func main() {
// catch errors
var tst testing.T
defer func() {
if mpi.Rank() == 0 {
if err := recover(); err != nil {
io.PfRed("ERROR: %v\n", err)
}
if tst.Failed() {
io.PfRed("test failed\n")
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// start global variables and log
analysis := fem.NewFEM("data/bh16.sim", "", true, true, false, true, true, 0)
// run simulation
err := analysis.Run()
if err != nil {
tst.Error("Run failed\n")
return
}
// check
skipK := true
tolK := 1e-12
tolu := 1e-15
tols := 1e-12
fem.TestingCompareResultsU(&tst, "data/bh16.sim", "cmp/bh16.cmp", "", tolK, tolu, tols, skipK, true)
}
func main() {
// catch errors
var tst testing.T
defer func() {
if mpi.Rank() == 0 {
if err := recover(); err != nil {
io.PfRed("ERROR: %v\n", err)
}
if tst.Failed() {
io.PfRed("test failed\n")
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// start global variables and log
analysis := fem.NewFEM("data/p01.sim", "", true, true, false, true, true, 0)
// run simulation
err := analysis.Run()
if err != nil {
tst.Error("Run failed\n")
return
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
io.PfYel("\nTest MPI 03\n")
}
if mpi.Size() != 3 {
chk.Panic("this test needs 3 processors")
}
x := []int{-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1}
n := len(x)
id, sz := mpi.Rank(), mpi.Size()
start, endp1 := (id*n)/sz, ((id+1)*n)/sz
for i := start; i < endp1; i++ {
x[i] = i
}
//io.Pforan("x = %v\n", x)
// IntAllReduceMax
w := make([]int, n)
mpi.IntAllReduceMax(x, w)
var tst testing.T
chk.Ints(&tst, fmt.Sprintf("IntAllReduceMax: x @ proc # %d", id), x, []int{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10})
//io.Pfred("x = %v\n", x)
}
func main() {
// catch errors
var tst testing.T
defer func() {
if mpi.Rank() == 0 {
if err := recover(); err != nil {
io.PfRed("ERROR: %v\n", err)
}
if tst.Failed() {
io.PfRed("test failed\n")
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// start global variables and log
if !fem.Start("data/p01.sim", true, true) {
tst.Error("Start failed\n")
return
}
// make sure to flush log
defer fem.End()
// run simulation
if !fem.Run() {
tst.Error("Run failed\n")
return
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
myrank := mpi.Rank()
if myrank == 0 {
chk.PrintTitle("Test MUMPS Sol 01a")
}
var t la.Triplet
switch mpi.Size() {
case 1:
t.Init(5, 5, 13)
t.Put(0, 0, 1.0)
t.Put(0, 0, 1.0)
t.Put(1, 0, 3.0)
t.Put(0, 1, 3.0)
t.Put(2, 1, -1.0)
t.Put(4, 1, 4.0)
t.Put(1, 2, 4.0)
t.Put(2, 2, -3.0)
t.Put(3, 2, 1.0)
t.Put(4, 2, 2.0)
t.Put(2, 3, 2.0)
t.Put(1, 4, 6.0)
t.Put(4, 4, 1.0)
case 2:
if myrank == 0 {
t.Init(5, 5, 6)
t.Put(0, 0, 1.0)
t.Put(0, 0, 1.0)
t.Put(1, 0, 3.0)
t.Put(0, 1, 3.0)
t.Put(2, 1, -1.0)
t.Put(4, 1, 4.0)
} else {
t.Init(5, 5, 7)
t.Put(1, 2, 4.0)
t.Put(2, 2, -3.0)
t.Put(3, 2, 1.0)
t.Put(4, 2, 2.0)
t.Put(2, 3, 2.0)
t.Put(1, 4, 6.0)
t.Put(4, 4, 1.0)
}
default:
chk.Panic("this test needs 1 or 2 procs")
}
b := []float64{8.0, 45.0, -3.0, 3.0, 19.0}
x_correct := []float64{1, 2, 3, 4, 5}
sum_b_to_root := false
la.RunMumpsTestR(&t, 1e-14, b, x_correct, sum_b_to_root)
}
func main() {
// catch errors
defer func() {
if err := recover(); err != nil {
if mpi.Rank() == 0 {
chk.Verbose = true
for i := 8; i > 3; i-- {
chk.CallerInfo(i)
}
io.PfRed("ERROR: %v\n", err)
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// default input parameters
// read input parameters
fnamepath, _ := io.ArgToFilename(0, "", ".sim", true)
verbose := io.ArgToBool(1, true)
erasePrev := io.ArgToBool(2, true)
saveSummary := io.ArgToBool(3, true)
allowParallel := io.ArgToBool(4, true)
alias := io.ArgToString(5, "")
// message
if mpi.Rank() == 0 && verbose {
io.PfWhite("\nGofem v3 -- Go Finite Element Method\n\n")
io.Pf("Copyright 2015 Dorival Pedroso and Raul Durand. All rights reserved.\n")
io.Pf("Use of this source code is governed by a BSD-style\n")
io.Pf("license that can be found in the LICENSE file.\n\n")
io.Pf("\n%v\n", io.ArgsTable(
"filename path", "fnamepath", fnamepath,
"show messages", "verbose", verbose,
"erase previous results", "erasePrev", erasePrev,
"save summary", "saveSummary", saveSummary,
"allow parallel run", "allowParallel", allowParallel,
"word to add to results", "alias", alias,
))
}
// profiling?
