func normalize(f *data.Slice) {
a := f.Vectors()
maxnorm := 0.
for i := range a[0] {
for j := range a[0][i] {
for k := range a[0][i][j] {
x, y, z := a[0][i][j][k], a[1][i][j][k], a[2][i][j][k]
norm := math.Sqrt(float64(x*x + y*y + z*z))
if norm > maxnorm {
maxnorm = norm
}
}
}
}
factor := float32(1 / maxnorm)
for i := range a[0] {
for j := range a[0][i] {
for k := range a[0][i][j] {
a[0][i][j][k] *= factor
a[1][i][j][k] *= factor
a[2][i][j][k] *= factor
}
}
}
}
func readOVF1DataBinary8(in io.Reader, t *data.Slice) {
size := t.Size()
data := t.Tensors()
// OOMMF requires this number to be first to check the format
var controlnumber float64
// OVF 1.0 is network byte order (MSB)
binary.Read(in, binary.BigEndian, &controlnumber)
if controlnumber != OVF_CONTROL_NUMBER_8 {
panic("invalid OVF1 control number: " + fmt.Sprint(controlnumber))
}
var tmp float64
for iz := 0; iz < size[Z]; iz++ {
for iy := 0; iy < size[Y]; iy++ {
for ix := 0; ix < size[X]; ix++ {
for c := 0; c < 3; c++ {
err := binary.Read(in, binary.BigEndian, &tmp)
if err != nil {
panic(err)
}
data[c][iz][iy][ix] = float32(tmp)
}
}
}
}
}
func (b *magnetization) SetArray(src *data.Slice) {
if src.Size() != b.Mesh().Size() {
src = data.Resample(src, b.Mesh().Size())
}
data.Copy(b.Buffer(), src)
M.normalize()
}
func assureGPU(s *data.Slice) *data.Slice {
if s.GPUAccess() {
return s
} else {
return cuda.GPUCopy(s)
}
}
func dumpGnuplot(f *data.Slice, m data.Meta, out io.Writer) {
buf := bufio.NewWriter(out)
defer buf.Flush()
data := f.Tensors()
cellsize := m.CellSize
// If no cell size is set, use generic cell index.
if cellsize == [3]float64{0, 0, 0} {
cellsize = [3]float64{1, 1, 1}
}
ncomp := f.NComp()
for iz := range data[0] {
z := float64(iz) * cellsize[Z]
for iy := range data[0][iz] {
y := float64(iy) * cellsize[Y]
for ix := range data[0][iz][iy] {
x := float64(ix) * cellsize[X]
fmt.Fprint(buf, x, DELIM, y, DELIM, z, DELIM)
for c := 0; c < ncomp-1; c++ {
fmt.Fprint(buf, data[c][iz][iy][ix], DELIM)
}
fmt.Fprint(buf, data[ncomp-1][iz][iy][ix])
fmt.Fprint(buf, "\n")
}
fmt.Fprint(buf, "\n")
}
}
}
// dst += LUT[region], for vectors. Used to add terms to excitation.
func RegionAddV(dst *data.Slice, lut LUTPtrs, regions *Bytes) {
util.Argument(dst.NComp() == 3)
N := dst.Len()
cfg := make1DConf(N)
k_regionaddv_async(dst.DevPtr(X), dst.DevPtr(Y), dst.DevPtr(Z),
lut[X], lut[Y], lut[Z], regions.Ptr, N, cfg)
}
// Write the slice to out in binary format. Add time stamp.
func Write(out io.Writer, s *data.Slice, info data.Meta) error {
w := newWriter(out)
// Writes the header.
w.writeString(MAGIC)
w.writeUInt64(uint64(s.NComp()))
size := s.Size()
w.writeUInt64(uint64(size[2])) // backwards compatible coordinates!
