Re: A question about ld relaxation.

[Nick Clifton <nickc at redhat dot com>]

Hi Qinwei,

> I see it is reliable to relax pc-relative instructions,
> like bl-32bits relax to bl-16bits (for function call) for the
> pc-relative offset can be calculated correctly?

This is possible for *some* architectures, not all.

> 1, Can ld relax direct jump instructions? 

During a final link, for architectures that have different sized direct 
jump instructions, then yes it is possible.  (Note: Just because it is 
possible, this does not mean that the code to perform this relaxation 
has actually been written).

> Such as jump-32bits relax to jump-16bits. It is not the final link,
> and all global symbol address is not fixed. In the relaxation stage,
> can i directly use the global symbol address as the final address? 

If you are not in the final link stage, then no, you cannot use the 
current address of the global symbol as the final address of the jump.
The address of the symbol might change during the final link (if it is 
not an absolute symbol) and the address of the jump instruction itself 
cannot be known until the final link stage.


> 2, Does ld handle pc-relative relaxation and direct jump relaxation 
> the same way?

In general yes.  Obviously the instructions will have different opcodes 
and may have different binary layouts, so there will be some differences 
in how they are handled.

> 3, Almost all the targets shrink the instruction size during relaxation.
> Can i enlarge the instruction? like bl32-bits relax to bl-48bits?

In theory you could do this, although I do not know of any port of the 
GNU linker that currently supports this feature.

Cheers
  Nick