SDPA 7.4.0 と超スレッド並列 その３

スレッド数から見ると 160, 80, 320, 640 の順で速くなっている。この規模の問題だとあまり多数のスレッドは要らないのかもしれない。

◯ 640 スレッド
SDPA start at [Thu May 1 16:58:29 2014]
param is /home2/fujisawa/param.sdpa
data is /home2/fujisawa/data/N.4P.DZ.pqgt1t2p.dat-s : sparse
out is /home2/fujisawa/out.N.4P.640
NumThreads is set as 640
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
0 1.0e+04 1.0e+00 1.0e+00 -0.00e+00 -1.35e+05 8.6e-01 1.0e+00 2.00e-01
1 2.0e+03 1.4e-01 4.1e-15 +6.47e+00 -2.09e+05 7.8e-01 7.8e-01 2.00e-01
2 6.0e+02 3.1e-02 4.8e-15 -1.18e+01 -2.66e+05 7.5e-01 7.5e-01 2.00e-01

中略

39 1.6e-09 4.2e-10 3.3e-12 -5.44e+01 -5.44e+01 5.2e-01 5.1e-01 1.00e-01

phase.value = pdOPT
Iteration = 39
mu = +1.6389359716880051e-09
relative gap = +2.6102464241977181e-16
gap = +1.4210854715202004e-14
digits = +1.5583318490509461e+01
objValPrimal = -5.4442578997382711e+01
objValDual = -5.4442578997382725e+01
p.feas.error = +5.1269546009545389e-08
d.feas.error = +6.9074010122793794e-09
total time = 1665.490751


◯ 320 スレッド
SDPA start at [Thu May 1 16:28:43 2014]
param is /home2/fujisawa/param.sdpa
data is /home2/fujisawa/data/N.4P.DZ.pqgt1t2p.dat-s : sparse
out is /home2/fujisawa/out.N.4P
NumThreads is set as 320
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
0 1.0e+04 1.0e+00 1.0e+00 -0.00e+00 -1.35e+05 8.6e-01 1.0e+00 2.00e-01
1 2.0e+03 1.4e-01 5.0e-15 +6.47e+00 -2.09e+05 7.8e-01 7.8e-01 2.00e-01
2 6.0e+02 3.1e-02 4.2e-15 -1.18e+01 -2.66e+05 7.5e-01 7.5e-01 2.00e-01

中略

39 1.6e-09 4.2e-10 1.7e-12 -5.44e+01 -5.44e+01 6.1e-01 4.8e-01 1.00e-01

phase.value = pdOPT
Iteration = 39
mu = +1.6371424909425418e-09
relative gap = +1.0440985699065082e-15
gap = -5.6843418860808015e-14
digits = +1.4981258499086902e+01
objValPrimal = -5.4442578985524307e+01
objValDual = -5.4442578985524250e+01
p.feas.error = +5.1265933374066133e-08
d.feas.error = +3.6663733715691649e-09
total time = 1125.381342


○ 160 スレッド
SDPA start at [Wed May 14 17:46:39 2014]
param is /home2/sgitesthpci2/fujisawa/param.sdpa
data is /home2/sgitesthpci2/fujisawa/data/N.4P.DZ.pqgt1t2p.dat-s : sparse
out is /home2/sgitesthpci2/fujisawa/out.N.4P.160
NumThreads is set as 160
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
0 1.0e+04 1.0e+00 1.0e+00 -0.00e+00 -1.35e+05 8.6e-01 1.0e+00 2.00e-01
1 2.0e+03 1.4e-01 3.4e-15 +6.47e+00 -2.09e+05 7.8e-01 7.8e-01 2.00e-01
2 6.0e+02 3.1e-02 5.3e-15 -1.18e+01 -2.66e+05 7.5e-01 7.5e-01 2.00e-01

中略

39 1.6e-09 4.2e-10 2.7e-12 -5.44e+01 -5.44e+01 5.4e-01 5.0e-01 1.00e-01

phase.value = pdOPT
Iteration = 39
mu = +1.6389411021402050e-09
relative gap = +0.0000000000000000e+00
gap = +0.0000000000000000e+00
digits = +inf
objValPrimal = -5.4442578952724780e+01
objValDual = -5.4442578952724780e+01
p.feas.error = +5.1258595481749482e-08
d.feas.error = +5.6941154499834390e-09
total time = 624.571501


○ 80 スレッド
SDPA start at [Wed May 14 18:01:44 2014]
param is /home2/sgitesthpci2/fujisawa/param.sdpa
data is /home2/sgitesthpci2/fujisawa/data/N.4P.DZ.pqgt1t2p.dat-s : sparse
out is /home2/sgitesthpci2/fujisawa/out.N.4P.80
NumThreads is set as 80
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
0 1.0e+04 1.0e+00 1.0e+00 -0.00e+00 -1.35e+05 8.6e-01 1.0e+00 2.00e-01
1 2.0e+03 1.4e-01 3.2e-15 +6.47e+00 -2.09e+05 7.8e-01 7.8e-01 2.00e-01
2 6.0e+02 3.1e-02 5.7e-15 -1.18e+01 -2.66e+05 7.5e-01 7.5e-01 2.00e-01

中略

39 1.6e-09 4.2e-10 2.1e-12 -5.44e+01 -5.44e+01 5.0e-01 5.0e-01 1.00e-01

phase.value = pdOPT
Iteration = 39
mu = +1.6371880571811611e-09
relative gap = +1.3051232124060353e-16
gap = +7.1054273576010019e-15
digits = +1.5884348486071225e+01
objValPrimal = -5.4442578984569010e+01
objValDual = -5.4442578984569018e+01
p.feas.error = +5.1268257492633434e-08
d.feas.error = +4.4527113593484380e-09
total time = 738.721715