page faults": 3251, + "cycles": 49455687, + "instructions": 78584469, + "branches": 17039294, + "branch mispredictions": 294966, + "ITLB accesses": 8375, + "ITLB misses": 14366, + "DTLB accesses": 17909445, + "DTLB misses": 14174, + "L1D cache accesses": 17933423, + "L1D cache misses": 585835, + "LL cache accesses": 73948, + "LL cache misses": 24797, + "branch miss rate": 0.017310928492694593, + "ITLB miss rate": 1.7153432835820897, + "DTLB miss rate": 0.0007914259766285332, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0326672158460769, + "L2D cache miss rate": null, + "LL cache miss rate": 0.33533023205495754 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 30.87, + "page faults": 2776, + "cycles": 31027198, + "instructions": 76269864, + "branch mispredictions": 316045, + "branches": 16843661, + "ITLB accesses": 22397832, + "ITLB misses": 5242, + "DTLB misses": 12258, + "DTLB accesses": 29979792, + "L1I cache accesses": 26500659, + "L1I cache misses": 242032, + "L1D cache misses": 394893, + "L1D cache accesses": 27722935, + "LL cache misses": 475227, + "LL cache accesses": 503677, + "L2D TLB accesses": 135439, + "L2D TLB misses": 18040, + "L2D cache misses": 263775, + "L2D cache accesses": 1471248, + "\u0394 watt": 3.1733333333333285, + "branch miss rate": 0.018763438660989437, + "ITLB miss rate": 0.0002340405089206848, + "DTLB miss rate": 0.0004088754184818894, + "L1I cache miss rate": 0.009133055898723123, + "L1D cache miss rate": 0.014244271034073412, + "L2D cache miss rate": 0.17928656487553424, + "LL cache miss rate": 0.9435153878378405 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell50", + "maxwell size": 50, + "matrix columns": 390150, + "task clock (msec)": 34.95, + "page faults": 2763, + "cycles": 39116016, + "instructions": 74243596, + "branch mispredictions": 312342, + "branches": 16630548, + "ITLB accesses": 22877751, + "ITLB misses": 5702, + "DTLB misses": 15680, + "DTLB accesses": 30809285, + "L1I cache accesses": 26242223, + "L1I cache misses": 238048, + "L1D cache misses": 386115, + "L1D cache accesses": 27462684, + "LL cache misses": 476534, + "LL cache accesses": 497710, + "L2D TLB accesses": 138762, + "L2D TLB misses": 18334, + "L2D cache misses": 258881, + "L2D cache accesses": 1470082, + "\u0394 watt": 19.074666666666662, + "branch miss rate": 0.01878122116000026, + "ITLB miss rate": 0.0002492377856547175, + "DTLB miss rate": 0.0005089374842681354, + "L1I cache miss rate": 0.00907118272716454, + "L1D cache miss rate": 0.014059623596877858, + "L2D cache miss rate": 0.17609970056092109, + "LL cache miss rate": 0.9574531353599486 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 27.37, + "page faults": 2759, + "cycles": 45348714, + "instructions": 78309411, + "branches": 17004157, + "branch mispredictions": 290236, + "ITLB accesses": 7105, + "ITLB misses": 14499, + "DTLB accesses": 17662985, + "DTLB misses": 12679, + "L1D cache accesses": 17593928, + "L1D cache misses": 590492, + "LL cache accesses": 65802, + "LL cache misses": 11589, + "branch miss rate": 0.017068532124232916, + "ITLB miss rate": 2.0406755805770582, + "DTLB miss rate": 0.0007178288381040917, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03356226079815718, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17611926689158383 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 28.84, + "page faults": 3251, + "cycles": 46945863, + "instructions": 78778600, + "branches": 17089248, + "branch mispredictions": 295353, + "ITLB accesses": 9379, + "ITLB misses": 15619, + "DTLB accesses": 18062818, + "DTLB misses": 14970, + "L1D cache accesses": 17623645, + "L1D cache misses": 587213, + "LL cache accesses": 66288, + "LL cache misses": 21398, + "branch miss rate": 0.017282972311010992, + "ITLB miss rate": 1.6653161317837724, + "DTLB miss rate": 0.0008287743363189509, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03331961123819732, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3228035240164132 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 36.01, + "page faults": 2758, + "cycles": 39830909, + "instructions": 74176381, + "branch mispredictions": 317516, + "branches": 16989939, + "ITLB accesses": 22906292, + "ITLB misses": 5412, + "DTLB misses": 12307, + "DTLB accesses": 30851528, + "L1I cache accesses": 26290542, + "L1I cache misses": 229175, + "L1D cache misses": 379049, + "L1D cache accesses": 27496626, + "LL cache misses": 462574, + "LL cache accesses": 487815, + "L2D TLB accesses": 136529, + "L2D TLB misses": 17906, + "L2D cache misses": 252703, + "L2D cache accesses": 1456088, + "\u0394 watt": 8.513333333333332, + "branch miss rate": 0.018688472042189205, + "ITLB miss rate": 0.00023626696106030605, + "DTLB miss rate": 0.0003989105499085815, + "L1I cache miss rate": 0.008717013137271952, + "L1D cache miss rate": 0.01378529132992535, + "L2D cache miss rate": 0.17354926350605182, + "LL cache miss rate": 0.9482570236667589 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 37.47, + "page faults": 2822, + "cycles": 25704354, + "instructions": 76290104, + "branch mispredictions": 316665, + "branches": 16856425, + "ITLB accesses": 22772192, + "ITLB misses": 5013, + "DTLB misses": 12272, + "DTLB accesses": 30469531, + "L1I cache accesses": 26519195, + "L1I cache misses": 236208, + "L1D cache misses": 380374, + "L1D cache accesses": 27752114, + "LL cache misses": 474912, + "LL cache accesses": 494530, + "L2D TLB accesses": 138674, + "L2D TLB misses": 17217, + "L2D cache misses": 255454, + "L2D cache accesses": 1459057, + "\u0394 watt": 8.971999999999998, + "branch miss rate": 0.018786011861945815, + "ITLB miss rate": 0.0002201369108428385, + "DTLB miss rate": 0.00040276300938140466, + "L1I cache miss rate": 0.008907057699149616, + "L1D cache miss rate": 0.013706127035944, + "L2D cache miss rate": 0.1750815766621866, + "LL cache miss rate": 0.9603300103128223 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 32.77, + "page faults": 3269, + "cycles": 51202082, + "instructions": 79122810, + "branches": 17153815, + "branch mispredictions": 292733, + "ITLB accesses": 8504, + "ITLB misses": 15168, + "DTLB accesses": 17800785, + "DTLB misses": 14899, + "L1D cache accesses": 17590854, + "L1D cache misses": 583954, + "LL cache accesses": 73737, + "LL cache misses": 22820, + "branch miss rate": 0.01706518345918969, + "ITLB miss rate": 1.7836312323612418, + "DTLB miss rate": 0.0008369855599064872, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033196455385281466, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3094782809173142 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell32", + "maxwell size": 32, + "matrix columns": 104544, + "task clock (msec)": 38.19, + "page faults": 2764, + "cycles": 28160037, + "instructions": 75192743, + "branch mispredictions": 316843, + "branches": 16871544, + "ITLB accesses": 22706983, + "ITLB misses": 5437, + "DTLB misses": 13067, + "DTLB accesses": 30572002, + "L1I cache accesses": 26441985, + "L1I cache misses": 225564, + "L1D cache misses": 376498, + "L1D cache accesses": 27735954, + "LL cache misses": 467890, + "LL cache accesses": 492897, + "L2D TLB accesses": 136869, + "L2D TLB misses": 18307, + "L2D cache misses": 257966, + "L2D cache accesses": 1471080, + "\u0394 watt": 11.0, + "branch miss rate": 0.018779727569687755, + "ITLB miss rate": 0.00023944176115338616, + "DTLB miss rate": 0.00042741721657613396, + "L1I cache miss rate": 0.008530524467054951, + "L1D cache miss rate": 0.013574366326105099, + "L2D cache miss rate": 0.17535824020447563, + "LL cache miss rate": 0.9492652623164678 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 29.14, + "page faults": 2775, + "cycles": 47219016, + "instructions": 79448945, + "branches": 17214993, + "branch mispredictions": 302959, + "ITLB accesses": 7393, + "ITLB misses": 14201, + "DTLB accesses": 17643944, + "DTLB misses": 13924, + "L1D cache accesses": 17778098, + "L1D cache misses": 587272, + "LL cache accesses": 71448, + "LL cache misses": 11360, + "branch miss rate": 0.01759855493406242, + "ITLB miss rate": 1.9208710942783713, + "DTLB miss rate": 0.0007891659597196635, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03303345498489208, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15899675288321577 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 29.05, + "page faults": 3258, + "cycles": 46083810, + "instructions": 79091262, + "branches": 17142768, + "branch mispredictions": 298230, + "ITLB accesses": 8268, + "ITLB misses": 13948, + "DTLB accesses": 17892377, + "DTLB misses": 14417, + "L1D cache accesses": 17736175, + "L1D cache misses": 584957, + "LL cache accesses": 71471, + "LL cache misses": 11675, + "branch miss rate": 0.01739684046357041, + "ITLB miss rate": 1.686985970004838, + "DTLB miss rate": 0.0008057621410503478, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03298101197129595, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16335296833680793 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 32.22, + "page faults": 2773, + "cycles": 24403039, + "instructions": 75230183, + "branch mispredictions": 316164, + "branches": 16836329, + "ITLB accesses": 22779660, + "ITLB misses": 5071, + "DTLB misses": 12526, + "DTLB accesses": 30516105, + "L1I cache accesses": 26602529, + "L1I cache misses": 240527, + "L1D cache misses": 390271, + "L1D cache accesses": 27837535, + "LL cache misses": 474380, + "LL cache accesses": 497098, + "L2D TLB accesses": 135216, + "L2D TLB misses": 17098, + "L2D cache misses": 261466, + "L2D cache accesses": 1451618, + "\u0394 watt": 3.979999999999997, + "branch miss rate": 0.018778677941016713, + "ITLB miss rate": 0.0002226108730332235, + "DTLB miss rate": 0.00041047178203116024, + "L1I cache miss rate": 0.009041508797904139, + "L1D cache miss rate": 0.01401959620347132, + "L2D cache miss rate": 0.18012038979952025, + "LL cache miss rate": 0.9542987499446789 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 29.43, + "page faults": 3250, + "cycles": 47251998, + "instructions": 79570608, + "branches": 17217585, + "branch mispredictions": 297938, + "ITLB accesses": 7313, + "ITLB misses": 13770, + "DTLB accesses": 17661992, + "DTLB misses": 14273, + "L1D cache accesses": 17655610, + "L1D cache misses": 575913, + "LL cache accesses": 66224, + "LL cache misses": 11211, + "branch miss rate": 0.017304285124772143, + "ITLB miss rate": 1.882948174483796, + "DTLB miss rate": 0.000808119491844408, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03261926379207515, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16928907948779898 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell40", + "maxwell size": 40, + "matrix columns": 201720, + "task clock (msec)": 36.17, + "page faults": 2777, + "cycles": 33198359, + "instructions": 75365745, + "branch mispredictions": 314158, + "branches": 16970945, + "ITLB accesses": 22549647, + "ITLB misses": 4902, + "DTLB misses": 12136, + "DTLB accesses": 30309488, + "L1I cache accesses": 26586747, + "L1I cache misses": 232631, + "L1D cache misses": 382824, + "L1D cache accesses": 27869172, + "LL cache misses": 444027, + "LL cache accesses": 472773, + "L2D TLB accesses": 160036, + "L2D TLB misses": 21546, + "L2D cache misses": 251471, + "L2D cache accesses": 1491358, + "\u0394 watt": 17.956, + "branch miss rate": 0.01851152071967707, + "ITLB miss rate": 0.00021738699501593084, + "DTLB miss rate": 0.00040040267258886064, + "L1I cache miss rate": 0.008749885798364125, + "L1D cache miss rate": 0.013736468381622532, + "L2D cache miss rate": 0.1686188024605762, + "LL cache miss rate": 0.9391970353636946 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell32", + "maxwell size": 32, + "matrix columns": 104544, + "task clock (msec)": 36.76, + "page faults": 2775, + "cycles": 34850201, + "instructions": 75677644, + "branch mispredictions": 311509, + "branches": 16732191, + "ITLB accesses": 22461744, + "ITLB misses": 5685, + "DTLB misses": 12876, + "DTLB accesses": 30259413, + "L1I cache accesses": 26404687, + "L1I cache misses": 225017, + "L1D cache misses": 376676, + "L1D cache accesses": 27688364, + "LL cache misses": 470244, + "LL cache accesses": 497265, + "L2D TLB accesses": 138358, + "L2D TLB misses": 19781, + "L2D cache misses": 257258, + "L2D cache accesses": 1475249, + "\u0394 watt": 10.506666666666668, + "branch miss rate": 0.018617346646353727, + "ITLB miss rate": 0.0002530969990575977, + "DTLB miss rate": 0.0004255204818414686, + "L1I cache miss rate": 0.008521858259482493, + "L1D cache miss rate": 0.013604126267626358, + "L2D cache miss rate": 0.17438276521455023, + "LL cache miss rate": 0.945660764381165 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 34.98, + "page faults": 2770, + "cycles": 35843362, + "instructions": 74725085, + "branch mispredictions": 315028, + "branches": 16768662, + "ITLB accesses": 22665706, + "ITLB misses": 5219, + "DTLB misses": 12254, + "DTLB accesses": 30493094, + "L1I cache accesses": 26628016, + "L1I cache misses": 240445, + "L1D cache misses": 377989, + "L1D cache accesses": 27940782, + "LL cache misses": 470547, + "LL cache accesses": 499243, + "L2D TLB accesses": 137928, + "L2D TLB misses": 17380, + "L2D cache misses": 260705, + "L2D cache accesses": 1463776, + "\u0394 watt": 11.36266666666667, + "branch miss rate": 0.018786710591459234, + "ITLB miss rate": 0.000230259758950372, + "DTLB miss rate": 0.00040186148378383644, + "L1I cache miss rate": 0.009029775256256417, + "L1D cache miss rate": 0.013528218358383814, + "L2D cache miss rate": 0.17810443674441992, + "LL cache miss rate": 0.9425209767588129 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 35.41, + "page faults": 2765, + "cycles": 37140362, + "instructions": 73876599, + "branch mispredictions": 316750, + "branches": 16782719, + "ITLB accesses": 22797676, + "ITLB misses": 5406, + "DTLB misses": 12331, + "DTLB accesses": 30733042, + "L1I cache accesses": 26630439, + "L1I cache misses": 237304, + "L1D cache misses": 387464, + "L1D cache accesses": 27899970, + "LL cache misses": 465815, + "LL cache accesses": 485283, + "L2D TLB accesses": 135965, + "L2D TLB misses": 17774, + "L2D cache misses": 248162, + "L2D cache accesses": 1468216, + "\u0394 watt": 2.949333333333332, + "branch miss rate": 0.018873580615870408, + "ITLB miss rate": 0.0002371294337194721, + "DTLB miss rate": 0.0004012293999402988, + "L1I cache miss rate": 0.008911005935726407, + "L1D cache miss rate": 0.013887613499225985, + "L2D cache miss rate": 0.1690228140818517, + "LL cache miss rate": 0.9598832021727528 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell50", + "maxwell size": 50, + "matrix columns": 390150, + "task clock (msec)": 35.49, + "page faults": 2748, + "cycles": 34782600, + "instructions": 74053260, + "branch mispredictions": 313370, + "branches": 16823321, + "ITLB accesses": 22512244, + "ITLB misses": 5600, + "DTLB misses": 12405, + "DTLB accesses": 30301609, + "L1I cache accesses": 26779507, + "L1I cache misses": 227711, + "L1D cache misses": 376129, + "L1D cache accesses": 28040627, + "LL cache misses": 459818, + "LL cache accesses": 479002, + "L2D TLB accesses": 134426, + "L2D TLB misses": 16836, + "L2D cache misses": 242825, + "L2D cache accesses": 1441765, + "\u0394 watt": 14.554666666666673, + "branch miss rate": 0.018627118866720787, + "ITLB miss rate": 0.00024875352274966457, + "DTLB miss rate": 0.0004093842013471958, + "L1I cache miss rate": 0.008503181182536334, + "L1D cache miss rate": 0.013413715748938139, + "L2D cache miss rate": 0.16842203826559807, + "LL cache miss rate": 0.9599500628389861 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 26.61, + "page faults": 2740, + "cycles": 44146579, + "instructions": 77589125, + "branches": 16882608, + "branch mispredictions": 287427, + "ITLB accesses": 8158, + "ITLB misses": 13803, + "DTLB accesses": 17689973, + "DTLB misses": 13768, + "L1D cache accesses": 18106537, + "L1D cache misses": 582639, + "LL cache accesses": 68276, + "LL cache misses": 10203, + "branch miss rate": 0.01702503546845369, + "ITLB miss rate": 1.6919588134346653, + "DTLB miss rate": 0.0007782940087019918, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03217837844972785, + "L2D cache miss rate": null, + "LL cache miss rate": 0.14943757689378406 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell46", + "maxwell size": 46, + "matrix columns": 304842, + "task clock (msec)": 37.23, + "page faults": 2814, + "cycles": 36640121, + "instructions": 75770899, + "branch mispredictions": 310974, + "branches": 16838522, + "ITLB accesses": 22880055, + "ITLB misses": 4914, + "DTLB misses": 11877, + "DTLB accesses": 30814344, + "L1I cache accesses": 26417580, + "L1I cache misses": 232026, + "L1D cache misses": 380204, + "L1D cache accesses": 27687572, + "LL cache misses": 462660, + "LL cache accesses": 484240, + "L2D TLB accesses": 136834, + "L2D