// return a 1-float CUDA reduction buffer from a pool
// initialized to initVal
func reduceBuf(initVal float32) unsafe.Pointer {
if reduceBuffers == nil {
initReduceBuf()
}
buf := <-reduceBuffers
cu.MemsetD32Async(cu.DevicePtr(uintptr(buf)), math.Float32bits(initVal), 1, stream0)
return buf
}
// Memset sets the Slice's components to the specified values.
// To be carefully used on unified slice (need sync)
func Memset(s *data.Slice, val ...float32) {
if Synchronous { // debug
Sync()
timer.Start("memset")
}
util.Argument(len(val) == s.NComp())
for c, v := range val {
cu.MemsetD32Async(cu.DevicePtr(uintptr(s.DevPtr(c))), math.Float32bits(v), int64(s.Len()), stream0)
}
if Synchronous { //debug
Sync()
timer.Stop("memset")
}
}
// zero 1-component slice
func zero1_async(dst *data.Slice) {
cu.MemsetD32Async(cu.DevicePtr(uintptr(dst.DevPtr(0))), 0, int64(dst.Len()), stream0)
}