w.writeUInt64(uint64(size[1]))
w.writeUInt64(uint64(size[0]))
cell := info.CellSize
w.writeFloat64(cell[2])
w.writeFloat64(cell[1])
w.writeFloat64(cell[0])
w.writeString(info.MeshUnit)
w.writeFloat64(info.Time)
w.writeString("s") // time unit
w.writeString(info.Name)
w.writeString(info.Unit)
w.writeUInt64(4) // precission
// return header write error before writing data
if w.err != nil {
return w.err
}
w.writeData(s)
w.writeHash()
return w.err
}
func crop(f *data.Slice) {
N := f.Size()
// default ranges
x1, x2 := 0, N[X]
y1, y2 := 0, N[Y]
z1, z2 := 0, N[Z]
havework := false
if *flag_cropz != "" {
z1, z2 = parseRange(*flag_cropz, N[Z])
havework = true
}
if *flag_cropy != "" {
y1, y2 = parseRange(*flag_cropy, N[Y])
havework = true
}
if *flag_cropx != "" {
x1, x2 = parseRange(*flag_cropx, N[X])
havework = true
}
if havework {
*f = *data.Crop(f, x1, x2, y1, y2, z1, z2)
}
}
// Calculate the demag field of m * vol * Bsat, store result in B.
// m: magnetization normalized to unit length
// vol: unitless mask used to scale m's length, may be nil
// Bsat: saturation magnetization in Tesla
// B: resulting demag field, in Tesla
func (c *DemagConvolution) Exec(B, m, vol *data.Slice, Bsat LUTPtr, regions *Bytes) {
util.Argument(B.Size() == c.inputSize && m.Size() == c.inputSize)
if c.is2D() {
c.exec2D(B, m, vol, Bsat, regions)
} else {
c.exec3D(B, m, vol, Bsat, regions)
}
}
func writeVTKHeader(out io.Writer, q *data.Slice) (err error) {
gridsize := q.Size()
_, err = fmt.Fprintln(out, "<?xml version=\"1.0\"?>")
_, err = fmt.Fprintln(out, "<VTKFile type=\"StructuredGrid\" version=\"0.1\" byte_order=\"LittleEndian\">")
_, err = fmt.Fprintf(out, "\t<StructuredGrid WholeExtent=\"0 %d 0 %d 0 %d\">\n", gridsize[0]-1, gridsize[1]-1, gridsize[2]-1)
_, err = fmt.Fprintf(out, "\t\t<Piece Extent=\"0 %d 0 %d 0 %d\">\n", gridsize[0]-1, gridsize[1]-1, gridsize[2]-1)
return
}
// Calculate the demag field of m * vol * Bsat, store result in B.
// m: magnetization normalized to unit length
// vol: unitless mask used to scale m's length, may be nil
// Bsat: saturation magnetization in Tesla
// B: resulting demag field, in Tesla
func (c *DemagConvolution) Exec(B, m, vol *data.Slice, Msat MSlice) {
util.Argument(B.Size() == c.inputSize && m.Size() == c.inputSize)
if c.is2D() {
c.exec2D(B, m, vol, Msat)
} else {
c.exec3D(B, m, vol, Msat)
}
}
// Set Bth to thermal noise (Brown).
// see temperature.cu
func SetTemperature(Bth, noise *data.Slice, temp_red LUTPtr, k2mu0_VgammaDt float64, regions *Bytes) {
util.Argument(Bth.NComp() == 1 && noise.NComp() == 1)
N := Bth.Len()
cfg := make1DConf(N)
k_settemperature_async(Bth.DevPtr(0), noise.DevPtr(0), float32(k2mu0_VgammaDt), unsafe.Pointer(temp_red),
regions.Ptr, N, cfg)
}
func writeOVF2Header(out io.Writer, q *data.Slice, meta data.Meta) {
gridsize := q.Size()
cellsize := meta.CellSize
fmt.Fprintln(out, "# OOMMF OVF 2.0")
hdr(out, "Segment count", "1")
hdr(out, "Begin", "Segment")
hdr(out, "Begin", "Header")
hdr(out, "Title", meta.Name)
hdr(out, "meshtype", "rectangular")
hdr(out, "meshunit", "m")
hdr(out, "xmin", 0)
hdr(out, "ymin", 0)
hdr(out, "zmin", 0)
hdr(out, "xmax", cellsize[X]*float64(gridsize[X]))
hdr(out, "ymax", cellsize[Y]*float64(gridsize[Y]))
hdr(out, "zmax", cellsize[Z]*float64(gridsize[Z]))
name := meta.Name
var labels []interface{}
if q.NComp() == 1 {
labels = []interface{}{name}
} else {
for i := 0; i < q.NComp(); i++ {
labels = append(labels, name+"_"+string('x'+i))
}
}
hdr(out, "valuedim", q.NComp())
hdr(out, "valuelabels", labels...) // TODO
unit := meta.Unit
if unit == "" {
unit = "1"
}
if q.NComp() == 1 {
hdr(out, "valueunits", unit)
} else {
hdr(out, "valueunits", unit, unit, unit)
}
// We don't really have stages
//fmt.Fprintln(out, "# Desc: Stage simulation time: ", meta.TimeStep, " s") // TODO
hdr(out, "Desc", "Total simulation time: ", meta.Time, " s")
hdr(out, "xbase", cellsize[X]/2)
hdr(out, "ybase", cellsize[Y]/2)
hdr(out, "zbase", cellsize[Z]/2)
hdr(out, "xnodes", gridsize[X])
hdr(out, "ynodes", gridsize[Y])
hdr(out, "znodes", gridsize[Z])
hdr(out, "xstepsize", cellsize[X])
hdr(out, "ystepsize", cellsize[Y])
hdr(out, "zstepsize", cellsize[Z])
hdr(out, "End", "Header")
}
// Finds the average exchange strength around each cell, for debugging.
