The new IRA heuristics would need more work on old-reload targets,
since flattening needs to be able to undo the cost propagation.
It's doable, but hardly seems worth it.
This patch therefore makes all the new calls to
ira_subloop_allocnos_can_differ_p return false if !ira_use_lra_p.
The color_pass code that predated the new function (and that was
the source of ira_subloop_allocnos_can_differ_p) continues to
behave as before.
It's a hack, but at least it has the advantage that the new parameter
would become obviously unused if reload and (!)ira_use_lra_p were
removed. The hack should therefore disappear alongside reload.
gcc/
PR rtl-optimization/103974
* ira-int.h (ira_subloop_allocnos_can_differ_p): Take an
extra argument, default true, that says whether old-reload
targets should be excluded.
* ira-color.c (color_pass): Pass false.
ira_assert (bitmap_bit_p (subloop_node->all_allocnos,
ALLOCNO_NUM (subloop_allocno)));
if (ira_single_region_allocno_p (a, subloop_allocno)
- || !ira_subloop_allocnos_can_differ_p (a, hard_regno >= 0))
+ || !ira_subloop_allocnos_can_differ_p (a, hard_regno >= 0,
+ false))
{
gcc_assert (!ALLOCNO_MIGHT_CONFLICT_WITH_PARENT_P
(subloop_allocno));
/* Return true if subloops that contain allocnos for A's register can
use a different assignment from A. ALLOCATED_P is true for the case
- in which allocation succeeded for A. */
+ in which allocation succeeded for A. EXCLUDE_OLD_RELOAD is true if
+ we should always return false for non-LRA targets. (This is a hack
+ and should be removed along with old reload.) */
inline bool
-ira_subloop_allocnos_can_differ_p (ira_allocno_t a, bool allocated_p = true)
+ira_subloop_allocnos_can_differ_p (ira_allocno_t a, bool allocated_p = true,
+ bool exclude_old_reload = true)
{
+ if (exclude_old_reload && !ira_use_lra_p)
+ return false;
+
auto regno = ALLOCNO_REGNO (a);
if (pic_offset_table_rtx != NULL