Local variables optimization

Local variables optimization

[Hanzac Chen]

Hi,

Some idea: Maybe local stack use in the code compiled by GCC should be
optimized 'cause some local variables are conditional. If the condition
is not satisfied, then these variables don't need to be allocated from
the stack (e.g. sub $VAR_SIZE, %esp).

For example:

int *func(unsigned int cond, unsigned i)
{
   switch (cond)
   {
	case 1:
	{  int d[0x10];
	   i = memset(d, i, 0x10);
	}
	case 2:
	{  int e[0x10];
	   i = memset(e, i, 0x10);
	}
	case 3:
	{  int f[0x10];
	   i = memset(f, i, 0x10);
	}
	default:
	  break;
   }
   return 1;
}

After compiled by GCC, you'll see GCC allocated all the local variables
in case 1/case 2/case 3, this is stupid.

Hanzac

Re: Local variables optimization

[Ian Lance Taylor]

Hanzac Chen <hanzac@gmail.com> writes:

> Some idea: Maybe local stack use in the code compiled by GCC should be
> optimized 'cause some local variables are conditional. If the condition
> is not satisfied, then these variables don't need to be allocated from
> the stack (e.g. sub $VAR_SIZE, %esp).

You neglected to mention which version of gcc you were using, or which
target.  The description and test case sound like GCC PR 9997:
    http://gcc.gnu.org/PR9997
which was fixed for the 4.0 release.

Ian

Re: Local variables optimization

[Hanzac Chen]

Ian Lance Taylor wrote:
> You neglected to mention which version of gcc you were using, or which
> target.  The description and test case sound like GCC PR 9997:
>     http://gcc.gnu.org/PR9997
> which was fixed for the 4.0 release.
Sorry, I'm using GCC 3.4.4, haven't tested on 4.0.X.
I saw the PR 9997, yes, that's it.

Thanks,
Hanzac