func ExchangeDecode(dst *data.Slice, Aex_red SymmLUT, regions *Bytes, mesh *data.Mesh) {
c := mesh.CellSize()
wx := float32(2 * 1e-18 / (c[X] * c[X]))
wy := float32(2 * 1e-18 / (c[Y] * c[Y]))
wz := float32(2 * 1e-18 / (c[Z] * c[Z]))
N := mesh.Size()
pbc := mesh.PBC_code()
cfg := make3DConf(N)
k_exchangedecode_async(dst.DevPtr(0), unsafe.Pointer(Aex_red), regions.Ptr, wx, wy, wz, N[X], N[Y], N[Z], pbc, cfg)
}
func checkNaN(s *data.Slice, name string) {
h := s.Host()
for _, h := range h {
for _, v := range h {
if math.IsNaN(float64(v)) || math.IsInf(float64(v), 0) {
util.Fatal("NaN or Inf in", name)
}
}
}
}
// Dot product.
func Dot(a, b *data.Slice) float32 {
nComp := a.NComp()
util.Argument(nComp == b.NComp())
out := reduceBuf(0)
// not async over components
for c := 0; c < nComp; c++ {
k_reducedot_async(a.DevPtr(c), b.DevPtr(c), out, 0, a.Len(), reducecfg) // all components add to out
}
return copyback(out)
}
// Set Bth to thermal noise (Brown).
// see temperature.cu
func SetTemperature(Bth, noise *data.Slice, k2mu0_Mu0VgammaDt float64, Msat, Temp, Alpha MSlice) {
util.Argument(Bth.NComp() == 1 && noise.NComp() == 1)
N := Bth.Len()
cfg := make1DConf(N)
k_settemperature2_async(Bth.DevPtr(0), noise.DevPtr(0), float32(k2mu0_Mu0VgammaDt),
Msat.DevPtr(0), Msat.Mul(0),
Temp.DevPtr(0), Temp.Mul(0),
Alpha.DevPtr(0), Alpha.Mul(0),
N, cfg)
}
// average of slice over the magnet volume
func sAverageMagnet(s *data.Slice) []float64 {
if geometry.Gpu().IsNil() {
return sAverageUniverse(s)
} else {
avg := make([]float64, s.NComp())
for i := range avg {
avg[i] = float64(cuda.Dot(s.Comp(i), geometry.Gpu())) / magnetNCell()
checkNaN1(avg[i])
}
return avg
}
}
func scale(f *data.Slice, factor float32) {
a := f.Vectors()
for i := range a[0] {
for j := range a[0][i] {
for k := range a[0][i][j] {
a[0][i][j][k] *= factor
a[1][i][j][k] *= factor
a[2][i][j][k] *= factor
}
}
}
}
// store MFM image in output, based on magnetization in inp.
func (c *MFMConvolution) Exec(outp, inp, vol *data.Slice, Bsat LUTPtr, regions *Bytes) {
for i := 0; i < 3; i++ {
zero1_async(c.fftRBuf)
copyPadMul(c.fftRBuf, inp.Comp(i), vol, c.kernSize, c.size, Bsat, regions)
c.fwPlan.ExecAsync(c.fftRBuf, c.fftCBuf)
Nx, Ny := c.fftKernSize[X]/2, c.fftKernSize[Y] // ??
