func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("demag1.out")
gpu.LockCudaThread()
N0, N1, N2 := 1, 3*64, 5*64
cx, cy, cz := 3e-9, 3.125e-9, 3.125e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
mbox := gpu.NewConst("m", "", mesh, nimble.UnifiedMemory, []float64{1, 0, 0})
m := mbox.Output()
const acc = 4
kernel := mag.BruteKernel(mesh, acc)
conv := gpu.NewConvolution("B", "T", mesh, nimble.UnifiedMemory, kernel, m)
B := conv.Output()
const probe = 24 * 121
outputc := B.NewReader()
nimble.RunStack()
output := host(outputc.ReadNext(mesh.NCell()))
if output[0][probe] > -0.97 || output[0][probe] < -0.99 || output[1][probe] != 0 || output[2][probe] != 0 {
fmt.Println("failed, got:", output[0][probe])
os.Exit(2)
} else {
fmt.Println("OK")
}
}
// Precision test for the kernel: thin film.
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("kernel-film.out")
gpu.LockCudaThread()
Y, X := core.IntArg(0), core.IntArg(1)
y, x := float64(Y), float64(X)
N0, N1, N2 := 1, 1024/Y, 1024/X
cx, cy, cz := 1e-9, 1e-9*y, 1e-9*x
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
mbox := gpu.NewConst("m", "", mesh, nimble.UnifiedMemory, []float64{1, 0, 0})
m := mbox.Output()
acc := 4.
kernel := mag.BruteKernel(mesh, acc)
conv := gpu.NewConvolution("B", "T", mesh, nimble.UnifiedMemory, kernel, m)
B := conv.Output()
outputc := B.NewReader()
nimble.RunStack()
output := host(outputc.ReadNext(mesh.NCell()))
Bz := core.Reshape(output[0], [3]int{N0, N1, N2})
probe := float64(Bz[N0/2][N1/2][N2/2])
fmt.Println(probe)
want := -1.
if math.Abs(probe-want) > 0.01 {
fmt.Println("FAIL")
os.Exit(2)
} else {
fmt.Println("OK")
}
}
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("demag2.out")
gpu.LockCudaThread()
N0, N1, N2 := 4, 32, 1024
cx, cy, cz := 3e-9, 3.125e-9, 3.125e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
mbox := gpu.NewConst("m", "", mesh, nimble.UnifiedMemory, []float64{1, 0, 0})
m := mbox.Output()
const acc = 4
kernel := mag.BruteKernel(mesh, acc)
conv := gpu.NewConvolution("B", "T", mesh, nimble.UnifiedMemory, kernel, m)
B := conv.Output()
outputc := B.NewReader()
nimble.RunStack()
output := host(outputc.ReadNext(mesh.NCell()))
out0 := core.Reshape(output[0], mesh.Size())
out1 := core.Reshape(output[1], mesh.Size())
out2 := core.Reshape(output[2], mesh.Size())
X, Y, Z := N0/2, N1/2, N2/2
if out0[X][Y][Z] < -0.95 || out0[X][Y][Z] > -0.90 || out1[X][Y][Z] > 0.001 || out2[X][Y][Z] > 0.001 {
fmt.Println("failed, got:", out0[X][Y][Z], out1[X][Y][Z], out2[X][Y][Z])
os.Exit(2)
} else {
fmt.Println("OK")
}
}
func main() {
nimble.Init()
nimble.SetOD("demag.out")
N0, N1, N2 := 1, 32, 128
cx, cy, cz := 3e-9, 3.125e-9, 3.125e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
//testM :=
m := nimble.NewConstant("m", "", mesh, testM).Output()
const acc = 10
kernel := mag.BruteKernel(ZeroPad(mesh), acc)
demag := conv.NewSymm2D(mesh, kernel, m)
hd := demag.Output()
// Msat := 1.0053
// aex := mag.Mu0 * 13e-12 / Msat
// hex := MakeChan3("Hex", "", mesh)
// Stack(mag.NewExchange6(m.NewReader(), hex, mesh, aex))
//
// heff := MakeChan3("Heff", "", mesh)
// Stack(NewAdder3(heff, hd.NewReader(), hex.NewReader()))
//
// const alpha = 1
// torque := MakeChan3("τ", "", mesh)
// Stack(mag.NewLLGTorque(torque, m.NewReader(), heff.NewReader(), alpha))
//
// const dt = 50e-15
// solver := mag.NewEuler(m, torque.NewReader(), mag.Gamma0, dt)
// mag.SetAll(m.UnsafeArray(), mag.Uniform(0, 0.1, 1))
// Stack(dump.NewAutosaver("h.dump", hd.NewReader(), 1))
// Stack(dump.NewAutosaver("m.dump", m.NewReader(), 1))
// Stack(dump.NewAutosaver("hex.dump", hex.NewReader(), 1))
// Stack(dump.NewAutotable("m.table", m.NewReader(), 1))
//
// RunStack()
//
// solver.Steps(100)
// res := m.UnsafeArray()
// got := [3]float32{res[0][0][0][0], res[1][0][0][0], res[2][0][0][0]}
// expect := [3]float32{-0.075877085, 0.17907967, 0.9809043}
// Log("result:", got)
// if got != expect {
// Fatal(fmt.Errorf("expected: %v", expect))
// }
//solver.Steps(10000)
ProfDump(os.Stdout)
Cleanup()
}
// Precision test for the kernel: cube.