defer utl.DoProf(false)()
// analysis data
readSummary := false
analysis := fem.NewFEM(fnamepath, alias, erasePrev, saveSummary, readSummary, allowParallel, verbose, 0)
// run simulation
err := analysis.Run()
if err != nil {
chk.Panic("Run failed:\n%v", err)
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("Test SumToRoot 01")
}
M := [][]float64{
{1000, 1000, 1000, 1011, 1021, 1000},
{1000, 1000, 1000, 1012, 1022, 1000},
{1000, 1000, 1000, 1013, 1023, 1000},
{1011, 1012, 1013, 1000, 1000, 1000},
{1021, 1022, 1023, 1000, 1000, 1000},
{1000, 1000, 1000, 1000, 1000, 1000},
}
id, sz, m := mpi.Rank(), mpi.Size(), len(M)
start, endp1 := (id*m)/sz, ((id+1)*m)/sz
if sz > 6 {
chk.Panic("this test works with at most 6 processors")
}
var J la.Triplet
J.Init(m, m, m*m)
for i := start; i < endp1; i++ {
for j := 0; j < m; j++ {
J.Put(i, j, M[i][j])
}
}
la.PrintMat(fmt.Sprintf("J @ proc # %d", id), J.ToMatrix(nil).ToDense(), "%10.1f", false)
la.SpTriSumToRoot(&J)
var tst testing.T
if mpi.Rank() == 0 {
chk.Matrix(&tst, "J @ proc 0", 1.0e-17, J.ToMatrix(nil).ToDense(), [][]float64{
{1000, 1000, 1000, 1011, 1021, 1000},
{1000, 1000, 1000, 1012, 1022, 1000},
{1000, 1000, 1000, 1013, 1023, 1000},
{1011, 1012, 1013, 1000, 1000, 1000},
{1021, 1022, 1023, 1000, 1000, 1000},
{1000, 1000, 1000, 1000, 1000, 1000},
})
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("TestJacobian 02b (MPI)")
}
if mpi.Size() > 6 {
io.Pf("this tests works with 6 or less MPI processors\n")
return
}
ffcn := func(fx, x []float64) error {
fx[0] = 2.0*x[0] - x[1] + sin(x[2]) - cos(x[3]) - x[5]*x[5] - 1.0 // 0
fx[1] = -x[0] + 2.0*x[1] + cos(x[2]) - sin(x[3]) + x[5] - 1.0 // 1
fx[2] = x[0] + 3.0*x[1] + sin(x[3]) - cos(x[4]) - x[5]*x[5] - 1.0 // 2
fx[3] = 2.0*x[0] + 4.0*x[1] + cos(x[3]) - cos(x[4]) + x[5] - 1.0 // 3
fx[4] = x[0] + 5.0*x[1] - sin(x[2]) + sin(x[4]) - x[5]*x[5]*x[5] - 1.0 // 4
fx[5] = x[0] + 6.0*x[1] - cos(x[2]) + cos(x[4]) + x[5] - 1.0 // 5
return nil
}
Jfcn := func(dfdx *la.Triplet, x []float64) error {
dfdx.Start()
J := [][]float64{
{2.0, -1.0, cos(x[2]), sin(x[3]), 0.0, -2.0 * x[5]},
{-1.0, 2.0, -sin(x[2]), -cos(x[3]), 0.0, 1.0},
{1.0, 3.0, 0.0, cos(x[3]), sin(x[4]), -2.0 * x[5]},
{2.0, 4.0, 0.0, -sin(x[3]), sin(x[4]), 1.0},
{1.0, 5.0, -cos(x[2]), 0.0, cos(x[4]), -3.0 * x[5] * x[5]},
{1.0, 6.0, sin(x[2]), 0.0, -sin(x[4]), 1.0},
}
id, sz, ndim := mpi.Rank(), mpi.Size(), 6
start, endp1 := (id*ndim)/sz, ((id+1)*ndim)/sz
for col := 0; col < 6; col++ {
for row := start; row < endp1; row++ {
dfdx.Put(row, col, J[row][col])
}
}
//la.PrintMat(fmt.Sprintf("J @ %d",mpi.Rank()), dfdx.ToMatrix(nil).ToDense(), "%12.6f", false)
return nil
}
x := []float64{5.0, 5.0, pi, pi, pi, 5.0}
var tst testing.T
num.CompareJac(&tst, ffcn, Jfcn, x, 1e-6, true)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
myrank := mpi.Rank()
if myrank == 0 {
chk.PrintTitle("Test MUMPS Sol 05")
}
ndim := 10
id, sz := mpi.Rank(), mpi.Size()
start, endp1 := (id*ndim)/sz, ((id+1)*ndim)/sz
if mpi.Size() > ndim {
chk.Panic("the number of processors must be smaller than or equal to %d", ndim)
}
n := 10
b := make([]complex128, n)
x_correct := make([]complex128, n)
// Let exact solution = 1 + 0.5i
for i := 0; i < ndim; i++ {
x_correct[i] = complex(float64(i+1), float64(i+1)/10.0)
}
var t la.TripletC
t.Init(ndim, ndim, ndim, true)
// assemble a and b
for i := start; i < endp1; i++ {
// Some very fake diagonals. Should take exactly 20 GMRES steps
ar := 10.0 + float64(i)/(float64(ndim)/10.0)
ac := 10.0 - float64(i)/(float64(ndim)/10.0)
t.Put(i, i, ar, ac)
// Generate RHS to match exact solution
b[i] = complex(ar*real(x_correct[i])-ac*imag(x_correct[i]),
ar*imag(x_correct[i])+ac*real(x_correct[i]))
}
sum_b_to_root := true
la.RunMumpsTestC(&t, 1e-14, b, x_correct, sum_b_to_root)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("TestJacobian 01b (MPI)")
}
if mpi.Size() != 2 {
io.Pf("this tests needs MPI 2 processors\n")