kernMulC_async(c.fftCBuf, c.gpuFFTKern[i], Nx, Ny)
c.bwPlan.ExecAsync(c.fftCBuf, c.fftRBuf)
copyUnPad(outp.Comp(i), c.fftRBuf, c.size, c.kernSize)
}
}
// Render on existing image buffer. Resize it if needed
func On(img *image.RGBA, f *data.Slice, fmin, fmax string, arrowSize int, colormap ...color.RGBA) {
dim := f.NComp()
switch dim {
default:
log.Fatalf("unsupported number of components: %v", dim)
case 3:
drawVectors(img, f.Vectors(), arrowSize)
case 1:
min, max := extrema(f.Host()[0])
if fmin != "auto" {
m, err := strconv.ParseFloat(fmin, 32)
if err != nil {
util.Fatal("draw: scale:", err)
}
min = float32(m)
}
if fmax != "auto" {
m, err := strconv.ParseFloat(fmax, 32)
if err != nil {
util.Fatal("draw: scale:", err)
}
max = float32(m)
}
if min == max {
min -= 1
max += 1 // make it gray instead of black
}
drawFloats(img, f.Scalars(), min, max, colormap...)
}
}
// comma-separated values
func dumpCSV(f *data.Slice, info data.Meta, out io.Writer) {
f2 := ", " + *flag_format
a := f.Tensors()
for _, a := range a {
for _, a := range a {
for _, a := range a {
fmt.Fprintf(out, *flag_format, a[0])
for i := 1; i < len(a); i++ {
fmt.Fprintf(out, f2, a[i])
}
fmt.Fprintln(out)
}
fmt.Fprintln(out)
}
}
}
// SetMaxAngle sets dst to the maximum angle of each cells magnetization with all of its neighbors,
// provided the exchange stiffness with that neighbor is nonzero.
func SetMaxAngle(dst, m *data.Slice, Aex_red SymmLUT, regions *Bytes, mesh *data.Mesh) {
N := mesh.Size()
pbc := mesh.PBC_code()
cfg := make3DConf(N)
k_setmaxangle_async(dst.DevPtr(0),
m.DevPtr(X), m.DevPtr(Y), m.DevPtr(Z),
unsafe.Pointer(Aex_red), regions.Ptr,
N[X], N[Y], N[Z], pbc, cfg)
}
func writeVTKPoints(out io.Writer, q *data.Slice, dataformat string, info data.Meta) (err error) {
_, err = fmt.Fprintln(out, "\t\t\t<Points>")
fmt.Fprintf(out, "\t\t\t\t<DataArray type=\"Float32\" NumberOfComponents=\"3\" format=\"%s\">\n\t\t\t\t\t", dataformat)
gridsize := q.Size()
cellsize := info.CellSize
switch dataformat {
case "ascii":
for k := 0; k < gridsize[2]; k++ {
for j := 0; j < gridsize[1]; j++ {
for i := 0; i < gridsize[0]; i++ {
x := (float32)(i) * (float32)(cellsize[0])
y := (float32)(j) * (float32)(cellsize[1])
z := (float32)(k) * (float32)(cellsize[2])
_, err = fmt.Fprint(out, x, " ", y, " ", z, " ")
}
}
}
case "binary":
buffer := new(bytes.Buffer)
for k := 0; k < gridsize[2]; k++ {
for j := 0; j < gridsize[1]; j++ {
for i := 0; i < gridsize[0]; i++ {
x := (float32)(i) * (float32)(cellsize[0])
y := (float32)(j) * (float32)(cellsize[1])
z := (float32)(k) * (float32)(cellsize[2])
binary.Write(buffer, binary.LittleEndian, x)
binary.Write(buffer, binary.LittleEndian, y)
binary.Write(buffer, binary.LittleEndian, z)
}
}
}
b64len := uint32(len(buffer.Bytes()))
bufLen := new(bytes.Buffer)
binary.Write(bufLen, binary.LittleEndian, b64len)
base64out := base64.NewEncoder(base64.StdEncoding, out)
base64out.Write(bufLen.Bytes())
base64out.Write(buffer.Bytes())
base64out.Close()
default:
log.Fatalf("Illegal VTK data format: %v. Options are: ascii, binary", dataformat)
}
_, err = fmt.Fprintln(out, "\n\t\t\t\t</DataArray>")
_, err = fmt.Fprintln(out, "\t\t\t</Points>")
return
}
func normpeak(f *data.Slice) {
a := f.Vectors()
maxnorm := 0.