func main() {
nimble.Init()
core.LOG = false
defer nimble.Cleanup()
nimble.SetOD("kernel-cube.out")
N0, N1, N2 := 2, 2, 2
cx, cy, cz := 1e-9, 1e-9, 1e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
for acc := 1; acc < 10; acc++ {
mag.BruteKernel(mesh, float64(acc))
//fmt.Println(acc, kernel[0][0][0][0][0], kernel[1][0][0][0][0])
}
}
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("gpu4.out")
mem := nimble.GPUMemory
const (
a = 8
N0, N1, N2 = 1, 32 * a, 128 * a
c = 1e-9
cx, cy, cz = c, c, c
Bsat = 800e3 * mag.Mu0
Aex_red = 13e-12 / (Bsat / mag.Mu0)
α = 1
)
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
m := nimble.MakeChanN(3, "m", "", mesh, mem, 0)
M := gpu.Device3(m.ChanN().UnsafeData())
M[0].Memset(float32(1 / math.Sqrt(3)))
M[1].Memset(float32(1 / math.Sqrt(3)))
M[2].Memset(float32(1 / math.Sqrt(3)))
const acc = 1
kernel := mag.BruteKernel(mesh, acc)
B := gpu.NewConvolution("B", "T", mesh, mem, kernel, m).Output()
Bex := gpu.NewExchange6("Bex", m, Aex_red).Output()
BeffBox := gpu.NewSum("Beff", B, Bex, Bsat, 1, mem)
Beff := BeffBox.Output()
tBox := gpu.NewLLGTorque("torque", m, Beff, α)
torque := tBox.Output()
solver := gpu.NewHeun(m, torque, 10e-15, mag.Gamma0)
solver.Maxerr = 2e-4
solver.Mindt = 1e-15
solver.Steps(100)
}
// Standard problem 4 on GPU
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("gpu4-3d.out")
gpu.LockCudaThread()
mem := nimble.GPUMemory
const (
N0, N1, N2 = 1 * 2, 32, 128
Sx, Sy, Sz = 3e-9, 125e-9, 500e-9
cx, cy, cz = Sx / N0, Sy / N1, Sz / N2
Bsat = 800e3 * mag.Mu0
Aex_red = 13e-12 / (Bsat / mag.Mu0)
α = 1
)
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
// TODO: MakeChanN -> NewQuant()
m := nimble.MakeChanN(3, "m", "", mesh, mem, 0)
M := gpu.Device3(m.ChanN().UnsafeData())
M[0].Memset(float32(1 / math.Sqrt(3)))
M[1].Memset(float32(1 / math.Sqrt(3)))
M[2].Memset(float32(1 / math.Sqrt(3)))
acc := 5.