return
}
ffcn := func(fx, x []float64) error {
fx[0] = math.Pow(x[0], 3.0) + x[1] - 1.0
fx[1] = -x[0] + math.Pow(x[1], 3.0) + 1.0
return nil
}
Jfcn := func(dfdx *la.Triplet, x []float64) error {
dfdx.Start()
if false {
if mpi.Rank() == 0 {
dfdx.Put(0, 0, 3.0*x[0]*x[0])
dfdx.Put(1, 0, -1.0)
} else {
dfdx.Put(0, 1, 1.0)
dfdx.Put(1, 1, 3.0*x[1]*x[1])
}
} else {
if mpi.Rank() == 0 {
dfdx.Put(0, 0, 3.0*x[0]*x[0])
dfdx.Put(0, 1, 1.0)
} else {
dfdx.Put(1, 0, -1.0)
dfdx.Put(1, 1, 3.0*x[1]*x[1])
}
}
return nil
}
x := []float64{0.5, 0.5}
var tst testing.T
num.CompareJacMpi(&tst, ffcn, Jfcn, x, 1e-8, true)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
myrank := mpi.Rank()
if myrank == 0 {
chk.PrintTitle("Test MUMPS Sol 04")
}
ndim := 10
id, sz := mpi.Rank(), mpi.Size()
start, endp1 := (id*ndim)/sz, ((id+1)*ndim)/sz
if mpi.Size() > ndim {
chk.Panic("the number of processors must be smaller than or equal to %d", ndim)
}
b := make([]complex128, ndim)
var t la.TripletC
t.Init(ndim, ndim, ndim*ndim, true)
for i := start; i < endp1; i++ {
j := i
if i > 0 {
j = i - 1
}
for ; j < 10; j++ {
val := 10.0 - float64(j)
if i > j {
val -= 1.0
}
t.Put(i, j, val, 0)
}
b[i] = complex(float64(i+1), 0.0)
}
x_correct := []complex128{-1, 8, -65, 454, -2725, 13624, -54497, 163490, -326981, 326991}
sum_b_to_root := true
la.RunMumpsTestC(&t, 1e-4, b, x_correct, sum_b_to_root)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
io.PfYel("\nTest MPI 04\n")
}
for i := 0; i < 60; i++ {
time.Sleep(1e9)
io.Pf("hello from %v\n", mpi.Rank())
if mpi.Rank() == 2 && i == 3 {
io.PfGreen("rank = 3 wants to abort (the following error is OK)\n")
mpi.Abort()
}
}
}
func main() {
// catch errors
var tst testing.T
defer func() {
if mpi.Rank() == 0 {
if err := recover(); err != nil {
io.PfRed("ERROR: %v\n", err)
}
if tst.Failed() {
io.PfRed("test failed\n")
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// start global variables and log
if !fem.Start("data/spo751.sim", true, true) {
tst.Error("Start failed\n")
return
}
// make sure to flush log
defer fem.End()
// run simulation
if !fem.Run() {
tst.Error("Run failed\n")
return
}
// check
skipK := true
tolK := 1e-17
tolu := 1e-12
tols := 1e-14
fem.TestingCompareResultsU(&tst, "data/spo751.sim", "cmp/spo751.cmp", tolK, tolu, tols, skipK, true)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("ode02: Hairer-Wanner VII-p5 Eq.(1.5) Van der Pol's Equation")
}
if mpi.Size() != 2 {
chk.Panic(">> error: this test requires 2 MPI processors\n")
return
}
eps := 1.0e-6
w := make([]float64, 2) // workspace
fcn := func(f []float64, dx, x float64, y []float64, args ...interface{}) error {
f[0], f[1] = 0, 0
switch mpi.Rank() {
case 0:
f[0] = y[1]
case 1:
f[1] = ((1.0-y[0]*y[0])*y[1] - y[0]) / eps
}
// join all f
mpi.AllReduceSum(f, w)
return nil
}
jac := func(dfdy *la.Triplet, dx, x float64, y []float64, args ...interface{}) error {
if dfdy.Max() == 0 {
dfdy.Init(2, 2, 4)
}
dfdy.Start()
if false { // per column
switch mpi.Rank() {
case 0:
dfdy.Put(0, 0, 0.0)
dfdy.Put(1, 0, (-2.0*y[0]*y[1]-1.0)/eps)
case 1:
dfdy.Put(0, 1, 1.0)
dfdy.Put(1, 1, (1.0-y[0]*y[0])/eps)
}
} else { // per row
switch mpi.Rank() {
case 0:
dfdy.Put(0, 0, 0.0)
dfdy.Put(0, 1, 1.0)
case 1:
dfdy.Put(1, 0, (-2.0*y[0]*y[1]-1.0)/eps)
dfdy.Put(1, 1, (1.0-y[0]*y[0])/eps)
}
}
return nil
}
// method and flags
silent := false
fixstp := false
//method := "Dopri5"
method := "Radau5"
numjac := false
xa, xb := 0.0, 2.0
ya := []float64{2.0, -0.6}
ndim := len(ya)
// structure to hold numerical results
res := ode.Results{Method: method}
// allocate ODE object
var o ode.Solver
o.Distr = true
if numjac {
o.Init(method, ndim, fcn, nil, nil, ode.SimpleOutput, silent)
} else {
o.Init(method, ndim, fcn, jac, nil, ode.SimpleOutput, silent)
}
// tolerances and initial step size
rtol := 1e-4
atol := rtol
o.IniH = 1.0e-4
o.SetTol(atol, rtol)
//o.NmaxSS = 2
// solve problem
y := make([]float64, ndim)
copy(y, ya)
