Small data sections

David Brown david at
Tue Jun 26 11:58:59 UTC 2007

Just a passing thought...

Some gcc targets support the concept of "small data section".  This is 
particularly important for 32-bit RISC cpus (like the PPC), were an 
addressing mode of the form "Rx + 16bit" can fit in one instruction, 
while an absolute 32-bit address takes a lot of messing around to get 
the address into a register and then doing the real load or store.  On a 
ColdFire, which has absolute long addressing modes, it is not nearly as 

However, there are a couple of reasons why small data sections could 
still be very interesting for ColdFires.  One is that (Ax + 16bit) 
addressing mode requires 2 instruction words instead of 3 - a minor 
gain, which must be weighed against the loss of the base pointer 
(typically A5).  The other is that many ColdFires have a certain amount 
of internal RAM, and there can be very significant speed gains by 
putting small data there.  I am working on a program using a 5234 at the 
moment, and I see that access to the sdram can be very slow (relative to 
the core).  I already have the stack at the top of the internal ram, but 
have plenty of space for more data.  I can always use the "section" 
attribute to put specific data into the internal ram, but it would be 
*very* convenient if all small data went into its own section 
automatically, leaving larger buffers and arrays in the main ram.  It 
would still save significant run time even if addressed using absolute 
long addressing - although to be *really* cunning, the internal ram 
could be placed at the very top or very bottom of memory, and accessed 
using absolute short addressing.



More information about the coldfire-gnu-discuss mailing list