kernel := mag.BruteKernel(mesh, acc)
B := gpu.NewConvolution("B", "T", mesh, mem, kernel, m).Output()
exch := gpu.NewExchange6("Bex", m, Aex_red)
Bex := exch.Output()
BeffBox := gpu.NewSum("Beff", B, Bex, Bsat, 1, mem)
Beff := BeffBox.Output()
tBox := gpu.NewLLGTorque("torque", m, Beff, α)
torque := tBox.Output()
solver := gpu.NewHeun(m, torque, 10e-15, mag.Gamma0)
solver.Maxerr = 5e-4
solver.Maxdt = 1e-12
solver.Mindt = 1e-15
solver.Headroom = 0.5
every := 100
uni.Autosave(m, every, gpu.GPUDevice)
uni.Autotable(m, every/10, gpu.GPUDevice)
solver.Advance(2e-9)
var avg [3]float32
for i := range avg {
avg[i] = gpu.Sum(m.UnsafeData()[i].Device()) / float32(mesh.NCell())
}
want := [3]float32{0, 0.12521397, 0.9669811}
err := math.Sqrt(float64(sqr(avg[0]-want[0]) + sqr(avg[1]-want[1]) + sqr(avg[2]-want[2])))
fmt.Println("avg:", avg, "err:", err)
if err > 1e-2 {
fmt.Println("FAILED")
os.Exit(2)
}
fmt.Println("OK")
const (
Bx = -24.6E-3
By = 4.3E-3
Bz = 0
)
Bext := gpu.NewConst("Bext", "T", mesh, mem, []float64{Bz, By, Bx}).Output()
BeffBox.MAdd(Bext, 1)
tBox.Alpha = 0.02
solver.Advance(1e-9)
for i := range avg {
avg[i] = gpu.Sum(m.UnsafeData()[i].Device()) / float32(mesh.NCell())
}
want = [3]float32{0.04303933, 0.13000599, -0.9842051}
err = math.Sqrt(float64(sqr(avg[0]-want[0]) + sqr(avg[1]-want[1]) + sqr(avg[2]-want[2])))
fmt.Println("avg:", avg, "err:", err)
if err > 1e-2 {
fmt.Println("FAILED")
os.Exit(2)
}
fmt.Println("OK")
}
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("gpueuler.out")
N0, N1, N2 := 1, 32, 128
cx, cy, cz := 3e-9, 3.125e-9, 3.125e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
m := nimble.MakeChanN(3, "m", "", mesh, nimble.UnifiedMemory, 0)
fmt.Println("m:", m)
acc := 8
kernel := mag.BruteKernel(mesh, acc)
B := conv.NewSymm2D("B", "T", mesh, nimble.UnifiedMemory, kernel, m).Output()
const (
Bsat = 1.0053
aex = mag.Mu0 * 13e-12 / Bsat
α = 1
)
//exch := cpu.NewExchange6("Bex", "T", nimble.UnifiedMemory, m.NewReader(), aex)
exch := gpu.NewExchange6("Bex", m, aex)
nimble.Stack(exch)
Bex := exch.Output()
// heff := MakeChan3("Heff", "", mesh)
Beff := gpu.NewSum("Beff", B, Bex, Bsat, 1, nimble.UnifiedMemory).Output()
tBox := gpu.NewLLGTorque("torque", m, Beff, α)
nimble.Stack(tBox)
torque := tBox.Output()
const dt = 100e-15
solver := gpu.NewEuler(m, torque, mag.Gamma0, dt)
M := cpu.Host(m.ChanN().UnsafeData())
for i := range M[2] {
M[2][i] = 1
M[1][i] = 0.1
}
every := 100
nimble.Autosave(B, every)
nimble.Autosave(m, every)
nimble.Autosave(Bex, every)
nimble.Autosave(Beff, every)
nimble.Autosave(torque, every)
nimble.Autotable(m, every)
nimble.RunStack()
solver.Steps(100)
res := cpu.Host(m.ChanN().UnsafeData())
got := [3]float32{res[0][0], res[1][0], res[2][0]}
expect := [3]float32{-0.033120323, 0.20761484, 0.9776498}
solver.Steps(10000)
fmt.Println("result:", got)
if got != expect {
fmt.Println("expected:", expect)
os.Exit(2)
} else {
fmt.Println("OK")
}
}