TLB misses": 17359, + "L2D cache misses": 250884, + "L2D cache accesses": 1453835, + "\u0394 watt": 13.985333333333333, + "branch miss rate": 0.018468010434645035, + "ITLB miss rate": 0.00021477221099337393, + "DTLB miss rate": 0.0003854373794230375, + "L1I cache miss rate": 0.008783014946864929, + "L1D cache miss rate": 0.013731937202727635, + "L2D cache miss rate": 0.17256703821272704, + "LL cache miss rate": 0.955435321328267 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell38", + "maxwell size": 38, + "matrix columns": 173394, + "task clock (msec)": 35.51, + "page faults": 2769, + "cycles": 34057119, + "instructions": 74858259, + "branch mispredictions": 312992, + "branches": 16734878, + "ITLB accesses": 22407965, + "ITLB misses": 4664, + "DTLB misses": 11467, + "DTLB accesses": 30143975, + "L1I cache accesses": 26380461, + "L1I cache misses": 224524, + "L1D cache misses": 380936, + "L1D cache accesses": 27654401, + "LL cache misses": 464375, + "LL cache accesses": 488020, + "L2D TLB accesses": 132180, + "L2D TLB misses": 18494, + "L2D cache misses": 253615, + "L2D cache accesses": 1460608, + "\u0394 watt": 11.418666666666667, + "branch miss rate": 0.018702974709466063, + "ITLB miss rate": 0.0002081402751209224, + "DTLB miss rate": 0.0003804076934113699, + "L1I cache miss rate": 0.008510996073950337, + "L1D cache miss rate": 0.013774878002239137, + "L2D cache miss rate": 0.1736365951713259, + "LL cache miss rate": 0.9515491168394737 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell50", + "maxwell size": 50, + "matrix columns": 390150, + "task clock (msec)": 32.64, + "page faults": 2756, + "cycles": 35412935, + "instructions": 74655974, + "branch mispredictions": 316213, + "branches": 16775798, + "ITLB accesses": 22688159, + "ITLB misses": 5553, + "DTLB misses": 12575, + "DTLB accesses": 30436074, + "L1I cache accesses": 26563956, + "L1I cache misses": 245142, + "L1D cache misses": 385092, + "L1D cache accesses": 27790301, + "LL cache misses": 469441, + "LL cache accesses": 501026, + "L2D TLB accesses": 138495, + "L2D TLB misses": 19115, + "L2D cache misses": 262301, + "L2D cache accesses": 1493319, + "\u0394 watt": 18.674666666666667, + "branch miss rate": 0.018849356674418706, + "ITLB miss rate": 0.0002447532212728234, + "DTLB miss rate": 0.00041316104041539656, + "L1I cache miss rate": 0.00922836944918897, + "L1D cache miss rate": 0.013857064736362517, + "L2D cache miss rate": 0.17564967699466758, + "LL cache miss rate": 0.9369593593945225 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 31.26, + "page faults": 3259, + "cycles": 50199983, + "instructions": 79261227, + "branches": 17172254, + "branch mispredictions": 298417, + "ITLB accesses": 9179, + "ITLB misses": 15360, + "DTLB accesses": 17844537, + "DTLB misses": 15261, + "L1D cache accesses": 17823138, + "L1D cache misses": 578648, + "LL cache accesses": 70731, + "LL cache misses": 24360, + "branch miss rate": 0.017377858491960343, + "ITLB miss rate": 1.6733849003159385, + "DTLB miss rate": 0.0008552197235490055, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03246611230861816, + "L2D cache miss rate": null, + "LL cache miss rate": 0.34440344403444034 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 34.07, + "page faults": 2759, + "cycles": 35815677, + "instructions": 75698838, + "branch mispredictions": 308569, + "branches": 16734313, + "ITLB accesses": 22865588, + "ITLB misses": 4963, + "DTLB misses": 12200, + "DTLB accesses": 30667114, + "L1I cache accesses": 26435684, + "L1I cache misses": 231434, + "L1D cache misses": 386049, + "L1D cache accesses": 27683683, + "LL cache misses": 472377, + "LL cache accesses": 499919, + "L2D TLB accesses": 134780, + "L2D TLB misses": 17580, + "L2D cache misses": 263006, + "L2D cache accesses": 1483782, + "\u0394 watt": 3.6693333333333342, + "branch miss rate": 0.018439298942239217, + "ITLB miss rate": 0.00021705105506143117, + "DTLB miss rate": 0.000397820283969336, + "L1I cache miss rate": 0.008754606084714888, + "L1D cache miss rate": 0.013945001465303588, + "L2D cache miss rate": 0.17725380143444253, + "LL cache miss rate": 0.9449070749461412 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell34", + "maxwell size": 34, + "matrix columns": 124950, + "task clock (msec)": 35.39, + "page faults": 2757, + "cycles": 35019558, + "instructions": 73884050, + "branch mispredictions": 311820, + "branches": 17013235, + "ITLB accesses": 22753628, + "ITLB misses": 5573, + "DTLB misses": 12737, + "DTLB accesses": 30471484, + "L1I cache accesses": 26688842, + "L1I cache misses": 234773, + "L1D cache misses": 379425, + "L1D cache accesses": 27966418, + "LL cache misses": 444727, + "LL cache accesses": 464326, + "L2D TLB accesses": 158917, + "L2D TLB misses": 21453, + "L2D cache misses": 244029, + "L2D cache accesses": 1473015, + "\u0394 watt": 12.477333333333334, + "branch miss rate": 0.01832808398872995, + "ITLB miss rate": 0.00024492797368402087, + "DTLB miss rate": 0.0004179973643554741, + "L1I cache miss rate": 0.00879667240714303, + "L1D cache miss rate": 0.013567164733073789, + "L2D cache miss rate": 0.165666337410006, + "LL cache miss rate": 0.9577904317225397 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell42", + "maxwell size": 42, + "matrix columns": 232974, + "task clock (msec)": 38.94, + "page faults": 2771, + "cycles": 38604757, + "instructions": 77657485, + "branch mispredictions": 311614, + "branches": 16694225, + "ITLB accesses": 22898357, + "ITLB misses": 5552, + "DTLB misses": 13048, + "DTLB accesses": 30770993, + "L1I cache accesses": 26383983, + "L1I cache misses": 223528, + "L1D cache misses": 375899, + "L1D cache accesses": 27687659, + "LL cache misses": 458067, + "LL cache accesses": 477325, + "L2D TLB accesses": 134565, + "L2D TLB misses": 17564, + "L2D cache misses": 245351, + "L2D cache accesses": 1438299, + "\u0394 watt": 12.877333333333336, + "branch miss rate": 0.01866597580899982, + "ITLB miss rate": 0.00024246281075974142, + "DTLB miss rate": 0.0004240357144145462, + "L1I cache miss rate": 0.008472109764473393, + "L1D cache miss rate": 0.013576409619895998, + "L2D cache miss rate": 0.17058414140592465, + "LL cache miss rate": 0.9596543235740848 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell34", + "maxwell size": 34, + "matrix columns": 124950, + "task clock (msec)": 31.69, + "page faults": 2762, + "cycles": 31887058, + "instructions": 74429157, + "branch mispredictions": 319093, + "branches": 16928644, + "ITLB accesses": 22662945, + "ITLB misses": 5028, + "DTLB misses": 12059, + "DTLB accesses": 30464264, + "L1I cache accesses": 26360446, + "L1I cache misses": 225534, + "L1D cache misses": 377754, + "L1D cache accesses": 27615755, + "LL cache misses": 454234, + "LL cache accesses": 483749, + "L2D TLB accesses": 155422, + "L2D TLB misses": 22005, + "L2D cache misses": 258239, + "L2D cache accesses": 1497679, + "\u0394 watt": 10.676000000000002, + "branch miss rate": 0.01884929472201081, + "ITLB miss rate": 0.0002218599568590931, + "DTLB miss rate": 0.000395840844866628, + "L1I cache miss rate": 0.00855577329761416, + "L1D cache miss rate": 0.013678930740803574, + "L2D cache miss rate": 0.17242613403806822, + "LL cache miss rate": 0.9389869539781994 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 35.66, + "page faults": 2783, + "cycles": 28797561, + "instructions": 72611886, + "branch mispredictions": 314831, + "branches": 16822795, + "ITLB accesses": 22365064, + "ITLB misses": 5315, + "DTLB misses": 12146, + "DTLB accesses": 29931897, + "L1I cache accesses": 26478185, + "L1I cache misses": 232023, + "L1D cache misses": 378229, + "L1D cache accesses": 27831494, + "LL cache misses": 448786, + "LL cache accesses": 468750, + "L2D TLB accesses": 130012, + "L2D TLB misses": 16264, + "L2D cache misses": 238552, + "L2D cache accesses": 1420866, + "\u0394 watt": 13.818666666666665, + "branch miss rate": 0.018714547731218268, + "ITLB miss rate": 0.00023764743083230168, + "DTLB miss rate": 0.0004057878456550883, + "L1I cache miss rate": 0.008762798507526101, + "L1D cache miss rate": 0.013589963945162268, + "L2D cache miss rate": 0.16789197573873962, + "LL cache miss rate": 0.9574101333333334 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 38.24, + "page faults": 2791, + "cycles": 38107904, + "instructions": 76348269, + "branch mispredictions": 316120, + "branches": 17064379, + "ITLB accesses": 22594643, + "ITLB misses": 5428, + "DTLB misses": 12324, + "DTLB accesses": 30768538, + "L1I cache accesses": 26740932, + "L1I cache misses": 233141, + "L1D cache misses": 388857, + "L1D cache accesses": 28098450, + "LL cache misses": 466769, + "LL cache accesses": 489164, + "L2D TLB accesses": 133866, + "L2D TLB misses": 16446, + "L2D cache misses": 250317, + "L2D cache accesses": 1469531, + "\u0394 watt": 1.4426666666666712, + "branch miss rate": 0.01852513941468365, + "ITLB miss rate": 0.00024023393509691655, + "DTLB miss rate": 0.00040053901813599335, + "L1I cache miss rate": 0.008718506894224928, + "L1D cache miss rate": 0.013839090768351991, + "L2D cache miss rate": 0.17033801940891347, + "LL cache miss rate": 0.9542178083423964 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 31.53, + "page faults": 2757, + "cycles": 38154360, + "instructions": 76003480, + "branch mispredictions": 314535, + "branches": 16819491, + "ITLB accesses": 22654359, + "ITLB misses": 5040, + "DTLB misses": 11917, + "DTLB accesses": 30403638, + "L1I cache accesses": 26635517, + "L1I cache misses": 238993, + "L1D cache misses": 384206, + "L1D cache accesses": 27836639, + "LL cache misses": 475418, + "LL cache accesses": 501247, + "L2D TLB accesses": 133593, + "L2D TLB misses": 17654, + "L2D cache misses": 262844, + "L2D cache accesses": 1466034, + "\u0394 watt": 6.961333333333332, + "branch miss rate": 0.01870062536375209, + "ITLB miss rate": 0.00022247374114624032, + "DTLB miss rate": 0.0003919596727207448, + "L1I cache miss rate": 0.008972718644807984, + "L1D cache miss rate": 0.013802169148365935, + "L2D cache miss rate": 0.17928915700454423, + "LL cache miss rate": 0.9484705145367454 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 28.84, + "page faults": 3238, + "cycles": 46524233, + "instructions": 78604006, + "branches": 17054156, + "branch mispredictions": 298876, + "ITLB accesses": 6035, + "ITLB misses": 13621, + "DTLB accesses": 17783780, + "DTLB misses": 12598, + "L1D cache accesses": 17494763, + "L1D cache misses": 576538, + "LL cache accesses": 67358, + "LL cache misses": 11047, + "branch miss rate": 0.01752511235384501, + "ITLB miss rate": 2.257000828500414, + "DTLB miss rate": 0.0007083983270148416, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032954890557820075, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16400427566139136 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 27.25, + "page faults": 2736, + "cycles": 45435644, + "instructions": 78271647, + "branches": 16994482, + "branch mispredictions": 291926, + "ITLB accesses": 7616, + "ITLB misses": 14515, + "DTLB accesses": 17745451, + "DTLB misses": 14184, + "L1D cache accesses": 17720302, + "L1D cache misses": 572377, + "LL cache accesses": 63704, + "LL cache misses": 11538, + "branch miss rate": 0.017177693324221355, + "ITLB miss rate": 1.9058560924369747, + "DTLB miss rate": 0.0007993034383854205, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03230063460543731, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1811189250282557 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 37.43, + "page faults": 2760, + "cycles": 38247751, + "instructions": 74492341, + "branch mispredictions": 315870, + "branches": 16808953, + "ITLB accesses": 22562493, + "ITLB misses": 4714, + "DTLB misses": 12202, + "DTLB accesses": 30307579, + "L1I cache accesses": 26528925, + "L1I cache misses": 229763, + "L1D cache misses": 371071, + "L1D cache accesses": 27793339, + "LL cache misses": 473991, + "LL cache accesses": 495456, + "L2D TLB accesses": 134781, + "L2D TLB misses": 17650, + "L2D cache misses": 257771, + "L2D cache accesses": 1472049, + "\u0394 watt": 18.604000000000006, + "branch miss rate": 0.018791771266181777, + "ITLB miss rate": 0.00020893081274307764, + "DTLB miss rate": 0.0004026055660862915, + "L1I cache miss rate": 0.008660848488960635, + "L1D cache miss rate": 0.01335107667344323, + "L2D cache miss rate": 0.1751103393976695, + "LL cache miss rate": 0.9566762739779112 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 35.44, + "page faults": 2764, + "cycles": 36127271, + "instructions": 76768925, + "branch mispredictions": 316239, + "branches": 16877036, + "ITLB accesses": 22436874, + "ITLB misses": 5008, + "DTLB misses": 11966, + "DTLB accesses": 30170027, + "L1I cache accesses": 26578511, + "L1I cache misses": 230855, + "L1D cache misses": 365266, + "L1D cache accesses": 27855340, + "LL cache misses": 459831, + "LL cache accesses": 478879, + "L2D TLB accesses": 134853, + "L2D TLB misses": 17671, + "L2D cache misses": 246499, + "L2D cache accesses": 1431433, + "\u0394 watt": 2.924000000000003, + "branch miss rate": 0.01873782813522469, + "ITLB miss rate": 0.0002232039989171397, + "DTLB miss rate": 0.00039661880315851225, + "L1I cache miss rate": 0.008685776264893093, + "L1D cache miss rate": 0.013112961464480419, + "L2D cache miss rate": 0.1722043574515887, + "LL cache miss rate": 0.9602237726022649 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 34.7, + "page faults": 2768, + "cycles": 26451940, + "instructions": 72629827, + "branch mispredictions": 318871, + "branches": 16901235, + "ITLB accesses": 22500323, + "ITLB misses": 5270, + "DTLB misses": 12429, + "DTLB accesses": 30168626, + "L1I cache accesses": 26444828, + "L1I cache misses": 226916, + "L1D cache misses": 373966, + "L1D cache accesses": 27747568, + "LL cache misses": 463884, + "LL cache accesses": 483483, + "L2D TLB accesses": 134714, + "L2D TLB misses": 17910, + "L2D cache misses": 244944, + "L2D cache accesses": 1458075, + "\u0394 watt": 5.628, + "branch miss rate": 0.018866727786460574, + "ITLB miss rate": 0.00023421885988036706, + "DTLB miss rate": 0.00041198429123023367, + "L1I cache miss rate": 0.008580732686179695, + "L1D cache miss rate": 0.013477433409659543, + "L2D cache miss rate": 0.16799135846921454, + "LL cache miss rate": 0.9594628973510961 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell48", + "maxwell size": 48, + "matrix columns": 345744, + "task clock (msec)": 35.09, + "page faults": 2777, + "cycles": 35498307, + "instructions": 74874275, + "branch mispredictions": 309679, + "branches": 16861431, + "ITLB accesses": 22803677, + "ITLB misses": 5037, + "DTLB misses": 12129, + "DTLB accesses": 30733658, + "L1I cache accesses": 26477891, + "L1I cache misses": 229830, + "L1D cache misses": 377265, + "L1D cache accesses": 27773004, + "LL cache misses": 470738, + "LL cache accesses": 496153, + "L2D TLB accesses": 133031, + "L2D TLB misses": 17983, + "L2D cache misses": 258141, + "L2D cache accesses": 1471401, + "\u0394 watt": 14.819999999999997, + "branch miss rate": 0.01836611613806681, + "ITLB miss rate": 0.0002208854300120108, + "DTLB miss rate": 0.0003946487593504164, + "L1I cache miss rate": 0.00868007198911726, + "L1D cache miss rate": 0.013583874470330973, + "L2D cache miss rate": 0.17543891841856843, + "LL cache miss rate": 0.9487758816332865 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 29.36, + "page faults": 3257, + "cycles": 47397126, + "instructions": 79359110, + "branches": 17182349, + "branch mispredictions": 299371, + "ITLB accesses": 7154, + "ITLB misses": 14997, + "DTLB accesses": 17594878, + "DTLB misses": 14791, + "L1D cache accesses": 17816160, + "L1D cache misses": 574506, + "LL cache accesses": 67016, + "LL cache misses": 11106, + "branch miss rate": 0.01742317072013844, + "ITLB miss rate": 2.096309756779424, + "DTLB miss rate": 0.0008406423733088687, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0322463426462268, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1657216187179181 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 27.95, + "page faults": 2757, + "cycles": 46254053, + "instructions": 78828949, + "branches": 17081818, + "branch mispredictions": 295191, + "ITLB accesses": 7066, + "ITLB misses": 14373, + "DTLB accesses": 17813443, + "DTLB misses": 14801, + "L1D cache accesses": 17558896, + "L1D cache misses": 577762, + "LL cache accesses": 58576, + "LL cache misses": 11107, + "branch miss rate": 0.017281006038116085, + "ITLB miss rate": 2.0341069912255874, + "DTLB miss rate": 0.0008308893457598287, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03290423270346837, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1896169079486479 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 33.23, + "page faults": 2759, + "cycles": 30945071, + "instructions": 69398421, + "branch mispredictions": 