| 1 |
NUM_THREADS=8 |
| 2 |
import os |
| 3 |
TEST_STR="timing: per iteration step:" |
| 4 |
REPEAT=10 |
| 5 |
HEADER="""from esys.escript import * |
| 6 |
from esys.finley import Rectangle,Brick |
| 7 |
from esys.escript.linearPDEs import LinearPDE |
| 8 |
SOLVER_TOL=1.e-2 |
| 9 |
REL_TOL=1. |
| 10 |
OPTIMIZE=False |
| 11 |
SOLVER_VERBOSE=True |
| 12 |
FAC_DIAG=1. |
| 13 |
FAC_OFFDIAG=-0.4 |
| 14 |
|
| 15 |
setNumberOfThreads(%d) |
| 16 |
""" |
| 17 |
|
| 18 |
DOM_2_1="dom=Rectangle(NE,NE,order=1, useFullElementOrder=False,optimize=OPTIMIZE)" |
| 19 |
DOM_2_2="dom=Rectangle(NE,NE,order=2, useFullElementOrder=False,optimize=OPTIMIZE)" |
| 20 |
DOM_3_1="dom=Brick(NE,NE,NE,order=1, useFullElementOrder=True,optimize=OPTIMIZE)" |
| 21 |
DOM_3_2="dom=Brick(NE,NE,NE,order=2, useFullElementOrder=True,optimize=OPTIMIZE)" |
| 22 |
|
| 23 |
TEST_2_s="""x=Solution(dom).getX() |
| 24 |
u_ex=Scalar(0,Solution(dom)) |
| 25 |
u_ex=1.+2.*x[0]+3.*x[1] |
| 26 |
g_ex=Data(0.,(2,),Solution(dom)) |
| 27 |
g_ex[0]=2. |
| 28 |
g_ex[1]=3. |
| 29 |
pde=LinearPDE(dom,numEquations=1) |
| 30 |
mask=whereZero(x[0]) |
| 31 |
pde.setValue(r=u_ex,q=mask) |
| 32 |
pde.setValue(A=kronecker(2),y=inner(g_ex,dom.getNormal())) |
| 33 |
""" |
| 34 |
TEST_2_v="""x=Solution(dom).getX() |
| 35 |
x=Solution(dom).getX() |
| 36 |
u_ex=Vector(0,Solution(dom)) |
| 37 |
u_ex[0]=1.+2.*x[0]+3.*x[1] |
| 38 |
u_ex[1]=-1.+3.*x[0]+2.*x[1] |
| 39 |
g_ex=Data(0.,(2,2),Solution(dom)) |
| 40 |
g_ex[0,0]=2. |
| 41 |
g_ex[0,1]=3. |
| 42 |
g_ex[1,0]=3. |
| 43 |
g_ex[1,1]=2. |
| 44 |
pde=LinearPDE(dom,numEquations=2) |
| 45 |
mask=whereZero(x[0]) |
| 46 |
pde.setValue(r=u_ex,q=mask*numarray.ones(2,)) |
| 47 |
A=Tensor4(0,Function(dom)) |
| 48 |
A[0,:,0,:]=kronecker(2) |
| 49 |
A[1,:,1,:]=kronecker(2) |
| 50 |
Y=Vector(0.,Function(dom)) |
| 51 |
Y[0]=u_ex[0]*FAC_DIAG+u_ex[1]*FAC_OFFDIAG |
| 52 |
Y[1]=u_ex[1]*FAC_DIAG+u_ex[0]*FAC_OFFDIAG |
| 53 |
pde.setValue(A=A, D=kronecker(2)*(FAC_DIAG-FAC_OFFDIAG)+numarray.ones((2,2))*FAC_OFFDIAG, Y=Y, y=matrixmult(g_ex,dom.getNormal())) |
| 54 |
""" |
| 55 |
|
| 56 |
TEST_3_s="""x=Solution(dom).getX() |
| 57 |
u_ex=1.+2.*x[0]+3.*x[1]+4.*x[2] |
| 58 |
g_ex=Data(0.,(3,),Solution(dom)) |
| 59 |
g_ex[0]=2. |
| 60 |
g_ex[1]=3. |
| 61 |
g_ex[2]=4. |
| 62 |
pde=LinearPDE(dom,numEquations=1) |
| 63 |
mask=whereZero(x[0]) |
| 64 |
pde.setValue(r=u_ex,q=mask) |
| 65 |
pde.setValue(A=kronecker(3),y=inner(g_ex,dom.getNormal())) |
| 66 |
""" |
| 67 |
|
| 68 |
TEST_3_v="""x=Solution(dom).getX() |
| 69 |
u_ex=Vector(0,Solution(dom)) |
| 70 |
u_ex[0]=1.+2.*x[0]+3.*x[1]+4.*x[2] |
| 71 |
u_ex[1]=-1.+4.*x[0]+1.*x[1]-2.*x[2] |
| 72 |
u_ex[2]=5.+8.*x[0]+4.*x[1]+5.*x[2] |
| 73 |
g_ex=Data(0.,(3,3),Solution(dom)) |
| 74 |
g_ex[0,0]=2. |
| 75 |
g_ex[0,1]=3. |
| 76 |
g_ex[0,2]=4. |
| 77 |
g_ex[1,0]=4. |
| 78 |
g_ex[1,1]=1. |
| 79 |
g_ex[1,2]=-2. |
| 80 |
g_ex[2,0]=8. |
| 81 |
g_ex[2,1]=4. |
| 82 |
g_ex[2,2]=5. |
| 83 |
pde=LinearPDE(dom,numEquations=3) |
| 84 |