t0 := time.Now()
if fixstp {
o.Solve(y, xa, xb, 0.05, fixstp, &res)
} else {
o.Solve(y, xa, xb, xb-xa, fixstp, &res)
}
// plot
if mpi.Rank() == 0 {
io.Pfmag("elapsed time = %v\n", time.Now().Sub(t0))
plt.SetForEps(1.5, 400)
args := "'b-', marker='.', lw=1, ms=4, clip_on=0"
ode.Plot("/tmp/gosl/ode", "vdpolA_mpi.eps", &res, nil, xa, xb, "", args, func() {
_, T, err := io.ReadTable("data/vdpol_radau5_for.dat")
if err != nil {
chk.Panic("%v", err)
}
plt.Subplot(3, 1, 1)
plt.Plot(T["x"], T["y0"], "'k+',label='reference',ms=7")
plt.Subplot(3, 1, 2)
plt.Plot(T["x"], T["y1"], "'k+',label='reference',ms=7")
})
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("Test ODE 02b")
io.Pfcyan("Hairer-Wanner VII-p5 Eq.(1.5) Van der Pol's Equation (MPI)\n")
}
if mpi.Size() != 2 {
chk.Panic(">> error: this test requires 2 MPI processors\n")
return
}
eps := 1.0e-6
w := make([]float64, 2) // workspace
fcn := func(f []float64, x float64, y []float64, args ...interface{}) error {
f[0], f[1] = 0, 0
switch mpi.Rank() {
case 0:
f[0] = y[1]
case 1:
f[1] = ((1.0-y[0]*y[0])*y[1] - y[0]) / eps
}
// join all f
mpi.AllReduceSum(f, w)
return nil
}
jac := func(dfdy *la.Triplet, x float64, y []float64, args ...interface{}) error {
if dfdy.Max() == 0 {
dfdy.Init(2, 2, 4)
}
dfdy.Start()
if false { // per column
switch mpi.Rank() {
case 0:
dfdy.Put(0, 0, 0.0)
dfdy.Put(1, 0, (-2.0*y[0]*y[1]-1.0)/eps)
case 1:
dfdy.Put(0, 1, 1.0)
dfdy.Put(1, 1, (1.0-y[0]*y[0])/eps)
}
} else { // per row
switch mpi.Rank() {
case 0:
dfdy.Put(0, 0, 0.0)
dfdy.Put(0, 1, 1.0)
case 1:
dfdy.Put(1, 0, (-2.0*y[0]*y[1]-1.0)/eps)
dfdy.Put(1, 1, (1.0-y[0]*y[0])/eps)
}
}
return nil
}
// data
silent := false
fixstp := false
//method := "Dopri5"
method := "Radau5"
xa, xb := 0.0, 2.0
ya := []float64{2.0, -0.6}
ndim := len(ya)
// output
var b bytes.Buffer
out := func(first bool, dx, x float64, y []float64, args ...interface{}) error {
if mpi.Rank() == 0 {
if first {
fmt.Fprintf(&b, "%23s %23s %23s %23s\n", "dx", "x", "y0", "y1")
}
fmt.Fprintf(&b, "%23.15E %23.15E %23.15E %23.15E\n", dx, x, y[0], y[1])
}
return nil
}
defer func() {
if mpi.Rank() == 0 {
extra := "d2 = Read('data/vdpol_radau5_for.dat')\n" +
"subplot(3,1,1)\n" +
"plot(d2['x'],d2['y0'],'k+',label='res',ms=10)\n" +
"subplot(3,1,2)\n" +
"plot(d2['x'],d2['y1'],'k+',label='res',ms=10)\n"
ode.Plot("/tmp/gosl", "vdpolB", method, &b, []int{0, 1}, ndim, nil, xa, xb, true, false, extra)
}
}()
// one run
var o ode.ODE
o.Distr = true
//numjac := true
numjac := false
if numjac {
o.Init(method, ndim, fcn, nil, nil, out, silent)
} else {
o.Init(method, ndim, fcn, jac, nil, out, silent)
}
// tolerances and initial step size
rtol := 1e-4
atol := rtol
o.SetTol(atol, rtol)
o.IniH = 1.0e-4
//o.NmaxSS = 2
y := make([]float64, ndim)
copy(y, ya)
t0 := time.Now()
if fixstp {
o.Solve(y, xa, xb, 0.05, fixstp)
} else {
o.Solve(y, xa, xb, xb-xa, fixstp)
}
if mpi.Rank() == 0 {
io.Pfmag("elapsed time = %v\n", time.Now().Sub(t0))
}
}
func main() {
// catch errors
defer func() {
if err := recover(); err != nil {
if mpi.Rank() == 0 {
chk.Verbose = true
for i := 8; i > 3; i-- {
chk.CallerInfo(i)
}
io.PfRed("ERROR: %v\n", err)
}
}
mpi.Stop(false)
}()
mpi.Start(false)
// message
if mpi.Rank() == 0 {
io.PfWhite("\nGofem v3 -- Go Finite Element Method\n\n")
io.Pf("Copyright 2015 Dorival Pedroso and Raul Durand. All rights reserved.\n")
io.Pf("Use of this source code is governed by a BSD-style\n")
io.Pf("license that can be found in the LICENSE file.\n\n")
}
// simulation filenamepath
flag.Parse()
var fnamepath string
if len(flag.Args()) > 0 {
fnamepath = flag.Arg(0)
} else {
chk.Panic("Please, provide a filename. Ex.: cylinder.sim")
}
// check extension
if io.FnExt(fnamepath) == "" {
fnamepath += ".sim"
}
// other options
erasefiles := true
verbose := true
if len(flag.Args()) > 1 {
erasefiles = io.Atob(flag.Arg(1))
}
if len(flag.Args()) > 2 {
verbose = io.Atob(flag.Arg(2))
}
// profiling?
