func main() {
cuda.Init()
N0, N1, N2 := 1, 64, 128
c := 1.
mesh := data.NewMesh(N0, N1, N2, c/2, c*2, c)
m := cuda.NewSlice(3, mesh)
conv := cuda.NewDemag(mesh)
cuda.Memset(m, 1, 1, 1)
B := cuda.NewSlice(3, mesh)
Bsat := 1.
vol := data.NilSlice(1, mesh)
conv.Exec(B, m, vol, Bsat)
out := B.HostCopy()
bx := out.Vectors()[0][N0/2][N1/2][N2/2]
by := out.Vectors()[1][N0/2][N1/2][N2/2]
bz := out.Vectors()[2][N0/2][N1/2][N2/2]
fmt.Println("demag tensor:", bx, by, bz)
check(bx, -1)
check(by, 0)
check(bz, 0)
fmt.Println("OK")
}
// returns host buffer for storing output before being flushed to disk.
// takes one from the pool or allocates a new one when the pool is empty
// and less than maxOutputQueLen buffers already are in use.
func hostbuf() *data.Slice {
select {
case b := <-hBuf:
cuda.Memset(b, 0, 0, 0) // not strictly needed
return b
default:
if nOutBuf < maxOutputQueLen {
nOutBuf++
return cuda.NewUnifiedSlice(3, mesh)
}
}
panic("unreachable")
}
// Memset with synchronization.
func (b *buffered) memset(val ...float32) {
s := b.Write()
cuda.Memset(s, val...)
b.WriteDone()
}