314531, + "branches": 16779111, + "ITLB accesses": 22691553, + "ITLB misses": 4951, + "DTLB misses": 12203, + "DTLB accesses": 30498706, + "L1I cache accesses": 26484775, + "L1I cache misses": 230256, + "L1D cache misses": 382347, + "L1D cache accesses": 27731498, + "LL cache misses": 459487, + "LL cache accesses": 479612, + "L2D TLB accesses": 136168, + "L2D TLB misses": 16861, + "L2D cache misses": 247064, + "L2D cache accesses": 1432491, + "\u0394 watt": 11.930666666666667, + "branch miss rate": 0.018745391218879238, + "ITLB miss rate": 0.00021818691739609007, + "DTLB miss rate": 0.0004001153360408143, + "L1I cache miss rate": 0.008693900552298444, + "L1D cache miss rate": 0.013787462905898556, + "L2D cache miss rate": 0.17247158969934193, + "LL cache miss rate": 0.9580389981902038 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 34.05, + "page faults": 2772, + "cycles": 35361566, + "instructions": 75377352, + "branch mispredictions": 313864, + "branches": 16971197, + "ITLB accesses": 22441338, + "ITLB misses": 5089, + "DTLB misses": 12043, + "DTLB accesses": 30113443, + "L1I cache accesses": 26484166, + "L1I cache misses": 234493, + "L1D cache misses": 375748, + "L1D cache accesses": 27804002, + "LL cache misses": 474052, + "LL cache accesses": 502246, + "L2D TLB accesses": 140197, + "L2D TLB misses": 18415, + "L2D cache misses": 265062, + "L2D cache accesses": 1466151, + "\u0394 watt": 4.187999999999999, + "branch miss rate": 0.018493922379193405, + "ITLB miss rate": 0.00022676900994049465, + "DTLB miss rate": 0.0003999210585119742, + "L1I cache miss rate": 0.008854082850862663, + "L1D cache miss rate": 0.013514169650829402, + "L2D cache miss rate": 0.18078765420478518, + "LL cache miss rate": 0.9438641621834718 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell36", + "maxwell size": 36, + "matrix columns": 147852, + "task clock (msec)": 33.29, + "page faults": 2784, + "cycles": 27771327, + "instructions": 74669795, + "branch mispredictions": 317060, + "branches": 16860901, + "ITLB accesses": 22406422, + "ITLB misses": 5222, + "DTLB misses": 12241, + "DTLB accesses": 30325313, + "L1I cache accesses": 26887951, + "L1I cache misses": 242387, + "L1D cache misses": 392627, + "L1D cache accesses": 28111869, + "LL cache misses": 471324, + "LL cache accesses": 510684, + "L2D TLB accesses": 136615, + "L2D TLB misses": 18298, + "L2D cache misses": 272960, + "L2D cache accesses": 1501134, + "\u0394 watt": 11.283999999999995, + "branch miss rate": 0.01880445179056564, + "ITLB miss rate": 0.00023305818305126985, + "DTLB miss rate": 0.00040365617990488674, + "L1I cache miss rate": 0.009014706996453542, + "L1D cache miss rate": 0.013966591833506339, + "L2D cache miss rate": 0.1818358654190765, + "LL cache miss rate": 0.9229268980426252 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell46", + "maxwell size": 46, + "matrix columns": 304842, + "task clock (msec)": 31.96, + "page faults": 2806, + "cycles": 27840159, + "instructions": 74835856, + "branch mispredictions": 312737, + "branches": 16758090, + "ITLB accesses": 22859003, + "ITLB misses": 5584, + "DTLB misses": 13002, + "DTLB accesses": 30585286, + "L1I cache accesses": 26237522, + "L1I cache misses": 226286, + "L1D cache misses": 373324, + "L1D cache accesses": 27436569, + "LL cache misses": 465023, + "LL cache accesses": 492346, + "L2D TLB accesses": 135926, + "L2D TLB misses": 17901, + "L2D cache misses": 256396, + "L2D cache accesses": 1470740, + "\u0394 watt": 14.650666666666666, + "branch miss rate": 0.01866185227552782, + "ITLB miss rate": 0.0002442801201784697, + "DTLB miss rate": 0.0004251063730448687, + "L1I cache miss rate": 0.008624518733133411, + "L1D cache miss rate": 0.013606803387114475, + "L2D cache miss rate": 0.17433128901097406, + "LL cache miss rate": 0.9445044744955783 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 31.55, + "page faults": 2771, + "cycles": 28283186, + "instructions": 74469098, + "branch mispredictions": 311442, + "branches": 16765071, + "ITLB accesses": 22534328, + "ITLB misses": 5621, + "DTLB misses": 12662, + "DTLB accesses": 30175541, + "L1I cache accesses": 26702763, + "L1I cache misses": 246882, + "L1D cache misses": 387738, + "L1D cache accesses": 27767626, + "LL cache misses": 459949, + "LL cache accesses": 482967, + "L2D TLB accesses": 131653, + "L2D TLB misses": 17297, + "L2D cache misses": 245227, + "L2D cache accesses": 1451665, + "\u0394 watt": 3.324000000000005, + "branch miss rate": 0.018576837521296508, + "ITLB miss rate": 0.00024944165186554487, + "DTLB miss rate": 0.0004196113667025887, + "L1I cache miss rate": 0.009245560094286872, + "L1D cache miss rate": 0.013963671219138432, + "L2D cache miss rate": 0.16892809291399874, + "LL cache miss rate": 0.9523404290562295 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 33.37, + "page faults": 2779, + "cycles": 36802889, + "instructions": 75056728, + "branch mispredictions": 313750, + "branches": 16722652, + "ITLB accesses": 22567618, + "ITLB misses": 5618, + "DTLB misses": 12389, + "DTLB accesses": 30324090, + "L1I cache accesses": 26480842, + "L1I cache misses": 225755, + "L1D cache misses": 369823, + "L1D cache accesses": 27799943, + "LL cache misses": 457189, + "LL cache accesses": 485992, + "L2D TLB accesses": 131135, + "L2D TLB misses": 17503, + "L2D cache misses": 250990, + "L2D cache accesses": 1464593, + "\u0394 watt": 6.472000000000001, + "branch miss rate": 0.018761976270271007, + "ITLB miss rate": 0.00024894076104974835, + "DTLB miss rate": 0.00040855306787441933, + "L1I cache miss rate": 0.008525219855169257, + "L1D cache miss rate": 0.013303012887472467, + "L2D cache miss rate": 0.17137184187006219, + "LL cache miss rate": 0.9407335923225074 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 34.81, + "page faults": 2765, + "cycles": 36773990, + "instructions": 74420116, + "branch mispredictions": 314951, + "branches": 17042284, + "ITLB accesses": 22487197, + "ITLB misses": 5607, + "DTLB misses": 12638, + "DTLB accesses": 30192260, + "L1I cache accesses": 27043845, + "L1I cache misses": 236758, + "L1D cache misses": 383812, + "L1D cache accesses": 28459501, + "LL cache misses": 457211, + "LL cache accesses": 476767, + "L2D TLB accesses": 134166, + "L2D TLB misses": 16622, + "L2D cache misses": 241461, + "L2D cache accesses": 1434082, + "\u0394 watt": 8.020000000000003, + "branch miss rate": 0.01848056281658022, + "ITLB miss rate": 0.00024934188107126026, + "DTLB miss rate": 0.0004185841006933565, + "L1I cache miss rate": 0.008754598319876482, + "L1D cache miss rate": 0.013486251919877302, + "L2D cache miss rate": 0.16837321715215728, + "LL cache miss rate": 0.9589820604194502 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell32", + "maxwell size": 32, + "matrix columns": 104544, + "task clock (msec)": 37.31, + "page faults": 2777, + "cycles": 34048763, + "instructions": 74343582, + "branch mispredictions": 317436, + "branches": 16874761, + "ITLB accesses": 22591179, + "ITLB misses": 4795, + "DTLB misses": 11739, + "DTLB accesses": 30359421, + "L1I cache accesses": 26705866, + "L1I cache misses": 229557, + "L1D cache misses": 377734, + "L1D cache accesses": 28114986, + "LL cache misses": 463225, + "LL cache accesses": 483313, + "L2D TLB accesses": 134440, + "L2D TLB misses": 17618, + "L2D cache misses": 247222, + "L2D cache accesses": 1452336, + "\u0394 watt": 17.98933333333333, + "branch miss rate": 0.018811288645806598, + "ITLB miss rate": 0.0002122509852186112, + "DTLB miss rate": 0.0003866674532429324, + "L1I cache miss rate": 0.008595751959513315, + "L1D cache miss rate": 0.013435325914798606, + "L2D cache miss rate": 0.17022369479238963, + "LL cache miss rate": 0.9584368721718638 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell48", + "maxwell size": 48, + "matrix columns": 345744, + "task clock (msec)": 34.94, + "page faults": 2771, + "cycles": 38735018, + "instructions": 75159781, + "branch mispredictions": 315149, + "branches": 16816929, + "ITLB accesses": 22633453, + "ITLB misses": 5768, + "DTLB misses": 12591, + "DTLB accesses": 30399400, + "L1I cache accesses": 26918987, + "L1I cache misses": 244722, + "L1D cache misses": 389557, + "L1D cache accesses": 28286479, + "LL cache misses": 462508, + "LL cache accesses": 481919, + "L2D TLB accesses": 136590, + "L2D TLB misses": 17961, + "L2D cache misses": 246707, + "L2D cache accesses": 1437179, + "\u0394 watt": 12.584000000000003, + "branch miss rate": 0.01873998516613824, + "ITLB miss rate": 0.00025484401341677735, + "DTLB miss rate": 0.00041418580629880856, + "L1I cache miss rate": 0.009091055320915307, + "L1D cache miss rate": 0.013771844845022953, + "L2D cache miss rate": 0.17166059342642775, + "LL cache miss rate": 0.9597214469651539 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell34", + "maxwell size": 34, + "matrix columns": 124950, + "task clock (msec)": 31.77, + "page faults": 2772, + "cycles": 21933963, + "instructions": 71670947, + "branch mispredictions": 313067, + "branches": 16949851, + "ITLB accesses": 22410986, + "ITLB misses": 5228, + "DTLB misses": 12020, + "DTLB accesses": 30113343, + "L1I cache accesses": 26523000, + "L1I cache misses": 232921, + "L1D cache misses": 380346, + "L1D cache accesses": 27768454, + "LL cache misses": 470177, + "LL cache accesses": 493751, + "L2D TLB accesses": 134281, + "L2D TLB misses": 16783, + "L2D cache misses": 257166, + "L2D cache accesses": 1461513, + "\u0394 watt": 17.726666666666663, + "branch miss rate": 0.018470191861863564, + "ITLB miss rate": 0.00023327844656187817, + "DTLB miss rate": 0.0003991586055390795, + "L1I cache miss rate": 0.008781849715341402, + "L1D cache miss rate": 0.013697053498188987, + "L2D cache miss rate": 0.17595874959716404, + "LL cache miss rate": 0.9522552865715715 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell44", + "maxwell size": 44, + "matrix columns": 267300, + "task clock (msec)": 33.08, + "page faults": 2779, + "cycles": 25039508, + "instructions": 73874444, + "branch mispredictions": 314391, + "branches": 16756549, + "ITLB accesses": 22832236, + "ITLB misses": 5366, + "DTLB misses": 12776, + "DTLB accesses": 30548671, + "L1I cache accesses": 26617695, + "L1I cache misses": 251490, + "L1D cache misses": 393877, + "L1D cache accesses": 27768409, + "LL cache misses": 473559, + "LL cache accesses": 506336, + "L2D TLB accesses": 138392, + "L2D TLB misses": 18953, + "L2D cache misses": 267893, + "L2D cache accesses": 1496868, + "\u0394 watt": 13.570666666666668, + "branch miss rate": 0.018762276170349874, + "ITLB miss rate": 0.00023501859388629304, + "DTLB miss rate": 0.0004182178661716577, + "L1I cache miss rate": 0.009448226076675685, + "L1D cache miss rate": 0.014184356042868714, + "L2D cache miss rate": 0.1789690206484473, + "LL cache miss rate": 0.9352663053782468 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 30.87, + "page faults": 2759, + "cycles": 37012840, + "instructions": 73960783, + "branch mispredictions": 315756, + "branches": 16812678, + "ITLB accesses": 22882240, + "ITLB misses": 5700, + "DTLB misses": 12892, + "DTLB accesses": 30639838, + "L1I cache accesses": 26461356, + "L1I cache misses": 231196, + "L1D cache misses": 378676, + "L1D cache accesses": 27671660, + "LL cache misses": 470732, + "LL cache accesses": 496250, + "L2D TLB accesses": 135873, + "L2D TLB misses": 19574, + "L2D cache misses": 261795, + "L2D cache accesses": 1476352, + "\u0394 watt": 9.581333333333337, + "branch miss rate": 0.01878082718291518, + "ITLB miss rate": 0.0002491014865677486, + "DTLB miss rate": 0.00042075940479841965, + "L1I cache miss rate": 0.008737118385013981, + "L1D cache miss rate": 0.013684614511742339, + "L2D cache miss rate": 0.1773255971475637, + "LL cache miss rate": 0.9485783375314861 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 38.13, + "page faults": 2794, + "cycles": 32415571, + "instructions": 75634143, + "branch mispredictions": 312624, + "branches": 16735629, + "ITLB accesses": 22746815, + "ITLB misses": 5116, + "DTLB misses": 12449, + "DTLB accesses": 30413011, + "L1I cache accesses": 26672713, + "L1I cache misses": 231027, + "L1D cache misses": 377361, + "L1D cache accesses": 27757160, + "LL cache misses": 469460, + "LL cache accesses": 493233, + "L2D TLB accesses": 136380, + "L2D TLB misses": 18404, + "L2D cache misses": 255871, + "L2D cache accesses": 1460513, + "\u0394 watt": 17.996, + "branch miss rate": 0.018680146410989393, + "ITLB miss rate": 0.00022491060836429187, + "DTLB miss rate": 0.00040933138780635693, + "L1I cache miss rate": 0.008661548602123826, + "L1D cache miss rate": 0.013595086817239227, + "L2D cache miss rate": 0.17519255220597146, + "LL cache miss rate": 0.9518016839911361 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 31.1, + "page faults": 3260, + "cycles": 49512609, + "instructions": 79157466, + "branches": 17163022, + "branch mispredictions": 296724, + "ITLB accesses": 7966, + "ITLB misses": 13125, + "DTLB accesses": 17723474, + "DTLB misses": 13123, + "L1D cache accesses": 17642308, + "L1D cache misses": 578691, + "LL cache accesses": 70574, + "LL cache misses": 30238, + "branch miss rate": 0.017288563750602896, + "ITLB miss rate": 1.647627416520211, + "DTLB miss rate": 0.0007404304596265946, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032801320552843766, + "L2D cache miss rate": null, + "LL cache miss rate": 0.42845807237792954 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 37.94, + "page faults": 2766, + "cycles": 38331378, + "instructions": 70059972, + "branch mispredictions": 313098, + "branches": 16918712, + "ITLB accesses": 22604179, + "ITLB misses": 5034, + "DTLB misses": 12111, + "DTLB accesses": 30294479, + "L1I cache accesses": 26278079, + "L1I cache misses": 221462, + "L1D cache misses": 380684, + "L1D cache accesses": 27551821, + "LL cache misses": 466481, + "LL cache accesses": 490057, + "L2D TLB accesses": 134286, + "L2D TLB misses": 18651, + "L2D cache misses": 255806, + "L2D cache accesses": 1444473, + "\u0394 watt": 9.651999999999994, + "branch miss rate": 0.01850601866146785, + "ITLB miss rate": 0.00022270218263622846, + "DTLB miss rate": 0.00039977581393626214, + "L1I cache miss rate": 0.008427632780919793, + "L1D cache miss rate": 0.01381701775719289, + "L2D cache miss rate": 0.1770929605468569, + "LL cache miss rate": 0.9518913106026442 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 34.09, + "page faults": 2782, + "cycles": 37064169, + "instructions": 74037049, + "branch mispredictions": 313553, + "branches": 16971332, + "ITLB accesses": 22830212, + "ITLB misses": 5471, + "DTLB misses": 12426, + "DTLB accesses": 30661633, + "L1I cache accesses": 26615111, + "L1I cache misses": 230524, + "L1D cache misses": 376346, + "L1D cache accesses": 27931850, + "LL cache misses": 461499, + "LL cache accesses": 488022, + "L2D TLB accesses": 135124, + "L2D TLB misses": 18143, + "L2D cache misses": 253051, + "L2D cache accesses": 1454207, + "\u0394 watt": 2.6693333333333378, + "branch miss rate": 0.01847545024751151, + "ITLB miss rate": 0.0002396385981873493, + "DTLB miss rate": 0.00040526217243549946, + "L1I cache miss rate": 0.008661395400530172, + "L1D cache miss rate": 0.013473722649949789, + "L2D cache miss rate": 0.1740130531623077, + "LL cache miss rate": 0.9456520402768728 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 35.49, + "page faults": 2781, + "cycles": 35904207, + "instructions": 74255441, + "branch mispredictions": 315604, + "branches": 16831247, + "ITLB accesses": 23016221, + "ITLB misses": 5250, + "DTLB misses": 12426, + "DTLB accesses": 30958111, + "L1I cache accesses": 26467387, + "L1I cache misses": 233063, + "L1D cache misses": 384211, + "L1D cache accesses": 27694892, + "LL cache misses": 464776, + "LL cache accesses": 484199, + "L2D TLB accesses": 134303, + "L2D TLB misses": 18242, + "L2D cache misses": 249032, + "L2D cache accesses": 1443567, + "\u0394 watt": 9.132000000000001, + "branch miss rate": 0.018751076494807544, + "ITLB miss rate": 0.00022809999956117905, + "DTLB miss rate": 0.00040138107909749404, + "L1I cache miss rate": 0.008805667140469892, + "L1D cache miss rate": 0.013872991452719873, + "L2D cache miss rate": 0.17251156337045664, + "LL cache miss rate": 0.9598863277288884 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell44", + "maxwell size": 44, + "matrix columns": 267300, + "task clock (msec)": 30.91, + "page faults": 2746, + "cycles": 33264810, + "instructions": 69273601, + "branch mispredictions": 315526, + "branches": 16760967, + "ITLB accesses": 22949528, + "ITLB misses": 5982, + "DTLB misses": 13101, + "DTLB accesses": 30809060, + "L1I cache accesses": 26535815, + "L1I cache misses": 240994, + "L1D cache misses": 384916, + "L1D cache accesses": 27816607, + "LL cache misses": 460494, + "LL cache accesses": 481127, + "L2D TLB accesses": 132497, + "L2D