Whatever method works for you is probably the best method! There are many ways of solving this problem. Each of the methods you presented have an application - choose the one the best solves your problem.

for example:

if I have a routine to print a string. I have to pass one pointer to the string, so it would seem reasonable to use the dptr.

mov dptr,#test_str
call pstring

say I have a routine that does a memory copy. For this I need two pointers - a source and destination. It would seem reasonable to pass these pointers in registers.

;
; source pointer in R3:R1
; destination pointer in R2:R0
; byte count in R4
;
mov r2,#high(src)
mov r0,#low(src)

mov r3,#high(dest)
mov r1,#low(dest)

mov r4,#NUM_BYTES_TO_COPY

;
; yes, I know this can be optimised using xch or using
; multiple dptrs if the micro has them!!!!! It is kept simple
; for illustration.
;
memcpy:
mov a,r4
jz memend
mov dpl,r0
mov dph,r2
movx a,@dptr
inc dptr
mov r2,dph
mov r0,dpl
mov dpl,r1
mov dph,r3
movx @dptr,a
inc dptr
mov r3,dph
mov r1,dpl
dec r4
sjmp memcpy
memend:
ret

I don't know what you mean by a 'floating pointer' - 'floating point' is a method of storing and manipulating numbers - nothing to do with pointers as such. Whilst I understand english is not your first language, it is best if you keep your english simple.