Small speedup in block_changes

[Andrew Pinski <pinskia at physics dot uc dot edu>]

Even though I see almost no improvement on ir.ii at -O0, I feel this
could speedup other cases where we have a small number of "real"
block changes but a large number of calls to record_block_change.
Also this improves by calling VARRAY_GROW since we know the size
of the array already.  The other thing which we can do is call
ggc_free when we know that we are no longer going to use
ib_boundaries_block anymore.

OK? Bootstrapped and tested on powerpc-darwin with no regressions.

Thanks,
Andrew Pinski

ChangeLog:
	* function.c (record_block_change): Return early if the last
	block is the same as the new one.  Grow the array to the new size.
	(finalize_block_changes): Inline record_block_change but do not
	return early if the last block is the same as the decl's initial.
	(free_block_changes): Call ggc_free on cfun->ib_boundaries_block.

Attachment: temp.diff.txt
Description: Text document