TLB misses": 18586, + "L2D cache misses": 246736, + "L2D cache accesses": 1446914, + "\u0394 watt": 18.18533333333333, + "branch miss rate": 0.01882504750471736, + "ITLB miss rate": 0.0002606589556003069, + "DTLB miss rate": 0.00042523205836205325, + "L1I cache miss rate": 0.009081839016438727, + "L1D cache miss rate": 0.013837633036983986, + "L2D cache miss rate": 0.1705256843184875, + "LL cache miss rate": 0.9571152730983711 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell42", + "maxwell size": 42, + "matrix columns": 232974, + "task clock (msec)": 34.51, + "page faults": 2760, + "cycles": 35380085, + "instructions": 74882678, + "branch mispredictions": 314478, + "branches": 16924367, + "ITLB accesses": 22759319, + "ITLB misses": 5648, + "DTLB misses": 12671, + "DTLB accesses": 30491137, + "L1I cache accesses": 26636435, + "L1I cache misses": 235128, + "L1D cache misses": 384387, + "L1D cache accesses": 27971398, + "LL cache misses": 465165, + "LL cache accesses": 484747, + "L2D TLB accesses": 132302, + "L2D TLB misses": 17635, + "L2D cache misses": 248647, + "L2D cache accesses": 1440208, + "\u0394 watt": 18.493333333333336, + "branch miss rate": 0.018581374417134772, + "ITLB miss rate": 0.000248162082529798, + "DTLB miss rate": 0.0004155633815819987, + "L1I cache miss rate": 0.008827307408067184, + "L1D cache miss rate": 0.013742144743712846, + "L2D cache miss rate": 0.1726465899370091, + "LL cache miss rate": 0.9596036695430812 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 35.58, + "page faults": 2792, + "cycles": 28042392, + "instructions": 75410629, + "branch mispredictions": 313846, + "branches": 16941333, + "ITLB accesses": 22593698, + "ITLB misses": 5385, + "DTLB misses": 12202, + "DTLB accesses": 30439426, + "L1I cache accesses": 26430404, + "L1I cache misses": 229508, + "L1D cache misses": 380816, + "L1D cache accesses": 27753131, + "LL cache misses": 457586, + "LL cache accesses": 478259, + "L2D TLB accesses": 132482, + "L2D TLB misses": 17723, + "L2D cache misses": 242435, + "L2D cache accesses": 1426434, + "\u0394 watt": 4.026666666666667, + "branch miss rate": 0.018525460776905808, + "ITLB miss rate": 0.00023834079750911072, + "DTLB miss rate": 0.00040086169824621526, + "L1I cache miss rate": 0.008683484368986565, + "L1D cache miss rate": 0.013721550912579917, + "L2D cache miss rate": 0.16995879234510675, + "LL cache miss rate": 0.9567744673910998 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell40", + "maxwell size": 40, + "matrix columns": 201720, + "task clock (msec)": 31.72, + "page faults": 2785, + "cycles": 30090507, + "instructions": 74992863, + "branch mispredictions": 317814, + "branches": 16791616, + "ITLB accesses": 22677480, + "ITLB misses": 5557, + "DTLB misses": 12446, + "DTLB accesses": 30413594, + "L1I cache accesses": 26843425, + "L1I cache misses": 239555, + "L1D cache misses": 385295, + "L1D cache accesses": 27801251, + "LL cache misses": 476478, + "LL cache accesses": 505239, + "L2D TLB accesses": 136570, + "L2D TLB misses": 19385, + "L2D cache misses": 267638, + "L2D cache accesses": 1488002, + "\u0394 watt": 12.689333333333337, + "branch miss rate": 0.0189269454470612, + "ITLB miss rate": 0.0002450448638914024, + "DTLB miss rate": 0.0004092249012070063, + "L1I cache miss rate": 0.008924159268051674, + "L1D cache miss rate": 0.013858908723208175, + "L2D cache miss rate": 0.17986400555913232, + "LL cache miss rate": 0.9430744657478936 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 34.04, + "page faults": 3255, + "cycles": 54898600, + "instructions": 81271980, + "branches": 17559162, + "branch mispredictions": 300852, + "ITLB accesses": 7217, + "ITLB misses": 13929, + "DTLB accesses": 17787712, + "DTLB misses": 12504, + "L1D cache accesses": 17720101, + "L1D cache misses": 590640, + "LL cache accesses": 79901, + "LL cache misses": 28153, + "branch miss rate": 0.017133619474551234, + "ITLB miss rate": 1.9300263267285576, + "DTLB miss rate": 0.0007029571875236118, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033331638459622774, + "L2D cache miss rate": null, + "LL cache miss rate": 0.352348531307493 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 32.29, + "page faults": 2770, + "cycles": 24172523, + "instructions": 55925355, + "branch mispredictions": 316558, + "branches": 16867022, + "ITLB accesses": 22725421, + "ITLB misses": 5397, + "DTLB misses": 12729, + "DTLB accesses": 30412650, + "L1I cache accesses": 26252660, + "L1I cache misses": 230453, + "L1D cache misses": 372224, + "L1D cache accesses": 27429044, + "LL cache misses": 468447, + "LL cache accesses": 497410, + "L2D TLB accesses": 134272, + "L2D TLB misses": 17416, + "L2D cache misses": 260097, + "L2D cache accesses": 1447346, + "\u0394 watt": 2.3826666666666654, + "branch miss rate": 0.018767865483308197, + "ITLB miss rate": 0.00023748734951928944, + "DTLB miss rate": 0.0004185429418350588, + "L1I cache miss rate": 0.008778272373161424, + "L1D cache miss rate": 0.013570432859417192, + "L2D cache miss rate": 0.17970616563005667, + "LL cache miss rate": 0.9417723809332341 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 40.34, + "page faults": 2792, + "cycles": 40725361, + "instructions": 68488967, + "branch mispredictions": 313193, + "branches": 17140574, + "ITLB accesses": 22851908, + "ITLB misses": 5193, + "DTLB misses": 12545, + "DTLB accesses": 30844809, + "L1I cache accesses": 26633420, + "L1I cache misses": 229672, + "L1D cache misses": 383161, + "L1D cache accesses": 27956131, + "LL cache misses": 463797, + "LL cache accesses": 489201, + "L2D TLB accesses": 135365, + "L2D TLB misses": 18059, + "L2D cache misses": 259969, + "L2D cache accesses": 1489942, + "\u0394 watt": 5.204000000000001, + "branch miss rate": 0.018272025195889007, + "ITLB miss rate": 0.0002272457949681926, + "DTLB miss rate": 0.0004067134926982365, + "L1I cache miss rate": 0.00862345128789318, + "L1D cache miss rate": 0.013705794982860826, + "L2D cache miss rate": 0.17448263086751029, + "LL cache miss rate": 0.9480704250400143 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.37, + "page faults": 2743, + "cycles": 45221267, + "instructions": 77850447, + "branches": 16924518, + "branch mispredictions": 289895, + "ITLB accesses": 7777, + "ITLB misses": 13661, + "DTLB accesses": 17612621, + "DTLB misses": 13387, + "L1D cache accesses": 17644914, + "L1D cache misses": 585931, + "LL cache accesses": 63994, + "LL cache misses": 11514, + "branch miss rate": 0.01712870050420343, + "ITLB miss rate": 1.7565899447087565, + "DTLB miss rate": 0.0007600799449440262, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03320679262024173, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17992311779229303 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell42", + "maxwell size": 42, + "matrix columns": 232974, + "task clock (msec)": 36.28, + "page faults": 2767, + "cycles": 36494687, + "instructions": 75735617, + "branch mispredictions": 318923, + "branches": 16870425, + "ITLB accesses": 22620706, + "ITLB misses": 5106, + "DTLB misses": 12161, + "DTLB accesses": 30340073, + "L1I cache accesses": 26386929, + "L1I cache misses": 231708, + "L1D cache misses": 380332, + "L1D cache accesses": 27654763, + "LL cache misses": 463631, + "LL cache accesses": 486406, + "L2D TLB accesses": 135704, + "L2D TLB misses": 17865, + "L2D cache misses": 252089, + "L2D cache accesses": 1453983, + "\u0394 watt": 10.441333333333336, + "branch miss rate": 0.01890426589727289, + "ITLB miss rate": 0.00022572239787741372, + "DTLB miss rate": 0.00040082303032032917, + "L1I cache miss rate": 0.008781165856777042, + "L1D cache miss rate": 0.013752856967170537, + "L2D cache miss rate": 0.17337823069458172, + "LL cache miss rate": 0.9531769756129653 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 31.43, + "page faults": 2786, + "cycles": 32825625, + "instructions": 75014125, + "branch mispredictions": 313433, + "branches": 16808190, + "ITLB accesses": 22734099, + "ITLB misses": 5182, + "DTLB misses": 12290, + "DTLB accesses": 30370350, + "L1I cache accesses": 26524678, + "L1I cache misses": 233449, + "L1D cache misses": 385270, + "L1D cache accesses": 27778791, + "LL cache misses": 460456, + "LL cache accesses": 481548, + "L2D TLB accesses": 130326, + "L2D TLB misses": 16370, + "L2D cache misses": 244717, + "L2D cache accesses": 1437795, + "\u0394 watt": 2.287999999999993, + "branch miss rate": 0.01864763546818545, + "ITLB miss rate": 0.00022793953699242712, + "DTLB miss rate": 0.0004046710031329899, + "L1I cache miss rate": 0.008801200150290232, + "L1D cache miss rate": 0.013869214106546249, + "L2D cache miss rate": 0.1702029844310211, + "LL cache miss rate": 0.9561995896566905 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 32.1, + "page faults": 3258, + "cycles": 52669502, + "instructions": 78956410, + "branches": 17149756, + "branch mispredictions": 297161, + "ITLB accesses": 8246, + "ITLB misses": 13925, + "DTLB accesses": 17627282, + "DTLB misses": 13397, + "L1D cache accesses": 17704168, + "L1D cache misses": 585596, + "LL cache accesses": 63570, + "LL cache misses": 21995, + "branch miss rate": 0.017327418535867216, + "ITLB miss rate": 1.6886975503274315, + "DTLB miss rate": 0.0007600150720910915, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0330767308579539, + "L2D cache miss rate": null, + "LL cache miss rate": 0.345996539248073 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell36", + "maxwell size": 36, + "matrix columns": 147852, + "task clock (msec)": 33.27, + "page faults": 2753, + "cycles": 35426575, + "instructions": 71870191, + "branch mispredictions": 316778, + "branches": 16824789, + "ITLB accesses": 22734331, + "ITLB misses": 5919, + "DTLB misses": 13018, + "DTLB accesses": 30547018, + "L1I cache accesses": 26772573, + "L1I cache misses": 239661, + "L1D cache misses": 388417, + "L1D cache accesses": 28077546, + "LL cache misses": 470825, + "LL cache accesses": 499788, + "L2D TLB accesses": 134784, + "L2D TLB misses": 17541, + "L2D cache misses": 259055, + "L2D cache accesses": 1481192, + "\u0394 watt": 18.52133333333333, + "branch miss rate": 0.018828051870368182, + "ITLB miss rate": 0.00026035514306534905, + "DTLB miss rate": 0.0004261627108741023, + "L1I cache miss rate": 0.008951735793194027, + "L1D cache miss rate": 0.013833723217833924, + "L2D cache miss rate": 0.1748962997369686, + "LL cache miss rate": 0.9420494289578781 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 30.27, + "page faults": 2770, + "cycles": 50550862, + "instructions": 79719924, + "branches": 17279759, + "branch mispredictions": 296867, + "ITLB accesses": 7116, + "ITLB misses": 13736, + "DTLB accesses": 17549541, + "DTLB misses": 12086, + "L1D cache accesses": 18242451, + "L1D cache misses": 585771, + "LL cache accesses": 74593, + "LL cache misses": 27394, + "branch miss rate": 0.017180042846662386, + "ITLB miss rate": 1.9302979201798762, + "DTLB miss rate": 0.0006886789802650679, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03211032333319684, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3672462563511321 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 36.52, + "page faults": 2774, + "cycles": 27452119, + "instructions": 75842779, + "branch mispredictions": 310172, + "branches": 16670816, + "ITLB accesses": 22606324, + "ITLB misses": 4704, + "DTLB misses": 11949, + "DTLB accesses": 30316550, + "L1I cache accesses": 26555286, + "L1I cache misses": 223942, + "L1D cache misses": 370100, + "L1D cache accesses": 27948620, + "LL cache misses": 457578, + "LL cache accesses": 478311, + "L2D TLB accesses": 133288, + "L2D TLB misses": 16236, + "L2D cache misses": 243738, + "L2D cache accesses": 1421114, + "\u0394 watt": 3.8346666666666636, + "branch miss rate": 0.018605687927933462, + "ITLB miss rate": 0.00020808336640667453, + "DTLB miss rate": 0.0003941411539241767, + "L1I cache miss rate": 0.008433047943825572, + "L1D cache miss rate": 0.013242156500034706, + "L2D cache miss rate": 0.17151192655902342, + "LL cache miss rate": 0.9566537252958849 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 27.88, + "page faults": 2760, + "cycles": 46921016, + "instructions": 78144329, + "branches": 16979990, + "branch mispredictions": 290347, + "ITLB accesses": 8645, + "ITLB misses": 14952, + "DTLB accesses": 18158039, + "DTLB misses": 13996, + "L1D cache accesses": 17789454, + "L1D cache misses": 588225, + "LL cache accesses": 72251, + "LL cache misses": 22822, + "branch miss rate": 0.017099362249329945, + "ITLB miss rate": 1.7295546558704453, + "DTLB miss rate": 0.0007707880790431169, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033065938954618845, + "L2D cache miss rate": null, + "LL cache miss rate": 0.31587106060815767 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell38", + "maxwell size": 38, + "matrix columns": 173394, + "task clock (msec)": 38.61, + "page faults": 2766, + "cycles": 39328079, + "instructions": 75494594, + "branch mispredictions": 310994, + "branches": 16926326, + "ITLB accesses": 22675910, + "ITLB misses": 5189, + "DTLB misses": 12161, + "DTLB accesses": 30479301, + "L1I cache accesses": 26347950, + "L1I cache misses": 229528, + "L1D cache misses": 378890, + "L1D cache accesses": 27541548, + "LL cache misses": 463900, + "LL cache accesses": 489009, + "L2D TLB accesses": 134701, + "L2D TLB misses": 18706, + "L2D cache misses": 254272, + "L2D cache accesses": 1461379, + "\u0394 watt": 19.607999999999997, + "branch miss rate": 0.0183733906578427, + "ITLB miss rate": 0.00022883315377420356, + "DTLB miss rate": 0.00039899208974641514, + "L1I cache miss rate": 0.00871141777633554, + "L1D cache miss rate": 0.013757033555267119, + "L2D cache miss rate": 0.17399456266991656, + "LL cache miss rate": 0.9486532967695891 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.22, + "page faults": 2725, + "cycles": 44632551, + "instructions": 77896058, + "branches": 16923307, + "branch mispredictions": 294777, + "ITLB accesses": 7728, + "ITLB misses": 15145, + "DTLB accesses": 17770114, + "DTLB misses": 14254, + "L1D cache accesses": 17635810, + "L1D cache misses": 580270, + "LL cache accesses": 67584, + "LL cache misses": 10728, + "branch miss rate": 0.017418404097969742, + "ITLB miss rate": 1.9597567287784678, + "DTLB miss rate": 0.0008021332896344953, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03290294009744945, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15873579545454544 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 28.86, + "page faults": 3264, + "cycles": 46795526, + "instructions": 78858968, + "branches": 17090137, + "branch mispredictions": 298327, + "ITLB accesses": 7578, + "ITLB misses": 14795, + "DTLB accesses": 17888783, + "DTLB misses": 14186, + "L1D cache accesses": 17505349, + "L1D cache misses": 581950, + "LL cache accesses": 64296, + "LL cache misses": 10879, + "branch miss rate": 0.017456091779720666, + "ITLB miss rate": 1.9523621008181578, + "DTLB miss rate": 0.000793010905213619, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033244124410201706, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16920181659823316 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell46", + "maxwell size": 46, + "matrix columns": 304842, + "task clock (msec)": 35.3, + "page faults": 2767, + "cycles": 22305313, + "instructions": 76129183, + "branch mispredictions": 314288, + "branches": 16855761, + "ITLB accesses": 22706804, + "ITLB misses": 4706, + "DTLB misses": 11592, + "DTLB accesses": 30488173, + "L1I cache accesses": 26439331, + "L1I cache misses": 237296, + "L1D cache misses": 383050, + "L1D cache accesses": 27780593, + "LL cache misses": 466156, + "LL cache accesses": 496369, + "L2D TLB accesses": 136658, + "L2D TLB misses": 17239, + "L2D cache misses": 257674, + "L2D cache accesses": 1463558, + "\u0394 watt": 18.817333333333327, + "branch miss rate": 0.018645731865799473, + "ITLB miss rate": 0.00020725065491383112, + "DTLB miss rate": 0.00038021300915604226, + "L1I cache miss rate": 0.00897511362900975, + "L1D cache miss rate": 0.013788402572976035, + "L2D cache miss rate": 0.1760599853234378, + "LL cache miss rate": 0.9391319764127091 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell44", + "maxwell size": 44, + "matrix columns": 267300, + "task clock (msec)": 36.17, + "page faults": 2781, + "cycles": 36947128, + "instructions": 75251973, + "branch mispredictions": 312703, + "branches": 16944024, + "ITLB accesses": 22792073, + "ITLB misses": 4719, + "DTLB misses": 12077, + "DTLB accesses": 30650016, + "L1I cache accesses": 26770419, + "L1I cache misses": 239352, + "L1D cache misses": 383301, + "L1D cache accesses": 28060808, + "LL cache misses": 465431, + "LL cache accesses": 487524, + "L2D TLB accesses": 137926, + "L2D TLB misses": 19434, + "L2D cache misses": 256811, + "L2D cache accesses": 