🚀 go-pugleaf

Eric Minso wrote:

> Hello,
> 
> I would like to know a bit more about the use of absolute addresses
> and relative addresses.
> 
> I have read that Windows DLL have a relocation table : this table is

quite common, other dll models have those tables,too.
they are required because i86 cpu-s (and others) do not provide a sufficient
set of operations to efficiently supporting relocatable code w.o. those
tables (as for instance, the 680xx processors do).

> needed to correct each absolute addresse in the DLL file because you
> can't assume the memory location where DLL while be loaded. But, why
> don't only use relative addresses ? (because it's not possible, I
> guess)
> 
> In fact, my questions are :
> 
> In which cases can you use relative addresses ?

at any known, constant displacement, e.g. usually not w. different dll-s.

> In which cases must you use absolute addresses ?

memory mapped port access, memory ranges which are dedicated to some static
program, by convention, #the# video memory, etc...

> In which cases can you choose ? (and which choice is better ? (i.e.
> for speed...))

direct (abs) is faster - with x86-s. relocating required once, at startup,
thus even code which could be programmed w. reg and displacement might
become faster executing.

> 
> Tell me if I am wrong : you can use relative addresses only inside a
> code section, and each time you reference another section, you must
> use absolute addresses. Is it right ?

depends on execution model and on what your 'section' is.

> 
> A last question before leaving :-)
> 
> This instruction :
> mov [ ebp + cst ], val
> (with 'ebp' initialised with loading address, and 'cst' initialised
> with offset) can solve the absolute addresses problem, so where is the
> trap ?

right. but, number of available registers is the 'trap'.
i86 cpu-s have no pc-rel subroutine call which complicates the code and can
slow down a program, considerably

hp
-- 
>>> hp -at- lxhp -dot- in-berlin -dot- de <<<
linux,asm,forth http://www.lxhp.in-berlin.de