lib/rbtree: use '+' instead of '|' for setting color.
authorNoah Goldstein <goldstein.w.n@gmail.com>
Tue, 4 Apr 2023 22:13:51 +0000 (17:13 -0500)
committerAndrew Morton <akpm@linux-foundation.org>
Tue, 18 Apr 2023 23:39:33 +0000 (16:39 -0700)
This has a slight benefit for x86 and has no effect on other targets.

The benefit to x86 is it change the codegen for setting a node to block
from `mov %r0, %r1; or $RB_BLACK, %r1` to `lea RB_BLACK(%r0), %r1` which
saves an instructions.

In all other cases it just replace ALU with ALU (or -> and) which
perform the same on all machines I am aware of.

Total instructions in rbtree.o:
    Before  - 802
    After   - 782

so it saves about 20 `mov` instructions.

Link: https://lkml.kernel.org/r/20230404221350.3806566-1-goldstein.w.n@gmail.com
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Cc: Michel Lespinasse <michel@lespinasse.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
include/linux/rbtree_augmented.h
lib/rbtree.c

index d1c53e9..7ee7ed5 100644 (file)
@@ -156,13 +156,13 @@ RB_DECLARE_CALLBACKS(RBSTATIC, RBNAME,                                          \
 
 static inline void rb_set_parent(struct rb_node *rb, struct rb_node *p)
 {
-       rb->__rb_parent_color = rb_color(rb) | (unsigned long)p;
+       rb->__rb_parent_color = rb_color(rb) + (unsigned long)p;
 }
 
 static inline void rb_set_parent_color(struct rb_node *rb,
                                       struct rb_node *p, int color)
 {
-       rb->__rb_parent_color = (unsigned long)p | color;
+       rb->__rb_parent_color = (unsigned long)p + color;
 }
 
 static inline void
index c4ac5c2..5114eda 100644 (file)
@@ -58,7 +58,7 @@
 
 static inline void rb_set_black(struct rb_node *rb)
 {
-       rb->__rb_parent_color |= RB_BLACK;
+       rb->__rb_parent_color += RB_BLACK;
 }
 
 static inline struct rb_node *rb_red_parent(struct rb_node *red)