1462171, + "\u0394 watt": 11.313333333333329, + "branch miss rate": 0.01845506120624003, + "ITLB miss rate": 0.0002070456689042721, + "DTLB miss rate": 0.00039402915809244604, + "L1I cache miss rate": 0.00894091347617682, + "L1D cache miss rate": 0.013659656557288015, + "L2D cache miss rate": 0.17563677572595818, + "LL cache miss rate": 0.9546832566191613 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 34.62, + "page faults": 2762, + "cycles": 31974608, + "instructions": 73403059, + "branch mispredictions": 315130, + "branches": 16935603, + "ITLB accesses": 22470994, + "ITLB misses": 5196, + "DTLB misses": 12525, + "DTLB accesses": 30180086, + "L1I cache accesses": 26316075, + "L1I cache misses": 227022, + "L1D cache misses": 373950, + "L1D cache accesses": 27566086, + "LL cache misses": 469059, + "LL cache accesses": 494454, + "L2D TLB accesses": 137297, + "L2D TLB misses": 17492, + "L2D cache misses": 256015, + "L2D cache accesses": 1464510, + "\u0394 watt": 8.707999999999995, + "branch miss rate": 0.018607545299686112, + "ITLB miss rate": 0.00023123142661156867, + "DTLB miss rate": 0.0004150087577616578, + "L1I cache miss rate": 0.008626742399844962, + "L1D cache miss rate": 0.013565581998111738, + "L2D cache miss rate": 0.174812736000437, + "LL cache miss rate": 0.948640318411824 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 36.02, + "page faults": 2754, + "cycles": 28216835, + "instructions": 72213172, + "branch mispredictions": 310681, + "branches": 16784941, + "ITLB accesses": 22988923, + "ITLB misses": 6290, + "DTLB misses": 13349, + "DTLB accesses": 30916878, + "L1I cache accesses": 26592108, + "L1I cache misses": 231135, + "L1D cache misses": 376017, + "L1D cache accesses": 27882306, + "LL cache misses": 442447, + "LL cache accesses": 466619, + "L2D TLB accesses": 159723, + "L2D TLB misses": 20452, + "L2D cache misses": 246420, + "L2D cache accesses": 1483875, + "\u0394 watt": 4.84, + "branch miss rate": 0.018509508016739527, + "ITLB miss rate": 0.0002736100338410808, + "DTLB miss rate": 0.0004317706335031629, + "L1I cache miss rate": 0.008691864518600783, + "L1D cache miss rate": 0.01348586447620222, + "L2D cache miss rate": 0.16606520090978014, + "LL cache miss rate": 0.9481975658942307 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 40.05, + "page faults": 2787, + "cycles": 35617476, + "instructions": 70762281, + "branch mispredictions": 315551, + "branches": 17092160, + "ITLB accesses": 22270930, + "ITLB misses": 5115, + "DTLB misses": 12145, + "DTLB accesses": 29954782, + "L1I cache accesses": 26790202, + "L1I cache misses": 235685, + "L1D cache misses": 385839, + "L1D cache accesses": 28190052, + "LL cache misses": 463555, + "LL cache accesses": 488806, + "L2D TLB accesses": 132085, + "L2D TLB misses": 17078, + "L2D cache misses": 248234, + "L2D cache accesses": 1456769, + "\u0394 watt": 5.119999999999997, + "branch miss rate": 0.01846173918334488, + "ITLB miss rate": 0.0002296715943159985, + "DTLB miss rate": 0.00040544444623232445, + "L1I cache miss rate": 0.00879743273305666, + "L1D cache miss rate": 0.01368706237221556, + "L2D cache miss rate": 0.17040038605983515, + "LL cache miss rate": 0.9483414688035744 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 32.53, + "page faults": 2785, + "cycles": 29159609, + "instructions": 70314265, + "branch mispredictions": 316033, + "branches": 16873830, + "ITLB accesses": 23069521, + "ITLB misses": 5362, + "DTLB misses": 12762, + "DTLB accesses": 30842612, + "L1I cache accesses": 26686872, + "L1I cache misses": 224715, + "L1D cache misses": 377368, + "L1D cache accesses": 28139495, + "LL cache misses": 474128, + "LL cache accesses": 497388, + "L2D TLB accesses": 133601, + "L2D TLB misses": 18428, + "L2D cache misses": 255989, + "L2D cache accesses": 1466756, + "\u0394 watt": 17.632, + "branch miss rate": 0.018729180037964113, + "ITLB miss rate": 0.0002324278861273279, + "DTLB miss rate": 0.0004137781845454594, + "L1I cache miss rate": 0.008420432338417182, + "L1D cache miss rate": 0.013410617354717986, + "L2D cache miss rate": 0.174527324244796, + "LL cache miss rate": 0.9532357033141129 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 34.13, + "page faults": 2773, + "cycles": 34485952, + "instructions": 69695625, + "branch mispredictions": 311099, + "branches": 16943287, + "ITLB accesses": 22528950, + "ITLB misses": 5140, + "DTLB misses": 12133, + "DTLB accesses": 30222074, + "L1I cache accesses": 26470998, + "L1I cache misses": 234215, + "L1D cache misses": 384868, + "L1D cache accesses": 27702279, + "LL cache misses": 460126, + "LL cache accesses": 479561, + "L2D TLB accesses": 130255, + "L2D TLB misses": 17812, + "L2D cache misses": 247457, + "L2D cache accesses": 1447544, + "\u0394 watt": 8.194666666666663, + "branch miss rate": 0.018361195203740573, + "ITLB miss rate": 0.0002281508902989265, + "DTLB miss rate": 0.0004014615277561692, + "L1I cache miss rate": 0.008847985255410468, + "L1D cache miss rate": 0.01389300858604449, + "L2D cache miss rate": 0.17094955317420404, + "LL cache miss rate": 0.9594733516695477 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 35.57, + "page faults": 2796, + "cycles": 37717911, + "instructions": 75108757, + "branch mispredictions": 316034, + "branches": 16843206, + "ITLB accesses": 22751785, + "ITLB misses": 5292, + "DTLB misses": 12280, + "DTLB accesses": 30599970, + "L1I cache accesses": 26975818, + "L1I cache misses": 232896, + "L1D cache misses": 377624, + "L1D cache accesses": 28434417, + "LL cache misses": 466672, + "LL cache accesses": 488601, + "L2D TLB accesses": 133066, + "L2D TLB misses": 18879, + "L2D cache misses": 253862, + "L2D cache accesses": 1461320, + "\u0394 watt": 16.958666666666666, + "branch miss rate": 0.01876329245156771, + "ITLB miss rate": 0.00023259713468635538, + "DTLB miss rate": 0.0004013075829812905, + "L1I cache miss rate": 0.008633510205325377, + "L1D cache miss rate": 0.013280525498377547, + "L2D cache miss rate": 0.17372101935236636, + "LL cache miss rate": 0.9551187983651281 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell38", + "maxwell size": 38, + "matrix columns": 173394, + "task clock (msec)": 35.22, + "page faults": 2756, + "cycles": 31538730, + "instructions": 73781559, + "branch mispredictions": 312797, + "branches": 16857034, + "ITLB accesses": 22645562, + "ITLB misses": 5056, + "DTLB misses": 11917, + "DTLB accesses": 30422134, + "L1I cache accesses": 26428674, + "L1I cache misses": 230943, + "L1D cache misses": 380047, + "L1D cache accesses": 27805181, + "LL cache misses": 468580, + "LL cache accesses": 494026, + "L2D TLB accesses": 140074, + "L2D TLB misses": 17751, + "L2D cache misses": 258749, + "L2D cache accesses": 1471487, + "\u0394 watt": 14.814666666666668, + "branch miss rate": 0.0185558740642037, + "ITLB miss rate": 0.00022326670453133379, + "DTLB miss rate": 0.00039172136971061926, + "L1I cache miss rate": 0.008738349869539426, + "L1D cache miss rate": 0.013668208094023917, + "L2D cache miss rate": 0.17584185249343012, + "LL cache miss rate": 0.9484925894588544 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 29.75, + "page faults": 3246, + "cycles": 47621399, + "instructions": 78692838, + "branches": 17085498, + "branch mispredictions": 288198, + "ITLB accesses": 8421, + "ITLB misses": 14846, + "DTLB accesses": 17760798, + "DTLB misses": 14557, + "L1D cache accesses": 17689718, + "L1D cache misses": 591509, + "LL cache accesses": 70978, + "LL cache misses": 26071, + "branch miss rate": 0.016867989449297877, + "ITLB miss rate": 1.7629735185844913, + "DTLB miss rate": 0.0008196140736469161, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03343801184394234, + "L2D cache miss rate": null, + "LL cache miss rate": 0.36731099777395815 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 28.3, + "page faults": 3228, + "cycles": 46109998, + "instructions": 78109004, + "branches": 16948597, + "branch mispredictions": 294052, + "ITLB accesses": 7273, + "ITLB misses": 14863, + "DTLB accesses": 17646854, + "DTLB misses": 14459, + "L1D cache accesses": 17507990, + "L1D cache misses": 578301, + "LL cache accesses": 59964, + "LL cache misses": 11600, + "branch miss rate": 0.017349636669041102, + "ITLB miss rate": 2.0435858655300425, + "DTLB miss rate": 0.0008193528432886678, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033030690558996205, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1934494029751184 + }, + { + "cpu": "altra", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell40", + "maxwell size": 40, + "matrix columns": 201720, + "task clock (msec)": 36.03, + "page faults": 2758, + "cycles": 24833611, + "instructions": 72242583, + "branch mispredictions": 316428, + "branches": 17120995, + "ITLB accesses": 22494089, + "ITLB misses": 5243, + "DTLB misses": 12074, + "DTLB accesses": 30209485, + "L1I cache accesses": 26372293, + "L1I cache misses": 222761, + "L1D cache misses": 371370, + "L1D cache accesses": 27608736, + "LL cache misses": 464046, + "LL cache accesses": 483309, + "L2D TLB accesses": 133913, + "L2D TLB misses": 18047, + "L2D cache misses": 246764, + "L2D cache accesses": 1439307, + "\u0394 watt": 9.798666666666662, + "branch miss rate": 0.018481869774507848, + "ITLB miss rate": 0.00023308345583588648, + "DTLB miss rate": 0.00039967579718753896, + "L1I cache miss rate": 0.008446781627975998, + "L1D cache miss rate": 0.013451177192610339, + "L2D cache miss rate": 0.1714463974676702, + "LL cache miss rate": 0.960143510673296 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 27.06, + "page faults": 3249, + "cycles": 45317452, + "instructions": 78195120, + "branches": 16987959, + "branch mispredictions": 292943, + "ITLB accesses": 6881, + "ITLB misses": 15043, + "DTLB accesses": 17690532, + "DTLB misses": 12893, + "L1D cache accesses": 17748819, + "L1D cache misses": 581005, + "LL cache accesses": 68511, + "LL cache misses": 10893, + "branch miss rate": 0.017244155110098865, + "ITLB miss rate": 2.1861648016276702, + "DTLB miss rate": 0.000728807929574984, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03273485407676984, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15899636554713842 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 35.48, + "page faults": 2788, + "cycles": 36826095, + "instructions": 75072991, + "branch mispredictions": 315151, + "branches": 16739311, + "ITLB accesses": 22682355, + "ITLB misses": 5397, + "DTLB misses": 12486, + "DTLB accesses": 30580327, + "L1I cache accesses": 27053327, + "L1I cache misses": 231094, + "L1D cache misses": 379632, + "L1D cache accesses": 28510815, + "LL cache misses": 466540, + "LL cache accesses": 494474, + "L2D TLB accesses": 134552, + "L2D TLB misses": 18143, + "L2D cache misses": 259011, + "L2D cache accesses": 1466702, + "\u0394 watt": 3.5280000000000022, + "branch miss rate": 0.018826999510314375, + "ITLB miss rate": 0.00023793825641120597, + "DTLB miss rate": 0.00040830171632893266, + "L1I cache miss rate": 0.008542165627170367, + "L1D cache miss rate": 0.013315368220796214, + "L2D cache miss rate": 0.17659415477718038, + "LL cache miss rate": 0.9435076465092199 + }, + { + "cpu": "altra", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell48", + "maxwell size": 48, + "matrix columns": 345744, + "task clock (msec)": 38.13, + "page faults": 2785, + "cycles": 36679069, + "instructions": 71192944, + "branch mispredictions": 317233, + "branches": 16868458, + "ITLB accesses": 22579878, + "ITLB misses": 4892, + "DTLB misses": 11824, + "DTLB accesses": 30360988, + "L1I cache accesses": 26394281, + "L1I cache misses": 232559, + "L1D cache misses": 379066, + "L1D cache accesses": 27646747, + "LL cache misses": 465751, + "LL cache accesses": 494631, + "L2D TLB accesses": 135055, + "L2D TLB misses": 18439, + "L2D cache misses": 258333, + "L2D cache accesses": 1471965, + "\u0394 watt": 19.902666666666676, + "branch miss rate": 0.018806283301058106, + "ITLB miss rate": 0.00021665307491918246, + "DTLB miss rate": 0.00038944714183873066, + "L1I cache miss rate": 0.008810961738264437, + "L1D cache miss rate": 0.01371105251550933, + "L2D cache miss rate": 0.1755021349013054, + "LL cache miss rate": 0.9416130408324589 + }, + { + "cpu": "altra", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 32.59, + "page faults": 2776, + "cycles": 38897101, + "instructions": 74428793, + "branch mispredictions": 315899, + "branches": 16920579, + "ITLB accesses": 22862806, + "ITLB misses": 5589, + "DTLB misses": 12828, + "DTLB accesses": 30658803, + "L1I cache accesses": 26893912, + "L1I cache misses": 251443, + "L1D cache misses": 389567, + "L1D cache accesses": 28166555, + "LL cache misses": 471274, + "LL cache accesses": 499214, + "L2D TLB accesses": 135744, + "L2D TLB misses": 17687, + "L2D cache misses": 261748, + "L2D cache accesses": 1495657, + "\u0394 watt": 2.847999999999999, + "branch miss rate": 0.018669514796154436, + "ITLB miss rate": 0.0002444581824295758, + "DTLB miss rate": 0.00041841163857571346, + "L1I cache miss rate": 0.009349439382414875, + "L1D cache miss rate": 0.013830835897396754, + "L2D cache miss rate": 0.17500536553501236, + "LL cache miss rate": 0.9440320183328192 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 31.81, + "page faults": 3265, + "cycles": 50932477, + "instructions": 78802385, + "branches": 17089783, + "branch mispredictions": 291874, + "ITLB accesses": 11277, + "ITLB misses": 13728, + "DTLB accesses": 17681908, + "DTLB misses": 12396, + "L1D cache accesses": 18127571, + "L1D cache misses": 590459, + "LL cache accesses": 75543, + "LL cache misses": 25465, + "branch miss rate": 0.017078859339524674, + "ITLB miss rate": 1.217345038574089, + "DTLB miss rate": 0.0007010555648180049, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03257242793311912, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3370927815945885 + } +] \ No newline at end of file diff --git a/analysis/analyze.py b/analysis/analyze.py new file mode 100755 index 0000000..148c2c3 --- /dev/null +++ b/analysis/analyze.py @@ -0,0 +1,384 @@ +#! /bin/python3 + +import argparse +import os +import re +import json + +from enum import Enum + +import math +import numpy as np +import matplotlib.pyplot as plt +import itertools + +class Stat(Enum): + CPU = 'cpu' + + SOLVER = 'solver' + LIN_ALG = 'linear algebra' + INPUT_FILE = 'input file' + MAXWELL_SIZE = 'maxwell size' + MATRIX_COLS = 'matrix columns' + + POWER_DELTA = 'Δ watt' + + TASK_CLK = 'task clock (msec)' + PAGE_FAULTS = 'page faults' + CYCLES = 'cycles' + INST = 'instructions' + + BR = 'branches' + BR_MISS = 'branch mispredictions' + ITLB = 'ITLB accesses' + ITLB_MISS = 'ITLB misses' + DTLB = 'DTLB accesses' + DTLB_MISS = 'DTLB misses' + L2D_TLB = 'L2D TLB accesses' + L2D_TLB_MISS = 'L2D TLB misses' + L1I_CACHE = 'L1I cache accesses' + L1I_CACHE_MISS = 'L1I cache misses' + L1D_CACHE = 'L1D cache accesses' + L1D_CACHE_MISS = 'L1D cache misses' + L2D_CACHE = 'L2D cache accesses' + L2D_CACHE_MISS = 'L2D cache misses' + LL_CACHE = 'LL cache accesses' + LL_CACHE_MISS = 'LL cache misses' + + BRANCH_MISS_RATE = 'branch miss rate' + ITLB_MISS_RATE = 'ITLB miss rate' + DTLB_MISS_RATE = 'DTLB miss rate' + L1I_CACHE_MISS_RATE = 'L1I cache miss rate' + L1D_CACHE_MISS_RATE = 'L1D cache miss rate' + L2D_CACHE_MISS_RATE = 'L2D cache miss rate' + LL_CACHE_MISS_RATE = 'LL cache miss rate' + +altra_names = { + Stat.TASK_CLK: 'task-clock:u', + Stat.PAGE_FAULTS: 'page-faults:u', + Stat.CYCLES: 'cycles:u', + Stat.INST: 'instructions:u', + + Stat.BR: 'BR_RETIRED:u', + Stat.BR_MISS: 'BR_MIS_PRED_RETIRED:u', + Stat.ITLB: 'L1I_TLB:u', + Stat.ITLB_MISS: 'ITLB_WALK:u', + Stat.DTLB: 'L1D_TLB:u', + Stat.DTLB_MISS: 'DTLB_WALK:u', + Stat.L2D_TLB: 'L2D_TLB:u', + Stat.L2D_TLB_MISS: 'L2D_TLB_REFILL:u', + Stat.L1I_CACHE: 'L1I_CACHE:u', + Stat.L1I_CACHE_MISS: 'L1I_CACHE_REFILL:u', + Stat.L1D_CACHE: 'L1D_CACHE:u', + Stat.L1D_CACHE_MISS: 'L1D_CACHE_REFILL:u', + Stat.L2D_CACHE: 'L2D_CACHE:u', + Stat.L2D_CACHE_MISS: 'L2D_CACHE_REFILL:u', + Stat.LL_CACHE: 'LL_CACHE_RD:u', + Stat.LL_CACHE_MISS: 'LL_CACHE_MISS_RD:u', +} + +xeon_names = { + Stat.TASK_CLK: 'task-clock:u', + Stat.PAGE_FAULTS: 'page-faults:u', + Stat.CYCLES: 'cycles:u', + Stat.INST: 'instructions:u', + + Stat.BR: 'branches:u', + Stat.BR_MISS: 'branch-misses:u', + Stat.ITLB: 'iTLB-loads:u', + Stat.ITLB_MISS: 'iTLB-load-misses:u', + Stat.DTLB: 'dTLB-loads:u', + Stat.DTLB_MISS: 'dTLB-load-misses:u', + Stat.L1I_CACHE: 'L1-icache-loads:u', + Stat.L1I_CACHE_MISS: 'L1-icache-load-misses:u', + Stat.L1D_CACHE: 'L1-dcache-loads:u', + Stat.L1D_CACHE_MISS: 'L1-dcache-load-misses:u', + Stat.LL_CACHE: 'LLC-loads:u', + Stat.LL_CACHE_MISS: 'LLC-load-misses:u', +} + +class Workload(Enum): + MINI_EM = 'mini_em' + +class CPU(Enum): + ALTRA = altra_names + XEON = xeon_names + +class Plot(Enum): + BOX = 'box' + LINE = 'line' + +def parse_input(filename: str, workload: Workload) -> dict[str, str | int]: + # Split filename into each input parameter. + filename_split = os.path.splitext(filename)[0].split('_') + + i = 0 + data = {Stat.CPU.value: filename_split[i]} + i += 1 + if workload == Workload.MINI_EM: + data[Stat.SOLVER.value] = filename_split[i] + i += 1 + data[Stat.LIN_ALG.value] = filename_split[i] + i += 1 + data[Stat.INPUT_FILE.value] = filename_split[i] + regex = r'^maxwell(\d+)' + data[Stat.MAXWELL_SIZE.value] = int(re.search(regex, filename_split[i]).group(1)) + x = data[Stat.MAXWELL_SIZE.value] + data[Stat.MATRIX_COLS.value] = 3 * x ** 3 + 6 * x ** 2 + 3 * x + i += 1 + + return data + +def parse_output(filename: str, data: dict[str, str]) -> dict[str, str | int | float]: + result: dict[str, int | float] = dict() + cpu: CPU = CPU[data['cpu'].upper()] + + with open(filename, 'r') as file: + for line in file: + for stat in [x for x in Stat if x in cpu.value]: + regex = r'^\W*([\d+(,|\.)