defer utl.DoProf(false)()
// start global variables and log
if !fem.Start(fnamepath, erasefiles, verbose) {
chk.Panic("Start failed\n")
return
}
// make sure to flush log
defer fem.End()
// run simulation
if !fem.Run() {
io.PfRed("ERROR: cannot run simulation\n")
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("Test ODE 04b (MPI)")
io.Pfcyan("Hairer-Wanner VII-p376 Transistor Amplifier (MPI)\n")
io.Pfcyan("(from E Hairer's website, not the system in the book)\n")
}
if mpi.Size() != 3 {
chk.Panic(">> error: this test requires 3 MPI processors\n")
return
}
// RIGHT-HAND SIDE OF THE AMPLIFIER PROBLEM
w := make([]float64, 8) // workspace
fcn := func(f []float64, x float64, y []float64, args ...interface{}) error {
d := args[0].(*HWtransData)
UET := d.UE * math.Sin(d.W*x)
FAC1 := d.BETA * (math.Exp((y[3]-y[2])/d.UF) - 1.0)
FAC2 := d.BETA * (math.Exp((y[6]-y[5])/d.UF) - 1.0)
la.VecFill(f, 0)
switch mpi.Rank() {
case 0:
f[0] = y[0] / d.R9
case 1:
f[1] = (y[1]-d.UB)/d.R8 + d.ALPHA*FAC1
f[2] = y[2]/d.R7 - FAC1
case 2:
f[3] = y[3]/d.R5 + (y[3]-d.UB)/d.R6 + (1.0-d.ALPHA)*FAC1
f[4] = (y[4]-d.UB)/d.R4 + d.ALPHA*FAC2
f[5] = y[5]/d.R3 - FAC2
f[6] = y[6]/d.R1 + (y[6]-d.UB)/d.R2 + (1.0-d.ALPHA)*FAC2
f[7] = (y[7] - UET) / d.R0
}
mpi.AllReduceSum(f, w)
return nil
}
// JACOBIAN OF THE AMPLIFIER PROBLEM
jac := func(dfdy *la.Triplet, x float64, y []float64, args ...interface{}) error {
d := args[0].(*HWtransData)
FAC14 := d.BETA * math.Exp((y[3]-y[2])/d.UF) / d.UF
FAC27 := d.BETA * math.Exp((y[6]-y[5])/d.UF) / d.UF
if dfdy.Max() == 0 {
dfdy.Init(8, 8, 16)
}
NU := 2
dfdy.Start()
switch mpi.Rank() {
case 0:
dfdy.Put(2+0-NU, 0, 1.0/d.R9)
dfdy.Put(2+1-NU, 1, 1.0/d.R8)
dfdy.Put(1+2-NU, 2, -d.ALPHA*FAC14)
dfdy.Put(0+3-NU, 3, d.ALPHA*FAC14)
dfdy.Put(2+2-NU, 2, 1.0/d.R7+FAC14)
case 1:
dfdy.Put(1+3-NU, 3, -FAC14)
dfdy.Put(2+3-NU, 3, 1.0/d.R5+1.0/d.R6+(1.0-d.ALPHA)*FAC14)
dfdy.Put(3+2-NU, 2, -(1.0-d.ALPHA)*FAC14)
dfdy.Put(2+4-NU, 4, 1.0/d.R4)
dfdy.Put(1+5-NU, 5, -d.ALPHA*FAC27)
case 2:
dfdy.Put(0+6-NU, 6, d.ALPHA*FAC27)
dfdy.Put(2+5-NU, 5, 1.0/d.R3+FAC27)
dfdy.Put(1+6-NU, 6, -FAC27)
dfdy.Put(2+6-NU, 6, 1.0/d.R1+1.0/d.R2+(1.0-d.ALPHA)*FAC27)
dfdy.Put(3+5-NU, 5, -(1.0-d.ALPHA)*FAC27)
dfdy.Put(2+7-NU, 7, 1.0/d.R0)
}
return nil
}
// MATRIX "M"
c1, c2, c3, c4, c5 := 1.0e-6, 2.0e-6, 3.0e-6, 4.0e-6, 5.0e-6
var M la.Triplet
M.Init(8, 8, 14)
M.Start()
NU := 1
switch mpi.Rank() {
case 0:
M.Put(1+0-NU, 0, -c5)
M.Put(0+1-NU, 1, c5)
M.Put(2+0-NU, 0, c5)
M.Put(1+1-NU, 1, -c5)
M.Put(1+2-NU, 2, -c4)
M.Put(1+3-NU, 3, -c3)
case 1:
M.Put(0+4-NU, 4, c3)
M.Put(2+3-NU, 3, c3)
M.Put(1+4-NU, 4, -c3)
case 2:
M.Put(1+5-NU, 5, -c2)
M.Put(1+6-NU, 6, -c1)
M.Put(0+7-NU, 7, c1)
M.Put(2+6-NU, 6, c1)
M.Put(1+7-NU, 7, -c1)
}
// WRITE FILE FUNCTION
idxstp := 1
var b bytes.Buffer
out := func(first bool, dx, x float64, y []float64, args ...interface{}) error {
if mpi.Rank() == 0 {
if first {
fmt.Fprintf(&b, "%6s%23s%23s%23s%23s%23s%23s%23s%23s%23s\n", "ns", "x", "y0", "y1", "y2", "y3", "y4", "y5", "y6", "y7")
}
fmt.Fprintf(&b, "%6d%23.15E", idxstp, x)
for j := 0; j < len(y); j++ {
