func (a *Array) Pin() {
if a.isPinned == 0 {
cu.MemHostRegister(cu.HostPtr(unsafe.Pointer(&a.List[0])), a.SizeInBytes, cu.MEMHOSTREGISTER_PORTABLE)
Debug("Successfully pinned.")
a.isPinned = 1
}
}
// Set a single value
func (b *Array) Set(comp, x, y, z int, value float64) {
b.checkBounds(comp, x, y, z)
acomp := b.Comp[comp]
index := acomp.indexOf(x, y, z)
cu.MemcpyHtoD(cu.DevicePtr(offset(uintptr(acomp.pointer), SIZEOF_FLOAT*index)),
cu.HostPtr(unsafe.Pointer(&value)),
1*SIZEOF_FLOAT)
}
func NewArrayPinned(components int, size3D []int) *Array {
t := new(Array)
t.Init(components, size3D)
cu.MemHostRegister(cu.HostPtr(unsafe.Pointer(&t.List[0])), t.SizeInBytes, cu.MEMHOSTREGISTER_PORTABLE)
Debug("Successfully pinned.")
t.isPinned = 1
return t
}
// Get a single value
func (b *Array) Get(comp, x, y, z int) float64 {
b.checkBounds(comp, x, y, z)
var value float64
acomp := b.Comp[comp]
index := acomp.indexOf(x, y, z)
cu.MemcpyDtoH(cu.HostPtr(unsafe.Pointer(&value)),
cu.DevicePtr(offset(uintptr(acomp.pointer), SIZEOF_FLOAT*index)),
1*SIZEOF_FLOAT)
return value
}
// Copy from host array to device array.
func (dst *Array) CopyFromHost(src *host.Array) {
CheckSize(dst.size4D, src.Size4D)
partPlaneN := dst.partSize[1] * dst.partSize[2] // floats per YZ plane per GPU
planeN := dst.size3D[1] * dst.size3D[2] // total floats per YZ plane
NPlane := dst.size4D[0] * dst.size3D[0] // total YZ planes (NComp * X size)
partPlaneBytes := SIZEOF_FLOAT * int64(partPlaneN) // bytes per YZ plane per GPU
for i := 0; i < NPlane; i++ {
dstOffset := i * partPlaneN
dstPtr := ArrayOffset(uintptr(dst.pointer), dstOffset)
srcOffset := i * planeN
cu.MemcpyHtoD(cu.DevicePtr(dstPtr), cu.HostPtr(&src.List[srcOffset]), partPlaneBytes)
}
}
// Copy from device array to host array.
func (src *Array) CopyToHost(dst *host.Array) {
CheckSize(dst.Size4D, src.size4D)
partPlaneN := src.partSize[1] * src.partSize[2] // floats per YZ plane per GPU
planeN := src.size3D[1] * src.size3D[2] // total floats per YZ plane
NPlane := src.size4D[0] * src.size3D[0] // total YZ planes (NComp * X size)
partPlaneBytes := SIZEOF_FLOAT * int64(partPlaneN) // bytes per YZ plane per GPU
for i := 0; i < NPlane; i++ {
srcOffset := i * partPlaneN
srcPtr := ArrayOffset(uintptr(src.pointer), srcOffset)
dstOffset := i * planeN
cu.MemcpyDtoH(cu.HostPtr(&dst.List[dstOffset]), cu.DevicePtr(srcPtr), partPlaneBytes)
}
}