?]+)\W*.*' + cpu.value[stat] + value = re.search(regex, line) + + if value is None: + continue + elif stat == Stat.TASK_CLK: + result[stat.value] = float(value.group(1).replace(',', '')) + else: + result[stat.value] = int(value.group(1).replace(',', '')) + + return result | parse_power(filename, cpu) + +# TODO CHANGE THIS +def parse_power(filename: str, cpu: CPU) -> dict[str, int]: + match cpu: + case CPU.ALTRA: + class Socket(Enum): + SOCKET1 = 'Socket1' + SOCKET2 = 'Socket2' + + data: dict[str, int] = {socket: list() for socket in Socket} + baseline_data: dict[str, int] = dict() + + filename = os.path.splitext(filename)[0] + "_power" + os.path.splitext(filename)[1] + with open(filename, 'r') as file: + for line in file: + regex = r'Start' + value = re.search(regex, line) + if value is not None: + for socket in Socket: + baseline_data[socket] = np.average(np.array(data[socket])) + data[socket] = list() + + continue + + for socket in Socket: + regex = r'^' + socket.value + r' (\d+\.\d+)' + value = re.search(regex, line) + if value is not None: + data[socket].append(float(value.group(1))) + break + + power_deltas: dict[str, int] = { + socket: np.max(np.array(data[socket])) - baseline_data[socket] + for socket in Socket + } + + return {Stat.POWER_DELTA.value: max(power_deltas.values())} + case CPU.XEON: + return {} + +def derive_stats(data: dict[str, str | int | float]) -> dict[str, int | float]: + result: dict[str, int | float] = dict() + + result[Stat.BRANCH_MISS_RATE.value] = data[Stat.BR_MISS.value] / data[Stat.BR.value] + result[Stat.ITLB_MISS_RATE.value] = data[Stat.ITLB_MISS.value] / data[Stat.ITLB.value] + result[Stat.DTLB_MISS_RATE.value] = data[Stat.DTLB_MISS.value] / data[Stat.DTLB.value] + result[Stat.L1I_CACHE_MISS_RATE.value] = ( + data[Stat.L1I_CACHE_MISS.value] / data[Stat.L1I_CACHE.value] + if Stat.L1I_CACHE_MISS.value in data and Stat.L1I_CACHE.value in data + else None) + result[Stat.L1D_CACHE_MISS_RATE.value] = ( + data[Stat.L1D_CACHE_MISS.value] / data[Stat.L1D_CACHE.value]) + result[Stat.L2D_CACHE_MISS_RATE.value] = ( + data[Stat.L2D_CACHE_MISS.value] / data[Stat.L2D_CACHE.value] + if Stat.L2D_CACHE_MISS.value in data and Stat.L2D_CACHE.value in data + else None) + result[Stat.LL_CACHE_MISS_RATE.value] = ( + data[Stat.LL_CACHE_MISS.value] / data[Stat.LL_CACHE.value]) + + return result + +def accumulate(stats_list: list[dict[str, str | int | float]], category: Stat, value: Stat): + category_list = np.array([stats[category.value] for stats in stats_list if value.value in stats]) + value_list = np.array([stats[value.value] for stats in stats_list if value.value in stats]) + + result: dict[np.ndarray] = dict() + for category in np.sort(np.unique(category_list)): + result[category] = value_list[category_list == category] + + return result + +def box_plot(ax, stats_list: list[dict[str, str | int | float]], x: Stat, y: Stat): + data: dict[str, np.ndarray] = accumulate(stats_list, x, y) + + print("Plotted data: " + str(data)) + + ax.boxplot(data.values(), tick_labels=data.keys()) + ax.set_ylabel(y.value) + +def line_plot( + ax, stats_list: list[dict[str, str | int | float]], + x: Stat, y: Stat, color: Stat +): + x_data: dict[str, np.ndarray] = accumulate(stats_list, color, x) + y_data: dict[str, np.ndarray] = accumulate(stats_list, color, y) + + for category in x_data.keys(): + sorted_indices = np.argsort(x_data[category]) + x_data[category] = x_data[category][sorted_indices] + y_data[category] = y_data[category][sorted_indices] + ax.plot(x_data[category], y_data[category], label=category) + print("Plotted x data: " + str(x_data[category])) + print("Plotted y data: " + str(y_data[category])) + + ax.set_ylabel(y.value) + ax.grid(True) + +def visualize( + stats_list: list[dict[str, str | int | float]], + plot: Plot, + rows: int, + size_multiplier: int, + font_size: int, + x: Stat, + y: Stat, + color: Stat, + filter_list: list[str] = [] +): + # Remove stats entries containing undesired values (like a specific CPU). + stats_list = [stats for stats in stats_list + if len([stats[key] for key in stats.keys() + if stats[key] in filter_list]) == 0] + + #x = Stat.MAXWELL_SIZE + #y = Stat.DTLB_MISS_RATE + #color = Stat.SOLVER + + if y is None: + ys = [stat for stat in Stat if stat.value in stats_list[0].keys() + and stat is not x + #and y != color + #and y != marker + and stat.value not in filter_list] + fig, axes = plt.subplots(rows, int(math.ceil(len(ys) / rows)), + figsize = (16 * size_multiplier, 9 * size_multiplier)) + match plot: + case Plot.BOX: + for i, y in enumerate(ys): + box_plot(axes[i % rows][int(i / rows)], stats_list, x, y) + case Plot.LINE: + for i, y in enumerate(ys): + line_plot(axes[i % rows][int(i / rows)], stats_list, x, y, color) + + handles, labels = axes[i % rows][int(i / rows)].get_legend_handles_labels() + else: + fig, ax = plt.subplots() + + match plot: + case Plot.BOX: + box_plot(ax, stats_list, x, y) + case Plot.LINE: + line_plot(ax, stats_list, x, y, color) + + handles, labels = ax.get_legend_handles_labels() + + #box_plot(ax, stats, x, y) + #line_plot(ax, stats, x, y, color) + + match plot: + case Plot.BOX: + title = f"{plot.value}_plot_of_{y.value.replace(' ', '_')}_vs_{x.value.replace(' ', '_')}_excluding_{filter_list}" + case Plot.LINE: + title = f"{plot.value}_plot_of_{y.value.replace(' ', '_')}_vs_{x.value.replace(' ', '_')}_by_{color.value.replace(' ', '_')}_excluding_{filter_list}" + fig.suptitle(title, fontsize = font_size) + fig.legend(handles, labels, fontsize = font_size) + fig.supxlabel(x.value, fontsize = font_size) + + plt.savefig(title + ".png", dpi = 100) + plt.show() + +def main(): + class Command(Enum): + PARSE = 'parse' + VISUALIZE = 'visualize' + + parser = argparse.ArgumentParser() + parser.add_argument('command', choices=[x.value for x in Command]) + parser.add_argument('workload', + choices=[x.name.lower() for x in Workload], + help='the workload to analyze') + parser.add_argument('filepath', + help='the output for the ' + Command.PARSE.value + ' command or the input for the ' + Command.VISUALIZE.value + ' command') + parser.add_argument('-i', '--input', + help='the input directory for the parse command') + parser.add_argument('-p', '--plot', + choices=[x.name.lower() for x in Plot], + help = 'the type of plot') + parser.add_argument('-r', '--rows', type=int, + help = 'the number of rows to display when -y is not specified', + default = 5) + parser.add_argument('-s', '--size', type=int, + help = 'figure size multiplier', + default = 4) + parser.add_argument('-fs', '--font_size', type=int, + help = 'font size', + default = 40) + parser.add_argument('-x', + choices=[x.name.lower() for x in Stat], + help = 'the name of the x axis') + parser.add_argument('-y', + choices=[x.name.lower() for x in Stat], + help = 'the name of the y axis') + parser.add_argument('-c', '--color', + choices=[x.name.lower() for x in Stat], + help = 'the name of the color') + parser.add_argument('-f', '--filter', nargs = '+', + help = 'a comma-separated string of names and values to filter out.', + default = []) + + args = parser.parse_args() + + stats_list: list[dict] = list() + if args.command == Command.PARSE.value: + if (args.input) is None: + print("An input directory is required with -i") + exit(-1) + + original_dir = os.getcwd() + os.chdir(args.input) + for filename in os.listdir(os.getcwd()): + if "output" not in filename: + continue + if "power" in filename: + continue + stats = parse_input(filename, Workload[args.workload.upper()]) + stats = stats | parse_output(filename, stats) + stats = stats | derive_stats(stats) + stats_list.append(stats) + print(filename + " parsed.") + + os.chdir(original_dir) + with open(args.filepath, 'w') as file: + json.dump(stats_list, file, indent = 2) + + elif args.command == Command.VISUALIZE.value: + with open(args.filepath, 'r') as file: + stats_list = json.load(file) + + x = Stat[args.x.upper()] if args.x is not None else None + y = Stat[args.y.upper()] if args.y is not None else None + color = Stat[args.color.upper()] if args.color is not None else None + visualize(stats_list, Plot[args.plot.upper()], args.rows, args.size, args.font_size, x, y, color, args.filter) + +if __name__ == '__main__': + main() diff --git a/analysis/analyze.py.bak b/analysis/analyze.py.bak new file mode 100644 index 0000000..b07a1da --- /dev/null +++ b/analysis/analyze.py.bak @@ -0,0 +1,302 @@ +#! /bin/python3 + +import os +import re +from dataclasses import dataclass +import json +import matplotlib.pyplot as plt +import numpy +from scipy.io import mmread +import math +import itertools +import argparse + +perf_stats = ["task-clock:u", "page-faults:u", "cycles:u", "instructions:u", + "branches:u", "branch-misses:u", "L1-dcache-loads:u", + "L1-dcache-load-misses:u", "LLC-loads:u", "LLC-load-misses:u", + "L1-icache-loads:u", "L1-icache-load-misses:u", "dTLB-loads:u", + "dTLB-load-misses:u", "iTLB-loads:u", "iTLB-load-misses:u"] + +def read_stats(filename): + # Split filename into each input parameter. + filename_split = os.path.splitext(filename)[0].split('_') + s = { + "solver": filename_split[0], + "linalg": filename_split[1], + "file": filename_split[2], + "time_steps": int(filename_split[3]), + "cores": int(re.match(r'\d+', filename_split[4]).group(0)), + "cpu": filename_split[5] } + + # Take filename and read SchurComplement matrix from corresponding dir. + print(os.path.splitext(filename)[0]) + m = mmread(os.path.splitext(filename)[0] + "/SchurComplement.mm") + s["cols"] = m.get_shape()[0] + s["nonzero"] = m.count_nonzero() + s["entries"] = m.get_shape()[0] * m.get_shape()[1] + s["trace"] = m.trace() + + # Compute ratio. + s["% nonzero entry"] = s["nonzero"] / s["entries"] + s["trace over cols"] = s["trace"] / s["cols"] + + # Obtain stats from perf in each log. + with open(filename) as file: + for line in file: + for perf_stat in perf_stats: + regex = r'^\W*([\d+(,|\.)?]+)\W*.*' + perf_stat + data = re.search(regex, line) + if data is None: + continue + match perf_stat: + case "task-clock:u": + s[perf_stat[:-2] + " (msec)"] = float(data.group(1).replace(',', '')) + case "page-faults:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "cycles:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "instructions:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "branches:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "branch-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "L1-dcache-loads:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "L1-dcache-load-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "LLC-loads:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "LLC-load-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "L1-icache-loads:u": + print("l1_icache_loads not expected! Exiting...") + exit(1) + case "L1-icache-load-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "dTLB-loads:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "dTLB-load-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "iTLB-loads:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case "iTLB-load-misses:u": + s[perf_stat[:-2]] = int(data.group(1).replace(',', '')) + case _: + print("Unexpected case! Exiting...") + exit(1) + + # Derive stats. + if s["cycles"] != 0: + s["insn per cycle"] = s["instructions"] / s["cycles"] + else: + s["insn per cycle"] = None + s["branch-miss-rate"] = s["branch-misses"] / s["branches"] + s["L1-dcache-miss-rate"] = s["L1-dcache-load-misses"] / s["L1-dcache-loads"] + s["LLC-miss-rate"] = s["LLC-load-misses"] / s["LLC-loads"] + s["dTLB-miss-rate"] = s["dTLB-load-misses"] / s["dTLB-loads"] + s["iTLB-miss-rate"] = s["iTLB-load-misses"] / s["iTLB-loads"] + + print(s) + return s + +def box_plot(ax, stat_list, x_name, y_name): + data = accumulate_to_dict(stat_list, x_name, y_name) + + for value in data.values(): + if type(value[0]) is str: + print(f"{y_name} values are strings... skipping box plot...") + return + + print("Plotted data: " + str(data)) + + ax.boxplot([numpy.array(y_axis) for y_axis in data.values()]) + ax.set_xticklabels(data.keys()) + ax.set_ylabel(y_name) + +def scatter_plot(ax, stat_list, x_name, y_name, color_name, size_name = None): + x_data = accumulate_to_dict(stat_list, color_name, x_name) + y_data = accumulate_to_dict(stat_list, color_name, y_name) + if size_name is not None: + size_data = accumulate_to_dict(stat_list, color_name, size_name) + + print(x_data) + print(y_data) + if size_name is not None: + print(size_data) + + for group in x_data.keys(): + print(f"Plotted x data for {group}: " + str(x_data[group])) + print(f"Plotted y data for {group}: " + str(y_data[group])) + ax.scatter(x_data[group], y_data[group], label = group, alpha = 0.3) + + #ax.legend() + #ax.set_xlabel(x_name) + ax.set_ylabel(y_name) + ax.grid(True) + +def line_plot(ax, stat_list, x_name, y_name, color_name = None, marker_name = None): + x_data = accumulate_to_dict(stat_list, color_name, x_name, marker_name) + y_data = accumulate_to_dict(stat_list, color_name, y_name, marker_name) + + for group in x_data.keys(): + sorted_data = [(x, y) for (x, y) in zip(x_data[group], y_data[group])] + sorted_data.sort() + x_data[group] = [x for (x, y) in sorted_data] + y_data[group] = [y for (x, y) in sorted_data] + print(f"Plotted x data for {group}: " + str(x_data[group])) + print(f"Plotted y data for {group}: " + str(y_data[group])) + ax.plot(x_data[group], y_data[group], label = group) + + ax.set_ylabel(y_name) + ax.grid(True) + +def basic_plot(ax, stat_list, x_name, y_name): + xs = accumulate_to_list(stat_list, x_name) + ys = accumulate_to_list(stat_list, y_name) + print("Plotted x data: " + str(xs)) + print("Plotted y data: " + str(ys)) + + ax.plot(xs, ys, 'o') + + ax.legend() + ax.set_xlabel(x_name) + ax.set_ylabel(y_name) + ax.grid(True) + +def accumulate_to_dict(stat_list, key_name, value_name, key2_name = None): + key_list = [stat[key_name] for stat in stat_list] + value_list = [stat[value_name] for stat in stat_list] + + if key2_name is not None: + key2_list = [stat[key2_name] for stat in stat_list] + data = {pair: list() for pair in + list(itertools.product(list(dict.fromkeys(key_list).keys()), + list(dict.fromkeys(key2_list).keys())))} + for key, key2, value in zip(key_list, key2_list, value_list): + data[(key, key2)].append(value) + else: + data = {key: list() for key in dict.fromkeys(key_list)} + for key, value in zip(key_list, value_list): + data[key].append(value) + + return data + +def accumulate_to_list(stat_list, name): + lst = [stat[name] for stat in stat_list] + return lst + +def show_visualizations(rows, plot, size_multiplier, stat_list, x_name, y_name = None, + color_name = None, marker_name = None, filter_list = []): + # Filter stats (removing values such as a specific CPU type) + stat_list = [stat for stat in stat_list + if len([stat[key] for key in stat.keys() + if stat[key] in filter_list]) + == 0] + + if y_name is None or y_name == "all": + # Filter y names (removing measurements such as cycles) + y_names = [y_name for y_name in stat_list[0].keys() if y_name != x_name + and y_name != color_name + and y_name != marker_name + and y_name not in filter_list] + + fig, axes = plt.subplots(rows, int(math.ceil(len(y_names) / rows)), + figsize = (16 * size_multiplier, 9 * size_multiplier)) + match plot: + case "box": + for i, y in enumerate(y_names): + box_plot(axes[i % rows][int(i / rows)], stat_list, x_name, y) + case "line": + for i, y in enumerate(y_names): + line_plot(axes[i % rows][int(i / rows)], stat_list, x_name, + y, color_name, marker_name) + case _: + print("Invalid plot type! Exiting...") + exit(1) + + handles, labels = axes[i % rows][int(i / rows)].get_legend_handles_labels() + else: + fig, ax = plt.subplots() + match plot: + case "box": + box_plot(ax, stat_list, x_name, y_name) + case "line": + line_plot(ax, stat_list, x_name, y_name, color_name, marker_name) + case _: + print("Invalid plot type! Exiting...") + exit(1) + + handles, labels = ax.get_legend_handles_labels() + + if y_name is None or y_name == "all": + y_name = "all" + fontsize = 'xx-large' + else: + fontsize = 'x-small' + + title = f"{plot}_plot_of_{y_name}_vs_{x_name}_by_{color_name}_and_{marker_name}_excluding_{filter_list}" + + fig.suptitle(title, fontsize = fontsize) + fig.legend(handles, labels, fontsize = fontsize) + fig.supxlabel(x_name, fontsize = fontsize) + + plt.savefig(title + ".png", dpi = 100) + plt.show() + +save_file = "stats.json" +samples_dir = "Samples/" +# save_file = "stats_avg_10.json" +# samples_dir = "Samples_Avg_10/" +# save_file = "tmp.json" +# samples_dir = "Old_Samples_Avg_10" + +def main(): + stat_list = list() + + # Get stats and accumulate into stat_list + if os.path.isfile(save_file): + with open(save_file, 'r') as file: + stat_list = json.load(file) + # print(*stat_list, sep='\n') + else: + os.chdir('./' + samples_dir) + for filename in os.listdir(os.getcwd()): + if os.path.splitext(filename)[1] == ".log": + stat_list.append(read_stats(filename)) + print(*stat_list, sep='\n') + os.chdir('../') + with open(save_file, 'w') as file: + json.dump(stat_list, file, indent = 2) + + # print(*stat_list, sep='\n') + + possible_args = [keys for keys in stat_list[0].keys()] + parser = argparse.ArgumentParser(epilog = 'Possible arguments are: ' + str(possible_args)) + parser.add_argument('-r', '--rows', help = 'Number of rows to display. Will split graphs into columns to evenly fill rows. Does not do anything if -y is not "all".', default = 0) + parser.add_argument('-p', '--plot', help = 'The type of plot. Only "box" and "line" are supported.') + parser.add_argument('-x', help = 'Name of the x axis. See below for possible arguments.') + parser.add_argument('-y', help = 'Name of the y axis. Can be "all" or omitted entirely to display all possible y values. See below for other possible arguments.') + parser.add_argument('-g1', '--group1', help = 'Name of the first group to group by. Does not need to be specified.') + parser.add_argument('-g2', '--group2', help = 'Name of the second group to group by. Does not need to be specified (and does not matter for box plots).') + parser.add_argument('-f', '--filter', nargs = '+', help = 'Names and string values to filter out of the visualization. Can be multiple space-separate arguments. Solver filters are "MueLu", "CG", "GMRES", and "ML". CPU filters are "Xeon,4216" and "Xeon,E5-2683"4216".', default = []) + parser.add_argument('-s', '--size', help = 'Multiplier for the resultant figure size if the visualization is too difficult to view. Default is 4.', default = 4) + + args = parser.parse_args() + + print(args) + + if args.plot == "line" and args.group1 is None: + print("Line plot requires at least one group! Exiting...") + exit(1) + + # Plot + show_visualizations(int(args.rows), args.plot, int(args.size), stat_list, args.x, args.y, args.group1, args.group2, args.filter) + #show_visualizations(4, "line", stat_list, "cols", None, "solver", "cpu", ["Xeon,E5-2683", "time_steps", "cpu", "linalg", "file", "cores"]) + #show_visualizations(4, "line", stat_list, "cols", None, "solver", "cpu", ["time_steps", "cpu", "linalg", "file", "cores"]) + #show_visualizations(4, "line", stat_list, "LLC-miss-rate", "branch-miss-rate", "solver", "cpu") + #show_visualizations(4, "box", stat_list, "solver", "cycles", None, None, ["Xeon,E5-2683", 0, None]) + #show_visualizations(4, "box", stat_list, "solver", None, None, None, [0, None]) + +if __name__ == '__main__': + main() diff --git a/analysis/xeon_mini-em.json b/analysis/xeon_mini-em.json new file mode 100644 index 0000000..13d8bf0 --- /dev/null +++ b/analysis/xeon_mini-em.json @@ -0,0 +1,1220 @@ +[ + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.15, + "page faults": 2747, + "cycles": 45356846, + "instructions": 77730625, + "branches": 16888960, + "branch mispredictions": 290099, + "ITLB accesses": 7996, + "ITLB misses": 14530, + "DTLB accesses": 17640131, + "DTLB misses": 14505, + "L1D cache accesses": 17593501, + "L1D cache misses": 574293, + "LL cache accesses": 62141, + "LL cache misses": 10948, + "branch miss rate": 0.017176842150138316, + "ITLB miss rate": 1.8171585792896447, + "DTLB miss rate": 0.0008222728051169234, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03264233764502017, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17617997779243977 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 31.09, + "page faults": 3226, + "cycles": 50299542, + "instructions": 77878301, + "branches": 16919929, + "branch mispredictions": 292094, + "ITLB accesses": 7694, + "ITLB misses": 13886, + "DTLB accesses": 17764513, + "DTLB misses": 12428, + "L1D cache accesses": 17864775, + "L1D cache misses": 574110, + "LL cache accesses": 65570, + "LL cache misses": 25239, + "branch miss rate": 0.01726331121129409, + "ITLB miss rate": 1.8047829477514947, + "DTLB miss rate": 0.0006995969999290158, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03213642489200116, + "L2D cache miss rate": null, + "LL cache miss rate": 0.38491688272075647 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 30.97, + "page faults": 3251, + "cycles": 49455687, + "instructions": 78584469, + "branches": 17039294, + "branch mispredictions": 294966, + "ITLB accesses": 8375, + "ITLB misses": 14366, + "DTLB accesses": 17909445, + "DTLB misses": 14174, + "L1D cache accesses": 17933423, + "L1D cache misses": 585835, + "LL cache accesses": 73948, + "LL cache misses": 24797, + "branch miss rate": 0.017310928492694593, + "ITLB miss rate": 1.7153432835820897, + "DTLB miss rate": 0.0007914259766285332, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0326672158460769, + "L2D cache miss rate": null, + "LL cache miss rate": 0.33533023205495754 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 27.37, + "page faults": 2759, + "cycles": 45348714, + "instructions": 78309411, + "branches": 17004157, + "branch mispredictions": 290236, + "ITLB accesses": 7105, + "ITLB misses": 14499, + "DTLB accesses": 17662985, + "DTLB misses": 12679, + "L1D cache accesses": 17593928, + "L1D cache misses": 590492, + "LL cache accesses": 65802, + "LL cache misses": 11589, + "branch miss rate": 0.017068532124232916, + "ITLB miss rate": 2.0406755805770582, + "DTLB miss rate": 0.0007178288381040917, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03356226079815718, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17611926689158383 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 28.93, + "page faults": 3248, + "cycles": 46661625, + "instructions": 77970945, + "branches": 16957089, + "branch mispredictions": 293037, + "ITLB accesses": 6602, + "ITLB misses": 13812, + "DTLB accesses": 17754589, + "DTLB misses": 12826, + "L1D cache accesses": 17698577, + "L1D cache misses": 582689, + "LL cache accesses": 67707, + "LL cache misses": 11248, + "branch miss rate": 0.017281091111805806, + "ITLB miss rate": 2.092093305059073, + "DTLB miss rate": 0.0007224047822227819, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03292292934059049, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16612757912771206 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 28.84, + "page faults": 3251, + "cycles": 46945863, + "instructions": 78778600, + "branches": 17089248, + "branch mispredictions": 295353, + "ITLB accesses": 9379, + "ITLB misses": 15619, + "DTLB accesses": 18062818, + "DTLB misses": 14970, + "L1D cache accesses": 17623645, + "L1D cache misses": 587213, + "LL cache accesses": 66288, + "LL cache misses": 21398, + "branch miss rate": 0.017282972311010992, + "ITLB miss rate": 1.6653161317837724, + "DTLB miss rate": 0.0008287743363189509, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03331961123819732, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3228035240164132 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 27.55, + "page faults": 2755, + "cycles": 45422244, + "instructions": 77660890, + "branches": 16872266, + "branch mispredictions": 287997, + "ITLB accesses": 6799, + "ITLB misses": 13737, + "DTLB accesses": 17603123, + "DTLB misses": 12447, + "L1D cache accesses": 17786984, + "L1D cache misses": 589987, + "LL cache accesses": 65223, + "LL cache misses": 11297, + "branch miss rate": 0.01706925436097321, + "ITLB miss rate": 2.0204441829680837, + "DTLB miss rate": 0.0007070904407132757, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03316959187684657, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17320577097036322 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 32.77, + "page faults": 3269, + "cycles": 51202082, + "instructions": 79122810, + "branches": 17153815, + "branch mispredictions": 292733, + "ITLB accesses": 8504, + "ITLB misses": 15168, + "DTLB accesses": 17800785, + "DTLB misses": 14899, + "L1D cache accesses": 17590854, + "L1D cache misses": 583954, + "LL cache accesses": 73737, + "LL cache misses": 22820, + "branch miss rate": 0.01706518345918969, + "ITLB miss rate": 1.7836312323612418, + "DTLB miss rate": 0.0008369855599064872, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033196455385281466, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3094782809173142 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 29.14, + "page faults": 2775, + "cycles": 47219016, + "instructions": 79448945, + "branches": 17214993, + "branch mispredictions": 302959, + "ITLB accesses": 7393, + "ITLB misses": 14201, + "DTLB accesses": 17643944, + "DTLB misses": 13924, + "L1D cache accesses": 17778098, + "L1D cache misses": 587272, + "LL cache accesses": 71448, + "LL cache misses": 11360, + "branch miss rate": 0.01759855493406242, + "ITLB miss rate": 1.9208710942783713, + "DTLB miss rate": 0.0007891659597196635, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03303345498489208, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15899675288321577 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell6", + "maxwell size": 6, + "matrix columns": 882, + "task clock (msec)": 29.05, + "page faults": 3258, + "cycles": 46083810, + "instructions": 79091262, + "branches": 17142768, + "branch mispredictions": 298230, + "ITLB accesses": 8268, + "ITLB misses": 13948, + "DTLB accesses": 17892377, + "DTLB misses": 14417, + "L1D cache accesses": 17736175, + "L1D cache misses": 584957, + "LL cache accesses": 71471, + "LL cache misses": 11675, + "branch miss rate": 0.01739684046357041, + "ITLB miss rate": 1.686985970004838, + "DTLB miss rate": 0.0008057621410503478, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03298101197129595, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16335296833680793 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 28.19, + "page faults": 3227, + "cycles": 46217337, + "instructions": 78002688, + "branches": 16938038, + "branch mispredictions": 295063, + "ITLB accesses": 7558, + "ITLB misses": 14020, + "DTLB accesses": 18065848, + "DTLB misses": 13921, + "L1D cache accesses": 17638435, + "L1D cache misses": 581076, + "LL cache accesses": 70695, + "LL cache misses": 11275, + "branch miss rate": 0.01742014039642608, + "ITLB miss rate": 1.8549880920878539, + "DTLB miss rate": 0.0007705699726910135, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032943739056214456, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15948794115566872 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 29.43, + "page faults": 3250, + "cycles": 47251998, + "instructions": 79570608, + "branches": 17217585, + "branch mispredictions": 297938, + "ITLB accesses": 7313, + "ITLB misses": 13770, + "DTLB accesses": 17661992, + "DTLB misses": 14273, + "L1D cache accesses": 17655610, + "L1D cache misses": 575913, + "LL cache accesses": 66224, + "LL cache misses": 11211, + "branch miss rate": 0.017304285124772143, + "ITLB miss rate": 1.882948174483796, + "DTLB miss rate": 0.000808119491844408, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03261926379207515, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16928907948779898 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 26.61, + "page faults": 2740, + "cycles": 44146579, + "instructions": 77589125, + "branches": 16882608, + "branch mispredictions": 287427, + "ITLB accesses": 8158, + "ITLB misses": 13803, + "DTLB accesses": 17689973, + "DTLB misses": 13768, + "L1D cache accesses": 18106537, + "L1D cache misses": 582639, + "LL cache accesses": 68276, + "LL cache misses": 10203, + "branch miss rate": 0.01702503546845369, + "ITLB miss rate": 1.6919588134346653, + "DTLB miss rate": 0.0007782940087019918, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03217837844972785, + "L2D cache miss rate": null, + "LL cache miss rate": 0.14943757689378406 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 28.52, + "page faults": 3233, + "cycles": 46252915, + "instructions": 77530146, + "branches": 16850069, + "branch mispredictions": 291772, + "ITLB accesses": 8471, + "ITLB misses": 14690, + "DTLB accesses": 17714701, + "DTLB misses": 14252, + "L1D cache accesses": 17666047, + "L1D cache misses": 580912, + "LL cache accesses": 64066, + "LL cache misses": 12171, + "branch miss rate": 0.017315774790002344, + "ITLB miss rate": 1.7341518120646913, + "DTLB miss rate": 0.000804529526069901, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03288296470625262, + "L2D cache miss rate": null, + "LL cache miss rate": 0.18997596228888958 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 34.35, + "page faults": 3283, + "cycles": 54352938, + "instructions": 79973336, + "branches": 17312182, + "branch mispredictions": 300210, + "ITLB accesses": 8875, + "ITLB misses": 14581, + "DTLB accesses": 17769566, + "DTLB misses": 14200, + "L1D cache accesses": 17717809, + "L1D cache misses": 580080, + "LL cache accesses": 70529, + "LL cache misses": 30347, + "branch miss rate": 0.017340968342407677, + "ITLB miss rate": 1.6429295774647887, + "DTLB miss rate": 0.0007991191231119545, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032739939797296606, + "L2D cache miss rate": null, + "LL cache miss rate": 0.43027690737143587 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 32.32, + "page faults": 3252, + "cycles": 51882346, + "instructions": 78686206, + "branches": 17074216, + "branch mispredictions": 299620, + "ITLB accesses": 7747, + "ITLB misses": 14130, + "DTLB accesses": 17740758, + "DTLB misses": 13623, + "L1D cache accesses": 17726291, + "L1D cache misses": 582972, + "LL cache accesses": 73138, + "LL cache misses": 27275, + "branch miss rate": 0.01754809708393053, + "ITLB miss rate": 1.8239318445850006, + "DTLB miss rate": 0.0007678927811314489, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03288742128852561, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3729251551860866 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 31.26, + "page faults": 3259, + "cycles": 50199983, + "instructions": 79261227, + "branches": 17172254, + "branch mispredictions": 298417, + "ITLB accesses": 9179, + "ITLB misses": 15360, + "DTLB accesses": 17844537, + "DTLB misses": 15261, + "L1D cache accesses": 17823138, + "L1D cache misses": 578648, + "LL cache accesses": 70731, + "LL cache misses": 24360, + "branch miss rate": 0.017377858491960343, + "ITLB miss rate": 1.6733849003159385, + "DTLB miss rate": 0.0008552197235490055, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03246611230861816, + "L2D cache miss rate": null, + "LL cache miss rate": 0.34440344403444034 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 28.69, + "page faults": 2779, + "cycles": 46774201, + "instructions": 79233239, + "branches": 17203509, + "branch mispredictions": 294550, + "ITLB accesses": 6215, + "ITLB misses": 14776, + "DTLB accesses": 17504586, + "DTLB misses": 12564, + "L1D cache accesses": 17818063, + "L1D cache misses": 586492, + "LL cache accesses": 70731, + "LL cache misses": 11390, + "branch miss rate": 0.01712150701348196, + "ITLB miss rate": 2.377473853580048, + "DTLB miss rate": 0.0007177547643800316, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03291558684016327, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1610326448092067 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 30.03, + "page faults": 2733, + "cycles": 49573320, + "instructions": 77908496, + "branches": 16944615, + "branch mispredictions": 291550, + "ITLB accesses": 8381, + "ITLB misses": 14712, + "DTLB accesses": 18063155, + "DTLB misses": 14597, + "L1D cache accesses": 17745590, + "L1D cache misses": 586554, + "LL cache accesses": 73017, + "LL cache misses": 23235, + "branch miss rate": 0.01720605631936754, + "ITLB miss rate": 1.7553991170504712, + "DTLB miss rate": 0.0008081091038636384, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03305350794197319, + "L2D cache miss rate": null, + "LL cache miss rate": 0.31821356670364437 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 31.91, + "page faults": 3239, + "cycles": 