fmt.Fprintf(&b, "%23.15E", y[j])
}
fmt.Fprintf(&b, "\n")
idxstp += 1
}
return nil
}
defer func() {
if mpi.Rank() == 0 {
io.WriteFileD("/tmp/gosl", "hwamplifierB.res", &b)
}
}()
// INITIAL DATA
D, xa, xb, ya := HWtransIni()
// SET ODE SOLVER
silent := false
fixstp := false
//method := "Dopri5"
method := "Radau5"
ndim := len(ya)
//numjac := true
numjac := false
var osol ode.ODE
osol.Pll = true
if numjac {
osol.Init(method, ndim, fcn, nil, &M, out, silent)
} else {
osol.Init(method, ndim, fcn, jac, &M, out, silent)
}
osol.IniH = 1.0e-6 // initial step size
// SET TOLERANCES
atol, rtol := 1e-11, 1e-5
osol.SetTol(atol, rtol)
// RUN
t0 := time.Now()
if fixstp {
osol.Solve(ya, xa, xb, 0.01, fixstp, &D)
} else {
osol.Solve(ya, xa, xb, xb-xa, fixstp, &D)
}
if mpi.Rank() == 0 {
io.Pfmag("elapsed time = %v\n", time.Now().Sub(t0))
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
chk.PrintTitle("ode04: Hairer-Wanner VII-p376 Transistor Amplifier\n")
}
if mpi.Size() != 3 {
chk.Panic(">> error: this test requires 3 MPI processors\n")
return
}
// data
UE, UB, UF, ALPHA, BETA := 0.1, 6.0, 0.026, 0.99, 1.0e-6
R0, R1, R2, R3, R4, R5 := 1000.0, 9000.0, 9000.0, 9000.0, 9000.0, 9000.0
R6, R7, R8, R9 := 9000.0, 9000.0, 9000.0, 9000.0
W := 2.0 * 3.141592654 * 100.0
// initial values
xa := 0.0
ya := []float64{0.0,
UB,
UB / (R6/R5 + 1.0),
UB / (R6/R5 + 1.0),
UB,
UB / (R2/R1 + 1.0),
UB / (R2/R1 + 1.0),
0.0}
// endpoint of integration
xb := 0.05
//xb = 0.0123 // OK
//xb = 0.01235 // !OK
// right-hand side of the amplifier problem
w := make([]float64, 8) // workspace
fcn := func(f []float64, dx, x float64, y []float64, args ...interface{}) error {
UET := UE * math.Sin(W*x)
FAC1 := BETA * (math.Exp((y[3]-y[2])/UF) - 1.0)
FAC2 := BETA * (math.Exp((y[6]-y[5])/UF) - 1.0)
la.VecFill(f, 0)
switch mpi.Rank() {
case 0:
f[0] = y[0] / R9
case 1:
f[1] = (y[1]-UB)/R8 + ALPHA*FAC1
f[2] = y[2]/R7 - FAC1
case 2:
f[3] = y[3]/R5 + (y[3]-UB)/R6 + (1.0-ALPHA)*FAC1
f[4] = (y[4]-UB)/R4 + ALPHA*FAC2
f[5] = y[5]/R3 - FAC2
f[6] = y[6]/R1 + (y[6]-UB)/R2 + (1.0-ALPHA)*FAC2
f[7] = (y[7] - UET) / R0
}
mpi.AllReduceSum(f, w)
return nil
}
// Jacobian of the amplifier problem
jac := func(dfdy *la.Triplet, dx, x float64, y []float64, args ...interface{}) error {
FAC14 := BETA * math.Exp((y[3]-y[2])/UF) / UF
FAC27 := BETA * math.Exp((y[6]-y[5])/UF) / UF
if dfdy.Max() == 0 {
dfdy.Init(8, 8, 16)
}
NU := 2
dfdy.Start()
switch mpi.Rank() {
case 0:
dfdy.Put(2+0-NU, 0, 1.0/R9)
dfdy.Put(2+1-NU, 1, 1.0/R8)
dfdy.Put(1+2-NU, 2, -ALPHA*FAC14)
dfdy.Put(0+3-NU, 3, ALPHA*FAC14)
dfdy.Put(2+2-NU, 2, 1.0/R7+FAC14)
case 1:
dfdy.Put(1+3-NU, 3, -FAC14)
dfdy.Put(2+3-NU, 3, 1.0/R5+1.0/R6+(1.0-ALPHA)*FAC14)
dfdy.Put(3+2-NU, 2, -(1.0-ALPHA)*FAC14)
dfdy.Put(2+4-NU, 4, 1.0/R4)
dfdy.Put(1+5-NU, 5, -ALPHA*FAC27)
case 2:
dfdy.Put(0+6-NU, 6, ALPHA*FAC27)
dfdy.Put(2+5-NU, 5, 1.0/R3+FAC27)
dfdy.Put(1+6-NU, 6, -FAC27)
dfdy.Put(2+6-NU, 6, 1.0/R1+1.0/R2+(1.0-ALPHA)*FAC27)
dfdy.Put(3+5-NU, 5, -(1.0-ALPHA)*FAC27)
dfdy.Put(2+7-NU, 7, 1.0/R0)
}
return nil
}
// matrix "M"
c1, c2, c3, c4, c5 := 1.0e-6, 2.0e-6, 3.0e-6, 4.0e-6, 5.0e-6
var M la.Triplet