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("test4.out")
N0, N1, N2 := 1, 32, 128
cx, cy, cz := 3e-9, 3.125e-9, 3.125e-9
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
m := nimble.MakeChanN(3, "m", "", mesh, nimble.UnifiedMemory, 0)
fmt.Println("m:", m)
acc := 8
kernel := mag.BruteKernel(mesh, acc)
B := gpu.NewConvolution("B", "T", mesh, nimble.UnifiedMemory, kernel, m).Output()
const Bsat = 1.0053
const aex = mag.Mu0 * 13e-12 / Bsat
exch := cpu.NewExchange6("Bex", "T", nimble.UnifiedMemory, m.NewReader(), aex)
Bex := exch.Output()
Beff := cpu.NewSum("Beff", B, Bex, Bsat, 1, nimble.UnifiedMemory).Output()
const alpha = 1
tbox := cpu.NewLLGTorque("torque", m, Beff, alpha)
nimble.Stack(tbox)
torque := tbox.Output()
const dt = 100e-15
solver := cpu.NewEuler(m, torque.NewReader(), mag.Gamma0, dt)
M := cpu.Host(m.ChanN().UnsafeData())
for i := range M[2] {
M[2][i] = 1
M[1][i] = 0.1
}
every := 100
uni.Autosave(B, every, cpu.CPUDevice)
uni.Autosave(m, every, cpu.CPUDevice)
uni.Autosave(Bex, every, cpu.CPUDevice)
uni.Autosave(Beff, every, cpu.CPUDevice)
uni.Autosave(torque, every, cpu.CPUDevice)
uni.Autotable(m, every, cpu.CPUDevice)
nimble.RunStack()
solver.Steps(100)
res := cpu.Host(m.ChanN().UnsafeData())
got := [3]float32{res[0][0], res[1][0], res[2][0]}
expect := [3]float32{-0.03450077, 0.21015842, 0.9770585}
fmt.Println("result:", got)
if got != expect {
fmt.Println("expected:", expect)
os.Exit(2)
} else {
fmt.Println("OK")
}
}
// Standard problem 4 on GPU
func main() {
nimble.Init()
defer nimble.Cleanup()
nimble.SetOD("gpu4.out")
mem := nimble.GPUMemory
const (
N0, N1, N2 = 1, 32 * 2, 128
cx, cy, cz = 3e-9, 3.125e-9 / 2, 3.125e-9
Bsat = 1.0053
Aex_red = mag.Mu0 * 13e-12 / Bsat
α = 1
)
mesh := nimble.NewMesh(N0, N1, N2, cx, cy, cz)
fmt.Println("mesh:", mesh)
// TODO: MakeChanN -> NewQuant()
m := nimble.MakeChanN(3, "m", "", mesh, mem, 0)
M := gpu.Device3(m.ChanN().UnsafeData())
M[0].Memset(0)
M[1].Memset(0.1)
M[2].Memset(0.99)
acc := 10
kernel := mag.BruteKernel(mesh, acc)
B := gpu.NewConvolution("B", "T", mesh, mem, kernel, m).Output()
exch := gpu.NewExchange6("Bex", m, Aex_red)
nimble.Stack(exch)
Bex := exch.Output()
BeffBox := gpu.NewSum("Beff", B, Bex, Bsat, 1, mem)
Beff := BeffBox.Output()
tBox := gpu.NewLLGTorque("torque", m, Beff, α)
nimble.Stack(tBox)
torque := tBox.Output()
solver := gpu.NewHeun(m, torque, 1e-15, mag.Gamma0)
every := 100
uni.Autosave(m, every, gpu.GPUDevice)
uni.Autotable(m, every/10, gpu.GPUDevice)
solver.Advance(0.3e-9)
// res := cpu.Host(m.ChanN().UnsafeData())
// got := [3]float32{res[0][0], res[1][0], res[2][0]}
// expect := [3]float32{1.090642e-06, 0.6730072, 0.739636}
// fmt.Println("result:", got)
// if got != expect {
// fmt.Println("expected:", expect)
// os.Exit(2)
// } else {
// fmt.Println("OK")
// }
const (
Bx = -24.6E-3
By = 4.3E-3
Bz = 0
)
Bext := gpu.RunConst("Bext", "T", mesh, mem, []float64{Bz, By, Bx})
BeffBox.MAdd(Bext, 1)
tBox.SetAlpha(0.02)
solver.Advance(1e-9)
}