52496644, + "instructions": 78693898, + "branches": 17079988, + "branch mispredictions": 300099, + "ITLB accesses": 7149, + "ITLB misses": 13328, + "DTLB accesses": 17483064, + "DTLB misses": 11695, + "L1D cache accesses": 18000078, + "L1D cache misses": 578543, + "LL cache accesses": 69713, + "LL cache misses": 23474, + "branch miss rate": 0.017570211407642675, + "ITLB miss rate": 1.8643166876486221, + "DTLB miss rate": 0.0006689330886165034, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032141138499510946, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3367234231778865 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 28.84, + "page faults": 3238, + "cycles": 46524233, + "instructions": 78604006, + "branches": 17054156, + "branch mispredictions": 298876, + "ITLB accesses": 6035, + "ITLB misses": 13621, + "DTLB accesses": 17783780, + "DTLB misses": 12598, + "L1D cache accesses": 17494763, + "L1D cache misses": 576538, + "LL cache accesses": 67358, + "LL cache misses": 11047, + "branch miss rate": 0.01752511235384501, + "ITLB miss rate": 2.257000828500414, + "DTLB miss rate": 0.0007083983270148416, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032954890557820075, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16400427566139136 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 27.25, + "page faults": 2736, + "cycles": 45435644, + "instructions": 78271647, + "branches": 16994482, + "branch mispredictions": 291926, + "ITLB accesses": 7616, + "ITLB misses": 14515, + "DTLB accesses": 17745451, + "DTLB misses": 14184, + "L1D cache accesses": 17720302, + "L1D cache misses": 572377, + "LL cache accesses": 63704, + "LL cache misses": 11538, + "branch miss rate": 0.017177693324221355, + "ITLB miss rate": 1.9058560924369747, + "DTLB miss rate": 0.0007993034383854205, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03230063460543731, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1811189250282557 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 29.36, + "page faults": 3257, + "cycles": 47397126, + "instructions": 79359110, + "branches": 17182349, + "branch mispredictions": 299371, + "ITLB accesses": 7154, + "ITLB misses": 14997, + "DTLB accesses": 17594878, + "DTLB misses": 14791, + "L1D cache accesses": 17816160, + "L1D cache misses": 574506, + "LL cache accesses": 67016, + "LL cache misses": 11106, + "branch miss rate": 0.01742317072013844, + "ITLB miss rate": 2.096309756779424, + "DTLB miss rate": 0.0008406423733088687, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0322463426462268, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1657216187179181 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell12", + "maxwell size": 12, + "matrix columns": 6084, + "task clock (msec)": 27.95, + "page faults": 2757, + "cycles": 46254053, + "instructions": 78828949, + "branches": 17081818, + "branch mispredictions": 295191, + "ITLB accesses": 7066, + "ITLB misses": 14373, + "DTLB accesses": 17813443, + "DTLB misses": 14801, + "L1D cache accesses": 17558896, + "L1D cache misses": 577762, + "LL cache accesses": 58576, + "LL cache misses": 11107, + "branch miss rate": 0.017281006038116085, + "ITLB miss rate": 2.0341069912255874, + "DTLB miss rate": 0.0008308893457598287, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03290423270346837, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1896169079486479 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 34.59, + "page faults": 3275, + "cycles": 54313195, + "instructions": 79842410, + "branches": 17275386, + "branch mispredictions": 297876, + "ITLB accesses": 11494, + "ITLB misses": 13129, + "DTLB accesses": 17538031, + "DTLB misses": 11745, + "L1D cache accesses": 17678682, + "L1D cache misses": 581897, + "LL cache accesses": 73167, + "LL cache misses": 31335, + "branch miss rate": 0.01724279851113023, + "ITLB miss rate": 1.142248129458848, + "DTLB miss rate": 0.0006696874922846242, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032915179989096474, + "L2D cache miss rate": null, + "LL cache miss rate": 0.42826684160892203 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 31.1, + "page faults": 3260, + "cycles": 49512609, + "instructions": 79157466, + "branches": 17163022, + "branch mispredictions": 296724, + "ITLB accesses": 7966, + "ITLB misses": 13125, + "DTLB accesses": 17723474, + "DTLB misses": 13123, + "L1D cache accesses": 17642308, + "L1D cache misses": 578691, + "LL cache accesses": 70574, + "LL cache misses": 30238, + "branch miss rate": 0.017288563750602896, + "ITLB miss rate": 1.647627416520211, + "DTLB miss rate": 0.0007404304596265946, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032801320552843766, + "L2D cache miss rate": null, + "LL cache miss rate": 0.42845807237792954 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell14", + "maxwell size": 14, + "matrix columns": 9450, + "task clock (msec)": 34.04, + "page faults": 3255, + "cycles": 54898600, + "instructions": 81271980, + "branches": 17559162, + "branch mispredictions": 300852, + "ITLB accesses": 7217, + "ITLB misses": 13929, + "DTLB accesses": 17787712, + "DTLB misses": 12504, + "L1D cache accesses": 17720101, + "L1D cache misses": 590640, + "LL cache accesses": 79901, + "LL cache misses": 28153, + "branch miss rate": 0.017133619474551234, + "ITLB miss rate": 1.9300263267285576, + "DTLB miss rate": 0.0007029571875236118, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033331638459622774, + "L2D cache miss rate": null, + "LL cache miss rate": 0.352348531307493 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell26", + "maxwell size": 26, + "matrix columns": 56862, + "task clock (msec)": 29.28, + "page faults": 3259, + "cycles": 46946807, + "instructions": 78697814, + "branches": 17067789, + "branch mispredictions": 295729, + "ITLB accesses": 7619, + "ITLB misses": 14205, + "DTLB accesses": 17816823, + "DTLB misses": 13612, + "L1D cache accesses": 17701764, + "L1D cache misses": 580555, + "LL cache accesses": 65809, + "LL cache misses": 11396, + "branch miss rate": 0.017326731658095843, + "ITLB miss rate": 1.8644179026118912, + "DTLB miss rate": 0.0007639970380802459, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03279644898666596, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1731678037958334 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell24", + "maxwell size": 24, + "matrix columns": 45000, + "task clock (msec)": 26.65, + "page faults": 2750, + "cycles": 45052601, + "instructions": 78090758, + "branches": 16972059, + "branch mispredictions": 289279, + "ITLB accesses": 8033, + "ITLB misses": 14485, + "DTLB accesses": 18107303, + "DTLB misses": 14607, + "L1D cache accesses": 17698985, + "L1D cache misses": 571365, + "LL cache accesses": 63864, + "LL cache misses": 11859, + "branch miss rate": 0.01704442578239918, + "ITLB miss rate": 1.8031868542263165, + "DTLB miss rate": 0.0008066910903296863, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03228235969463786, + "L2D cache miss rate": null, + "LL cache miss rate": 0.18569146937241637 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.37, + "page faults": 2743, + "cycles": 45221267, + "instructions": 77850447, + "branches": 16924518, + "branch mispredictions": 289895, + "ITLB accesses": 7777, + "ITLB misses": 13661, + "DTLB accesses": 17612621, + "DTLB misses": 13387, + "L1D cache accesses": 17644914, + "L1D cache misses": 585931, + "LL cache accesses": 63994, + "LL cache misses": 11514, + "branch miss rate": 0.01712870050420343, + "ITLB miss rate": 1.7565899447087565, + "DTLB miss rate": 0.0007600799449440262, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03320679262024173, + "L2D cache miss rate": null, + "LL cache miss rate": 0.17992311779229303 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 32.1, + "page faults": 3258, + "cycles": 52669502, + "instructions": 78956410, + "branches": 17149756, + "branch mispredictions": 297161, + "ITLB accesses": 8246, + "ITLB misses": 13925, + "DTLB accesses": 17627282, + "DTLB misses": 13397, + "L1D cache accesses": 17704168, + "L1D cache misses": 585596, + "LL cache accesses": 63570, + "LL cache misses": 21995, + "branch miss rate": 0.017327418535867216, + "ITLB miss rate": 1.6886975503274315, + "DTLB miss rate": 0.0007600150720910915, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.0330767308579539, + "L2D cache miss rate": null, + "LL cache miss rate": 0.345996539248073 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 30.27, + "page faults": 2770, + "cycles": 50550862, + "instructions": 79719924, + "branches": 17279759, + "branch mispredictions": 296867, + "ITLB accesses": 7116, + "ITLB misses": 13736, + "DTLB accesses": 17549541, + "DTLB misses": 12086, + "L1D cache accesses": 18242451, + "L1D cache misses": 585771, + "LL cache accesses": 74593, + "LL cache misses": 27394, + "branch miss rate": 0.017180042846662386, + "ITLB miss rate": 1.9302979201798762, + "DTLB miss rate": 0.0006886789802650679, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03211032333319684, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3672462563511321 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell22", + "maxwell size": 22, + "matrix columns": 34914, + "task clock (msec)": 29.22, + "page faults": 3273, + "cycles": 46785271, + "instructions": 79142630, + "branches": 17154865, + "branch mispredictions": 298022, + "ITLB accesses": 7805, + "ITLB misses": 14231, + "DTLB accesses": 17923713, + "DTLB misses": 14087, + "L1D cache accesses": 17649159, + "L1D cache misses": 578416, + "LL cache accesses": 59413, + "LL cache misses": 10987, + "branch miss rate": 0.01737244799070118, + "ITLB miss rate": 1.8233183856502242, + "DTLB miss rate": 0.0007859420645711076, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.032773006351180814, + "L2D cache miss rate": null, + "LL cache miss rate": 0.18492585797721037 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell28", + "maxwell size": 28, + "matrix columns": 70644, + "task clock (msec)": 28.17, + "page faults": 3226, + "cycles": 46421141, + "instructions": 79459038, + "branches": 17205687, + "branch mispredictions": 297234, + "ITLB accesses": 6222, + "ITLB misses": 14597, + "DTLB accesses": 17730244, + "DTLB misses": 12585, + "L1D cache accesses": 17733537, + "L1D cache misses": 585471, + "LL cache accesses": 71081, + "LL cache misses": 12017, + "branch miss rate": 0.0172753346030298, + "ITLB miss rate": 2.3460302153648342, + "DTLB miss rate": 0.0007098041064747897, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03301490277996995, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16906064911861116 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell18", + "maxwell size": 18, + "matrix columns": 19494, + "task clock (msec)": 27.88, + "page faults": 2760, + "cycles": 46921016, + "instructions": 78144329, + "branches": 16979990, + "branch mispredictions": 290347, + "ITLB accesses": 8645, + "ITLB misses": 14952, + "DTLB accesses": 18158039, + "DTLB misses": 13996, + "L1D cache accesses": 17789454, + "L1D cache misses": 588225, + "LL cache accesses": 72251, + "LL cache misses": 22822, + "branch miss rate": 0.017099362249329945, + "ITLB miss rate": 1.7295546558704453, + "DTLB miss rate": 0.0007707880790431169, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033065938954618845, + "L2D cache miss rate": null, + "LL cache miss rate": 0.31587106060815767 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell8", + "maxwell size": 8, + "matrix columns": 1944, + "task clock (msec)": 27.22, + "page faults": 2725, + "cycles": 44632551, + "instructions": 77896058, + "branches": 16923307, + "branch mispredictions": 294777, + "ITLB accesses": 7728, + "ITLB misses": 15145, + "DTLB accesses": 17770114, + "DTLB misses": 14254, + "L1D cache accesses": 17635810, + "L1D cache misses": 580270, + "LL cache accesses": 67584, + "LL cache misses": 10728, + "branch miss rate": 0.017418404097969742, + "ITLB miss rate": 1.9597567287784678, + "DTLB miss rate": 0.0008021332896344953, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03290294009744945, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15873579545454544 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell4", + "maxwell size": 4, + "matrix columns": 300, + "task clock (msec)": 28.86, + "page faults": 3264, + "cycles": 46795526, + "instructions": 78858968, + "branches": 17090137, + "branch mispredictions": 298327, + "ITLB accesses": 7578, + "ITLB misses": 14795, + "DTLB accesses": 17888783, + "DTLB misses": 14186, + "L1D cache accesses": 17505349, + "L1D cache misses": 581950, + "LL cache accesses": 64296, + "LL cache misses": 10879, + "branch miss rate": 0.017456091779720666, + "ITLB miss rate": 1.9523621008181578, + "DTLB miss rate": 0.000793010905213619, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033244124410201706, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16920181659823316 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell30", + "maxwell size": 30, + "matrix columns": 86490, + "task clock (msec)": 28.01, + "page faults": 3229, + "cycles": 46636132, + "instructions": 79019996, + "branches": 17133654, + "branch mispredictions": 297073, + "ITLB accesses": 7653, + "ITLB misses": 14613, + "DTLB accesses": 17723280, + "DTLB misses": 13985, + "L1D cache accesses": 17644898, + "L1D cache misses": 579478, + "LL cache accesses": 68634, + "LL cache misses": 11498, + "branch miss rate": 0.017338566542781826, + "ITLB miss rate": 1.9094472755782046, + "DTLB miss rate": 0.0007890751599026817, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03284110795086489, + "L2D cache miss rate": null, + "LL cache miss rate": 0.16752629891890317 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell16", + "maxwell size": 16, + "matrix columns": 13872, + "task clock (msec)": 29.75, + "page faults": 3246, + "cycles": 47621399, + "instructions": 78692838, + "branches": 17085498, + "branch mispredictions": 288198, + "ITLB accesses": 8421, + "ITLB misses": 14846, + "DTLB accesses": 17760798, + "DTLB misses": 14557, + "L1D cache accesses": 17689718, + "L1D cache misses": 591509, + "LL cache accesses": 70978, + "LL cache misses": 26071, + "branch miss rate": 0.016867989449297877, + "ITLB miss rate": 1.7629735185844913, + "DTLB miss rate": 0.0008196140736469161, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03343801184394234, + "L2D cache miss rate": null, + "LL cache miss rate": 0.36731099777395815 + }, + { + "cpu": "xeon", + "solver": "GMRES", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 28.3, + "page faults": 3228, + "cycles": 46109998, + "instructions": 78109004, + "branches": 16948597, + "branch mispredictions": 294052, + "ITLB accesses": 7273, + "ITLB misses": 14863, + "DTLB accesses": 17646854, + "DTLB misses": 14459, + "L1D cache accesses": 17507990, + "L1D cache misses": 578301, + "LL cache accesses": 59964, + "LL cache misses": 11600, + "branch miss rate": 0.017349636669041102, + "ITLB miss rate": 2.0435858655300425, + "DTLB miss rate": 0.0008193528432886678, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.033030690558996205, + "L2D cache miss rate": null, + "LL cache miss rate": 0.1934494029751184 + }, + { + "cpu": "xeon", + "solver": "MueLu", + "linear algebra": "Tpetra", + "input file": "maxwell20", + "maxwell size": 20, + "matrix columns": 26460, + "task clock (msec)": 27.06, + "page faults": 3249, + "cycles": 45317452, + "instructions": 78195120, + "branches": 16987959, + "branch mispredictions": 292943, + "ITLB accesses": 6881, + "ITLB misses": 15043, + "DTLB accesses": 17690532, + "DTLB misses": 12893, + "L1D cache accesses": 17748819, + "L1D cache misses": 581005, + "LL cache accesses": 68511, + "LL cache misses": 10893, + "branch miss rate": 0.017244155110098865, + "ITLB miss rate": 2.1861648016276702, + "DTLB miss rate": 0.000728807929574984, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03273485407676984, + "L2D cache miss rate": null, + "LL cache miss rate": 0.15899636554713842 + }, + { + "cpu": "xeon", + "solver": "CG", + "linear algebra": "Tpetra", + "input file": "maxwell10", + "maxwell size": 10, + "matrix columns": 3630, + "task clock (msec)": 31.81, + "page faults": 3265, + "cycles": 50932477, + "instructions": 78802385, + "branches": 17089783, + "branch mispredictions": 291874, + "ITLB accesses": 11277, + "ITLB misses": 13728, + "DTLB accesses": 17681908, + "DTLB misses": 12396, + "L1D cache accesses": 18127571, + "L1D cache misses": 590459, + "LL cache accesses": 75543, + "LL cache misses": 25465, + "branch miss rate": 0.017078859339524674, + "ITLB miss rate": 1.217345038574089, + "DTLB miss rate": 0.0007010555648180049, + "L1I cache miss rate": null, + "L1D cache miss rate": 0.03257242793311912, + "L2D cache miss rate": null, + "LL cache miss rate": 0.3370927815945885 + } +] \ No newline at end of file