Using this shadow register method compiles down nicely, usually to around 6 instructions, depending on whether "val" and "mask" are constant integers (i.e., literal, not const integer) or variables. From my previous example, the line PORT_WR(P2, i, 0x0F) results in:

    MOV  A,P2_OLR
    XRL  A,R7
    ANL  A,#0FH
    XRL  A,P2_OLR
    MOV  P2_OLR,A
    MOV  P2,A

This is using Keil's C51, with a different compiler YMMV, so check the compiler output to learn which way is more efficient (if you care), the AND-XOR method or the AND-CPL-OR method of updating the shadow register. The nice thing about preprocessor macros is that you can change the macro body in one place only.