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") }
// 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) }