DSO: pc relative symbol address

DSO: pc relative symbol address

[kum]

Hi,
Consider the scrap of code

  static int my_foo1(int i)
    {...}

  int (*my_foo)(int) = my_foo1;
  ...

The compiler emits code like this (excerpt from .s)
...
	lw	$2,%got(my_foo1)($28)
...

There are no direct invocation of my_foo1. What prevents the compiler
from emitting a PC relative address to store in my_foo? Something like
  lw $2, PC - <offset>

I guess that this offset is compile-time known quantity. Kindly enlighten me.

If this is possible what compile option(s) turns it on? Is it possible
to do this optimization while building (elf) shared objects too?

Thanks for clearing my ignorance,
kumaresh

Re: DSO: pc relative symbol address

[Ian Lance Taylor]

kum <a.kumaresh@gmail.com> writes:

> There are no direct invocation of my_foo1. What prevents the compiler
> from emitting a PC relative address to store in my_foo? Something like
>   lw $2, PC - <offset>

You need to get the address of my_foo1 into $2.  Using a PC relative
value only works if your processor has a PC-relative load instruction.
Not all processors have such an instruction.  If your processor has
one and gcc is not using it, then this may be a bug.

Ian

Re: DSO: pc relative symbol address

[kum]

I am on mips and it neither supports PC relative load nor the PC can
be read . You are correct, I should have looked up MIPS addressing
modes before coming here... Thanks Ian.

kumaresh



On Fri, Mar 19, 2010 at 9:20 PM, Ian Lance Taylor <iant@google.com> wrote:
> kum <a.kumaresh@gmail.com> writes:
>
>> There are no direct invocation of my_foo1. What prevents the compiler
>> from emitting a PC relative address to store in my_foo? Something like
>>   lw $2, PC - <offset>
>
> You need to get the address of my_foo1 into $2.  Using a PC relative
> value only works if your processor has a PC-relative load instruction.
> Not all processors have such an instruction.  If your processor has
> one and gcc is not using it, then this may be a bug.
>
> Ian
>