[patch] Update profile in loop versioning and unrolling

[roger@eyesopen.com]

On Fri, December 22, 2006 12:56 am, Zdenek Dvorak wrote:
> 	* tree-ssa-loop-niter.c (zero_p, nonzero_p): Removed.
> 	(number_of_iterations_ne, number_of_iterations_lt_to_ne,
> 	assert_no_overflow_lt, assert_loop_rolls_lt,
> 	number_of_iterations_lt, number_of_iterations_le,
> 	number_of_iterations_cond, tree_simplify_using_condition_1,
> 	number_of_iterations_exit, find_loop_niter, loop_niter_by_eval,
> 	implies_nonnegative_p, implies_ge_p, record_nonwrapping_iv,
> 	idx_infer_loop_bounds, n_of_executions_at_most, scev_probably_wraps_p):
> 	Do not use zero_p/nonzero_p.
> 	* tree-ssa-loop-manip.c (determine_exit_conditions): Ditto.
> 	* tree-ssa-loop-ivopts.c (niter_for_exit, determine_biv_step,
> 	find_interesting_uses_op, find_interesting_uses_cond,
> 	find_interesting_uses_address, find_interesting_uses_stmt,
> 	strip_offset_1, add_candidate_1, add_old_ivs_candidates,
> 	difference_cost, determine_use_iv_cost_condition,
> 	rewrite_use_compare, remove_unused_ivs): Ditto.
> 	* tree-ssa-address.c (tree_mem_ref_addr, create_mem_ref_raw): Ditto.
> 	* tree-ssa-loop-prefetch.c (idx_analyze_ref): Ditto.
> 	* tree-cfg.c (find_taken_edge_cond_expr): Ditto.
> 	* tree.h (zero_p): Declaration removed.
> 	(null_or_integer_zerop, nonnull_and_integer_nonzerop): New.

Cool!  Many thanks for addressing this.  Its difficult to clean-up this
kind of code duplication outside of stage 1.  Ok for mainline.  I look
forward to the next installment.

Roger
--