@@ -2211,7 +2211,7 @@ static inline void mast_rebalance_next(struct maple_subtree_state *mast)
{
unsigned char b_end = mast->bn->b_end;
- mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node),
+ mas_mab_cp(mast->orig_r, 0, mt_pivot_count(mast->orig_r->node),
mast->bn, b_end);
mast->orig_r->last = mast->orig_r->max;
}
@@ -2692,7 +2692,7 @@ static inline void mast_combine_cp_right(struct maple_subtree_state *mast)
return;
mas_mab_cp(mast->orig_r, mast->orig_r->offset + 1,
- mt_slot_count(mast->orig_r->node), mast->bn,
+ mt_pivot_count(mast->orig_r->node), mast->bn,
mast->bn->b_end);
mast->orig_r->last = mast->orig_r->max;
}
@@ -2963,7 +2963,7 @@ static inline int mas_rebalance(struct ma_state *mas,
l_mas = r_mas = *mas;
if (mas_next_sibling(&r_mas)) {
- mas_mab_cp(&r_mas, 0, mt_slot_count(r_mas.node), b_node, b_end);
+ mas_mab_cp(&r_mas, 0, mt_pivot_count(r_mas.node), b_node, b_end);
r_mas.last = r_mas.index = r_mas.max;
} else {
mas_prev_sibling(&l_mas);
mas_mab_cp() copy a maple_node's range inclusively, and the maximum valid data index is (mt_slot_count() - 1) instead of mt_slot_count(). But if using (mt_slot_count() - 1) directly, it would have more instructions. The good news is we have defined mt_pivot_count() == (mt_slot_count - 1), except for maple_dense node. And mas_mab_cp() doesn't handle maple_dense node. So it is safe to use it. By doing so have almost identical generated code: Current: lib/maple_tree.c:2209: mas_mab_cp(mast->orig_r, 0, mt_slot_count(mast->orig_r->node), movl %esi, (%esp) # _135, movl -88(%ebp), %eax # %sfp, _137 movzbl mt_slots(%edx), %ecx # mt_slots[_142], mt_slots[_142] xorl %edx, %edx # movl %eax, 4(%esp) # _137, movl %ebx, %eax # _138, call mas_mab_cp # After applying this patch: lib/maple_tree.c:2209: mas_mab_cp(mast->orig_r, 0, mt_pivot_count(mast->orig_r->node), movl %esi, (%esp) # _135, movl -88(%ebp), %eax # %sfp, _137 movzbl mt_pivots(%edx), %ecx # mt_pivots[_142], mt_pivots[_142] xorl %edx, %edx # movl %eax, 4(%esp) # _137, movl %ebx, %eax # _138, call mas_mab_cp # The difference is we access mt_pivots to get ecx instead of mt_slots, which doesn't expect performance difference. Signed-off-by: Wei Yang <richard.weiyang@gmail.com> --- lib/maple_tree.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)