Blackfin function epilogue

Hi,
  deas anybody know what's diference between:

   SP + = 16; 
   P0=[FP+4]; 
   (R7:4)=[SP++]; 
   UNLINK; 
   JUMP (P0); 

and

   SP + = 16; 
   (R7:4)=[SP++]; 
   UNLINK; 
   RTS; 

Thanks

JkH

jenikh <jenikh@centrum.cz> wrote:
[ADI blackfin]
>   deas anybody know what's diference between:
>    SP + = 16; 
>    P0=[FP+4]; 
>    (R7:4)=[SP++]; 
>    UNLINK; 
>    JUMP (P0); 

> and

>    SP + = 16; 
>    (R7:4)=[SP++]; 
>    UNLINK; 
>    RTS; 

	your solution is one cycle faster. ;-)

WD
-- 

>
>	your solution is one cycle faster. ;-)
>
>WD
>-- 

So why VisualDSP++ use longer solution? ;-)

JkH

jenikh wrote:

> Hi,
>   deas anybody know what's diference between:
> 
>    SP + = 16;
>    P0=[FP+4];
>    (R7:4)=[SP++];
>    UNLINK;
>    JUMP (P0);
> 
> and
> 
>    SP + = 16;
>    (R7:4)=[SP++];
>    UNLINK;
>    RTS;
> 
> Thanks
> 
> JkH

It has been long time since I have programmed BF asm, but IIRC RTS returns
to address pointed by register RETS, while the longer version fetches the
return address from memory. Functionally they seem identical. 

All I can think of is that the the longer version migth not issue a pipeline
stall as the memory fetch for the return address is done earlier and stored
in P0. This is only a guess though.. perhaps you should benchmark the two
version.

-- 
----
Jani Huhtanen
Tampere University of Technology, Pori