M.Init(8, 8, 14)
M.Start()
NU := 1
switch mpi.Rank() {
case 0:
M.Put(1+0-NU, 0, -c5)
M.Put(0+1-NU, 1, c5)
M.Put(2+0-NU, 0, c5)
M.Put(1+1-NU, 1, -c5)
M.Put(1+2-NU, 2, -c4)
M.Put(1+3-NU, 3, -c3)
case 1:
M.Put(0+4-NU, 4, c3)
M.Put(2+3-NU, 3, c3)
M.Put(1+4-NU, 4, -c3)
case 2:
M.Put(1+5-NU, 5, -c2)
M.Put(1+6-NU, 6, -c1)
M.Put(0+7-NU, 7, c1)
M.Put(2+6-NU, 6, c1)
M.Put(1+7-NU, 7, -c1)
}
// flags
silent := false
fixstp := false
//method := "Dopri5"
method := "Radau5"
ndim := len(ya)
numjac := false
// structure to hold numerical results
res := ode.Results{Method: method}
// ODE solver
var osol ode.Solver
osol.Pll = true
// solve problem
if numjac {
osol.Init(method, ndim, fcn, nil, &M, ode.SimpleOutput, silent)
} else {
osol.Init(method, ndim, fcn, jac, &M, ode.SimpleOutput, silent)
}
osol.IniH = 1.0e-6 // initial step size
// set tolerances
atol, rtol := 1e-11, 1e-5
osol.SetTol(atol, rtol)
// run
t0 := time.Now()
if fixstp {
osol.Solve(ya, xa, xb, 0.01, fixstp, &res)
} else {
osol.Solve(ya, xa, xb, xb-xa, fixstp, &res)
}
// plot
if mpi.Rank() == 0 {
io.Pfmag("elapsed time = %v\n", time.Now().Sub(t0))
plt.SetForEps(2.0, 400)
args := "'b-', marker='.', lw=1, clip_on=0"
ode.Plot("/tmp/gosl/ode", "hwamplifier_mpi.eps", &res, nil, xa, xb, "", args, func() {
_, T, err := io.ReadTable("data/radau5_hwamplifier.dat")
if err != nil {
chk.Panic("%v", err)
}
for j := 0; j < ndim; j++ {
plt.Subplot(ndim+1, 1, j+1)
plt.Plot(T["x"], T[io.Sf("y%d", j)], "'k+',label='reference',ms=10")
}
})
}
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
myrank := mpi.Rank()
if myrank == 0 {
chk.PrintTitle("Test MUMPS Sol 01b")
}
var t la.Triplet
b := []float64{8.0, 45.0, -3.0, 3.0, 19.0}
switch mpi.Size() {
case 1:
t.Init(5, 5, 13)
t.Put(0, 0, 1.0)
t.Put(0, 0, 1.0)
t.Put(1, 0, 3.0)
t.Put(0, 1, 3.0)
t.Put(2, 1, -1.0)
t.Put(4, 1, 4.0)
t.Put(1, 2, 4.0)
t.Put(2, 2, -3.0)
t.Put(3, 2, 1.0)
t.Put(4, 2, 2.0)
t.Put(2, 3, 2.0)
t.Put(1, 4, 6.0)
t.Put(4, 4, 1.0)
case 2:
la.VecFill(b, 0)
if myrank == 0 {
t.Init(5, 5, 8)
t.Put(0, 0, 1.0)
t.Put(0, 0, 1.0)
t.Put(1, 0, 3.0)
t.Put(0, 1, 3.0)
t.Put(2, 1, -1.0)
t.Put(4, 1, 1.0)
t.Put(4, 1, 1.5)
t.Put(4, 1, 1.5)
b[0] = 8.0
b[1] = 40.0
b[2] = 1.5
} else {
t.Init(5, 5, 8)
t.Put(1, 2, 4.0)
t.Put(2, 2, -3.0)
t.Put(3, 2, 1.0)
t.Put(4, 2, 2.0)
t.Put(2, 3, 2.0)
t.Put(1, 4, 6.0)
t.Put(4, 4, 0.5)
t.Put(4, 4, 0.5)
b[1] = 5.0
b[2] = -4.5
b[3] = 3.0
b[4] = 19.0
}
default:
chk.Panic("this test needs 1 or 2 procs")
}
x_correct := []float64{1, 2, 3, 4, 5}
sum_b_to_root := true
la.RunMumpsTestR(&t, 1e-14, b, x_correct, sum_b_to_root)
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
io.PfYel("\nTest MPI 01\n")
}
if mpi.Size() != 3 {
chk.Panic("this test needs 3 processors")
}
n := 11
x := make([]float64, n)
id, sz := mpi.Rank(), mpi.Size()
start, endp1 := (id*n)/sz, ((id+1)*n)/sz
for i := start; i < endp1; i++ {
x[i] = float64(i)
}
// Barrier
mpi.Barrier()
io.Pfgrey("x @ proc # %d = %v\n", id, x)
// SumToRoot
r := make([]float64, n)
mpi.SumToRoot(r, x)
var tst testing.T
if id == 0 {
