For the code:

unsigned char idata * data a;
unsigned char idata * data b;

void Func0()
{
    *a++ = *b++;
}


void Func1()
{
    *a = *b;
    ++a;
    ++b;
}

Keil produces:

; FUNCTION Func0 (BEGIN)
    R     MOV     R7,b
    R     INC     b
          XCH     A,R0
          MOV     A,R7
          XCH     A,R0
          MOV     A,@R0
    R     MOV     R6,a
    R     INC     a
          XCH     A,R0
          MOV     A,R6
          XCH     A,R0
          MOV     @R0,A
          RET
; FUNCTION Func0 (END)

; FUNCTION Func1 (BEGIN)
    R     MOV     R0,b
          MOV     A,@R0
    R     MOV     R0,a
          MOV     @R0,A
    R     INC     a
    R     INC     b
          RET
; FUNCTION Func1 (END)

The problem is that Keil is determined to apply the post-increment prior to the dereference. There is no requirement to do this, and so I believe that the optimiser needs work in this area.

The code of Func0 is improved (but not to the point of Func1) if I allow the use of absolute register addresses.