Avoid full DOM walk in LIM fill_always_executed_in

Avoid full DOM walk in LIM fill_always_executed_in

This avoids a full DOM walk via get_loop_body_in_dom_order in the
loop body walk of fill_always_executed_in which is often terminating
the walk of a loop body early by integrating the DOM walk of
get_loop_body_in_dom_order with the actual processing done by
fill_always_executed_in.  This trades the fully populated loop
body array with a worklist allocation of the same size and thus
should be a strict improvement over the recursive approach of
get_loop_body_in_dom_order.

2021-09-09  Richard Biener  <rguenther@suse.de>

	* tree-ssa-loop-im.c (fill_always_executed_in_1): Integrate
	DOM walk from get_loop_body_in_dom_order using a worklist
	approach.