chk.Vector(&tst, fmt.Sprintf("SumToRoot: r @ proc # %d", id), 1e-17, r, []float64{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10})
} else {
chk.Vector(&tst, fmt.Sprintf("SumToRoot: r @ proc # %d", id), 1e-17, r, make([]float64, n))
}
// BcastFromRoot
r[0] = 666
mpi.BcastFromRoot(r)
chk.Vector(&tst, fmt.Sprintf("BcastFromRoot: r @ proc # %d", id), 1e-17, r, []float64{666, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10})
// AllReduceSum
setslice(x)
w := make([]float64, n)
mpi.AllReduceSum(x, w)
chk.Vector(&tst, fmt.Sprintf("AllReduceSum: w @ proc # %d", id), 1e-17, w, []float64{110, 110, 110, 1021, 1021, 1021, 2032, 2032, 2032, 3043, 3043})
// AllReduceSumAdd
setslice(x)
y := []float64{-1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000}
mpi.AllReduceSumAdd(y, x, w)
chk.Vector(&tst, fmt.Sprintf("AllReduceSumAdd: y @ proc # %d", id), 1e-17, y, []float64{-890, -890, -890, 21, 21, 21, 1032, 1032, 1032, 2043, 2043})
// AllReduceMin
setslice(x)
mpi.AllReduceMin(x, w)
chk.Vector(&tst, fmt.Sprintf("AllReduceMin: x @ proc # %d", id), 1e-17, x, []float64{0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3})
// AllReduceMax
setslice(x)
mpi.AllReduceMax(x, w)
chk.Vector(&tst, fmt.Sprintf("AllReduceMax: x @ proc # %d", id), 1e-17, x, []float64{100, 100, 100, 1000, 1000, 1000, 2000, 2000, 2000, 3000, 3000})
}
func main() {
mpi.Start(false)
defer func() {
mpi.Stop(false)
}()
if mpi.Rank() == 0 {
io.PfYel("\nTest MPI 02\n")
}
if mpi.Size() != 3 {
chk.Panic("this test needs 3 processors")
}
var myints []int
var mydbls []float64
switch mpi.Rank() {
case 1:
myints = []int{1, 2, 3, 4}
mydbls = []float64{-1, -2, -3}
case 2:
myints = []int{20, 30, 40, 50, 60}
mydbls = []float64{-20, -50}
}
// SingleIntSend
if mpi.Rank() == 0 {
v1 := make([]int, mpi.Size())
v2 := make([]int, mpi.Size())
allints := []int{}
alldbls := []float64{}
for proc := 1; proc < mpi.Size(); proc++ {
// SingleIntRecv
val := mpi.SingleIntRecv(proc)
io.Pf("root recieved val=%d from proc=%d\n", val, proc)
v1[proc] = val
val = mpi.SingleIntRecv(proc)
io.Pf("root recieved val=%d from proc=%d\n", val, proc)
v2[proc] = val
// IntRecv
n := mpi.SingleIntRecv(proc)
io.Pf("root recieved n=%d from proc=%d\n", n, proc)
ints := make([]int, n)
mpi.IntRecv(ints, proc)
io.Pf("root recieved ints=%v from proc=%d\n", ints, proc)
allints = append(allints, ints...)
// DblRecv
n = mpi.SingleIntRecv(proc)
io.Pf("root recieved n=%d from proc=%d\n", n, proc)
dbls := make([]float64, n)
mpi.DblRecv(dbls, proc)
io.Pf("root recieved dbls=%v from proc=%d\n", dbls, proc)
alldbls = append(alldbls, dbls...)
}
var tst testing.T
chk.Ints(&tst, "SingleIntRecv: vals", v1, []int{0, 1001, 1002})
chk.Ints(&tst, "SingleIntRecv: vals", v2, []int{0, 2001, 2002})
chk.Ints(&tst, "IntRecv: allints", allints, []int{1, 2, 3, 4, 20, 30, 40, 50, 60})
chk.Vector(&tst, "IntRecv: alldbls", 1e-17, alldbls, []float64{-1, -2, -3, -20, -50})
} else {
// SingleIntSend
mpi.SingleIntSend(1000+mpi.Rank(), 0)
mpi.SingleIntSend(2000+mpi.Rank(), 0)
// IntSend
mpi.SingleIntSend(len(myints), 0)
mpi.IntSend(myints, 0)
// DblSend
mpi.SingleIntSend(len(mydbls), 0)
mpi.DblSend(mydbls, 0)
}
}
