// 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("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")
}
}
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 Live(in_ nimble.ChanN) {
go func() {
in := in_.NewReader()
Frame = new(dump.Frame)
Frame.Components = in.NComp()
Frame.MeshSize = in.Mesh().Size()
Frame.MeshStep = in.Mesh().CellSize()
Crop2 = Frame.MeshSize
ncell := in.Mesh().NCell()
n := ncell * in.NComp()
Frame.Data = make([]float32, n)
Init(800, 600, true, 2, true)
InitInputHandlers()
defer glfw.CloseWindow()
defer glfw.Terminate()
Viewpos[2] = -20
wantframe := new(atomicbool)
wantframe.set(true)
go func() {
gpu.LockCudaThread()
for {
data := in.ReadNext(ncell)
if wantframe.get() {
for i, d := range data {
d.Device().CopyDtoH(Frame.Data[i*ncell : (i+1)*ncell])
}
wantframe.set(false)
}
in.ReadDone()
}
}()
for glfw.WindowParam(glfw.Opened) == 1 { // window open
Render()
if Wantscrot {
Screenshot()
Wantscrot = false
}
if wantframe.get() == false {
PreRender(Frame)
wantframe.set(true)
}
}
}()
}
// 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")
}