Christoph,

this is exactly what I consider absolutely unnecessary (even harmful) - to modify the C code so that it compiles into effective (in any way) binary - unless you are the compiler maker.

Just for the record, the same line compiles into 21 bytes in SDCC:

   018F AA 82               458 	mov	r2,dpl
   0191 7B 07               461 	mov	r3,#0x07
   0193                     462 00101$:
   0193 ED                  466 	mov	a,r5
   0194 2D                  468 	add	a,r5
   0195 FC                  469 	mov	r4,a
   0196 74 01               471 	mov	a,#0x01
   0198 5A                  472 	anl	a,r2
   0199 4C                  474 	orl	a,r4
   019A FD                  475 	mov	r5,a
   019B EA                  479 	mov	a,r2
   019C C3                  480 	clr	c
   019D 13                  481 	rrc	a
   019E FA                  482 	mov	r2,a
   019F DB F2               486 	djnz	r3,00101$
   01A1 8D 82               490 	mov	dpl,r5
   01A3 22                  492 	ret

... and the following spares 2 more bytes, yielding 19 bytes:

UC r,i=7;DO{r+=r+(c&1);c/=2;}WH(--i);RT r;   
   018F AA 82               458 	mov	r2,dpl
   0191 7B 07               461 	mov	r3,#0x07
   0193                     462 00101$:
   0193 74 01               464 	mov	a,#0x01
   0195 5A                  465 	anl	a,r2
   0196 2C                  468 	add	a,r4
   0197 2C                  471 	add	a,r4
   0198 FC                  472 	mov	r4,a
   0199 EA                  476 	mov	a,r2
   019A C3                  477 	clr	c
   019B 13                  478 	rrc	a
   019C FA                  479 	mov	r2,a
   019D DB F4               483 	djnz	r3,00101$
   019F 8C 82               487 	mov	dpl,r4
   01A1 22                  489 	ret 

(but in Keil's makes again that stupid assumption that r cannot be in register, so it is less effective there).

JW

PS. would not be those stupid "features" in those two compilers,

UC r,i=1;DO{r+=r+(c&i);}WH(i+=i);RT r;

would have even more potential to be the best binary-size-wise...