for i := range a[0] {
for j := range a[0][i] {
for k := range a[0][i][j] {
x, y, z := a[0][i][j][k], a[1][i][j][k], a[2][i][j][k]
norm := math.Sqrt(float64(x*x + y*y + z*z))
if norm > maxnorm {
maxnorm = norm
}
}
}
}
scale(f, float32(1/maxnorm))
}
// normalize vector data to given length
func normalize(f *data.Slice, length float64) {
a := f.Vectors()
for i := range a[0] {
for j := range a[0][i] {
for k := range a[0][i][j] {
x, y, z := a[0][i][j][k], a[1][i][j][k], a[2][i][j][k]
norm := math.Sqrt(float64(x*x + y*y + z*z))
invnorm := float32(1)
if norm != 0 {
invnorm = float32(length / norm)
}
a[0][i][j][k] *= invnorm
a[1][i][j][k] *= invnorm
a[2][i][j][k] *= invnorm
}
}
}
}
// Writes data in OMF Binary 4 format
func writeOVF1Binary4(out io.Writer, array *data.Slice) (err error) {
data := array.Tensors()
gridsize := array.Size()
var bytes []byte
// OOMMF requires this number to be first to check the format
var controlnumber float32 = OVF_CONTROL_NUMBER_4
// Conversion form float32 [4]byte in big-endian
// Inlined for performance, terabytes of data will pass here...
bytes = (*[4]byte)(unsafe.Pointer(&controlnumber))[:]
bytes[0], bytes[1], bytes[2], bytes[3] = bytes[3], bytes[2], bytes[1], bytes[0] // swap endianess
_, err = out.Write(bytes)
ncomp := array.NComp()
for iz := 0; iz < gridsize[Z]; iz++ {
for iy := 0; iy < gridsize[Y]; iy++ {
for ix := 0; ix < gridsize[X]; ix++ {
for c := 0; c < ncomp; c++ {
// dirty conversion from float32 to [4]byte
bytes = (*[4]byte)(unsafe.Pointer(&data[c][iz][iy][ix]))[:]
bytes[0], bytes[1], bytes[2], bytes[3] = bytes[3], bytes[2], bytes[1], bytes[0]
out.Write(bytes)
}
}
}
}
return
}
func writeOVF2DataBinary4(out io.Writer, array *data.Slice) {
//w.count(w.out.Write((*(*[1<<31 - 1]byte)(unsafe.Pointer(&list[0])))[0 : 4*len(list)])) // (shortcut)
data := array.Tensors()
size := array.Size()
var bytes []byte
// OOMMF requires this number to be first to check the format
var controlnumber float32 = OVF_CONTROL_NUMBER_4
bytes = (*[4]byte)(unsafe.Pointer(&controlnumber))[:]
out.Write(bytes)
ncomp := array.NComp()
for iz := 0; iz < size[Z]; iz++ {
for iy := 0; iy < size[Y]; iy++ {
for ix := 0; ix < size[X]; ix++ {
for c := 0; c < ncomp; c++ {
bytes = (*[4]byte)(unsafe.Pointer(&data[c][iz][iy][ix]))[:]
out.Write(bytes)
}
}
}
}
}
// Sets vector dst to zero where mask != 0.
func ZeroMask(dst *data.Slice, mask LUTPtr, regions *Bytes) {
N := dst.Len()
cfg := make1DConf(N)
for c := 0; c < dst.NComp(); c++ {
k_zeromask_async(dst.DevPtr(c), unsafe.Pointer(mask), regions.Ptr, N, cfg)
}
}
// average of slice over universe
func sAverageUniverse(s *data.Slice) []float64 {
nCell := float64(prod(s.Size()))
avg := make([]float64, s.NComp())
for i := range avg {
avg[i] = float64(cuda.Sum(s.Comp(i))) / nCell
checkNaN1(avg[i])
}
return avg
}