mask=whereZero(x[0]) |
| 85 |
pde.setValue(r=u_ex,q=mask*numarray.ones(3,)) |
| 86 |
A=Tensor4(0,Function(dom)) |
| 87 |
A[0,:,0,:]=kronecker(3) |
| 88 |
A[1,:,1,:]=kronecker(3) |
| 89 |
A[2,:,2,:]=kronecker(3) |
| 90 |
Y=Vector(0.,Function(dom)) |
| 91 |
Y[0]=u_ex[0]*FAC_DIAG+u_ex[2]*FAC_OFFDIAG+u_ex[1]*FAC_OFFDIAG |
| 92 |
Y[1]=u_ex[1]*FAC_DIAG+u_ex[0]*FAC_OFFDIAG+u_ex[2]*FAC_OFFDIAG |
| 93 |
Y[2]=u_ex[2]*FAC_DIAG+u_ex[1]*FAC_OFFDIAG+u_ex[0]*FAC_OFFDIAG |
| 94 |
pde.setValue(A=A, |
| 95 |
D=kronecker(3)*(FAC_DIAG-FAC_OFFDIAG)+numarray.ones((3,3))*FAC_OFFDIAG, |
| 96 |
Y=Y, |
| 97 |
y=matrixmult(g_ex,dom.getNormal())) |
| 98 |
""" |
| 99 |
|
| 100 |
SOLVE_AND_TEST="""pde.setTolerance(SOLVER_TOL) |
| 101 |
pde.setSolverMethod(pde.PCG,pde.JACOBI) |
| 102 |
pde.setSolverPackage(pde.PASO) |
| 103 |
u=pde.getSolution(verbose=SOLVER_VERBOSE) |
| 104 |
error=Lsup(u-u_ex)/Lsup(u_ex) |
| 105 |
if error>REL_TOL*Lsup(u_ex): raise RuntimeError("solution error %s is too big."%error) |
| 106 |
""" |
| 107 |
|
| 108 |
|
| 109 |
#for n in [10000, 50000, 100000]: |
| 110 |
for n in [100000]: |
| 111 |
# for n in [1000, 10000]: |
| 112 |
#for prop in [ (1,2), (2,2), (1,3), (2,3) ]: |
| 113 |
for prop in [ (1,2), (1,3) ]: |
| 114 |
for tp in [ "s", "v" ]: |
| 115 |
# create code: |
| 116 |
prog=HEADER%NUM_THREADS |
| 117 |
dim=prop[1] |
| 118 |
if isinstance(prop[0], int): |
| 119 |
o=prop[0] |
| 120 |
if tp=="s": |
| 121 |
q=1 |
| 122 |
else: |
| 123 |
q=dim |
| 124 |
NE=int(float(n/q-1)**(1./dim)/o+0.5) |
| 125 |
prog+="NE=%d\n"%NE |
| 126 |
if dim==2: |
| 127 |
if o==1: |
| 128 |
prog+=DOM_2_1 |
| 129 |
else: |
| 130 |
prog+=DOM_2_2 |
| 131 |
else: |
| 132 |
if o==1: |
| 133 |
prog+=DOM_3_1 |
| 134 |
else: |
| 135 |
prog+=DOM_3_2 |
| 136 |
prog+="\n" |
| 137 |
if dim==2: |
| 138 |
if tp =="s": |
| 139 |
prog+=TEST_2_s |
| 140 |
else: |
| 141 |
prog+=TEST_2_v |
| 142 |
else: |
| 143 |
if tp =="s": |
| 144 |
prog+=TEST_3_s |
| 145 |
else: |
| 146 |
prog+=TEST_3_v |
| 147 |
print "l= %d, dim= %d, type=%s, order=%s"%(q*(o*NE+1)**dim,dim,tp,o) |
| 148 |
|
| 149 |
prog+=SOLVE_AND_TEST |
| 150 |
# run code: |
| 151 |
print >> file("__prog","w"), prog |
| 152 |
# activate for dynamic |
| 153 |
# for CHUNK in [1,10,100,1000,10000, 100000]: |
| 154 |
# for CHUNK_PCG in [1,10,100,1000,10000, 100000]: |
| 155 |
# activate for static |
| 156 |
for CHUNK in [-1]: |
| 157 |
for CHUNK_PCG in [-1]: |
| 158 |
if CHUNK*NUM_THREADS <= n and CHUNK_PCG*NUM_THREADS <=n: |
| 159 |
time_per_iter=0 |
| 160 |
for i in range(REPEAT): |
| 161 |
os.system("export OMP_NUM_THREADS=%d;export PASO_CHUNK_SIZE_MVM=%d; export PASO_CHUNK_SIZE_PCG=%d; python __prog > __out;"%(NUM_THREADS,CHUNK,CHUNK_PCG)) |
| 162 |
out=file("__out","r").read() |
| 163 |
for i in out.split("\n"): |
| 164 |
if i.startswith(TEST_STR): time_per_iter+=float(i[len(TEST_STR):-3].strip()) |
| 165 |
print CHUNK,CHUNK_PCG,time_per_iter/REPEAT |
| 166 |
|
Parent Directory
| 
Revision Log