/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _EYTZINGER_H
-#define _EYTZINGER_H
+#ifndef _LINUX_EYTZINGER_H
+#define _LINUX_EYTZINGER_H
#include <linux/bitops.h>
#include <linux/log2.h>
-#include "util.h"
+#ifdef EYTZINGER_DEBUG
+#define EYTZINGER_BUG_ON(cond) BUG_ON(cond)
+#else
+#define EYTZINGER_BUG_ON(cond)
+#endif
/*
* Traversal for trees in eytzinger layout - a full binary tree layed out in an
- * array
- */
-
-/*
- * One based indexing version:
+ * array.
*
- * With one based indexing each level of the tree starts at a power of two -
- * good for cacheline alignment:
+ * Consider using an eytzinger tree any time you would otherwise be doing binary
+ * search over an array. Binary search is a worst case scenario for branch
+ * prediction and prefetching, but in an eytzinger tree every node's children
+ * are adjacent in memory, thus we can prefetch children before knowing the
+ * result of the comparison, assuming multiple nodes fit on a cacheline.
+ *
+ * Two variants are provided, for one based indexing and zero based indexing.
+ *
+ * Zero based indexing is more convenient, but one based indexing has better
+ * alignment and thus better performance because each new level of the tree
+ * starts at a power of two, and thus if element 0 was cacheline aligned, each
+ * new level will be as well.
*/
static inline unsigned eytzinger1_child(unsigned i, unsigned child)
{
- EBUG_ON(child > 1);
+ EYTZINGER_BUG_ON(child > 1);
return (i << 1) + child;
}
static inline unsigned eytzinger1_next(unsigned i, unsigned size)
{
- EBUG_ON(i > size);
+ EYTZINGER_BUG_ON(i > size);
if (eytzinger1_right_child(i) <= size) {
i = eytzinger1_right_child(i);
static inline unsigned eytzinger1_prev(unsigned i, unsigned size)
{
- EBUG_ON(i > size);
+ EYTZINGER_BUG_ON(i > size);
if (eytzinger1_left_child(i) <= size) {
i = eytzinger1_left_child(i) + 1;
unsigned shift = __fls(size) - b;
int s;
- EBUG_ON(!i || i > size);
+ EYTZINGER_BUG_ON(!i || i > size);
i ^= 1U << b;
i <<= 1;
unsigned shift;
int s;
- EBUG_ON(!i || i > size);
+ EYTZINGER_BUG_ON(!i || i > size);
/*
* sign bit trick:
}
#define eytzinger1_for_each(_i, _size) \
- for ((_i) = eytzinger1_first((_size)); \
+ for (unsigned (_i) = eytzinger1_first((_size)); \
(_i) != 0; \
(_i) = eytzinger1_next((_i), (_size)))
static inline unsigned eytzinger0_child(unsigned i, unsigned child)
{
- EBUG_ON(child > 1);
+ EYTZINGER_BUG_ON(child > 1);
return (i << 1) + 1 + child;
}
}
#define eytzinger0_for_each(_i, _size) \
- for ((_i) = eytzinger0_first((_size)); \
+ for (unsigned (_i) = eytzinger0_first((_size)); \
(_i) != -1; \
(_i) = eytzinger0_next((_i), (_size)))
-typedef int (*eytzinger_cmp_fn)(const void *l, const void *r, size_t size);
-
/* return greatest node <= @search, or -1 if not found */
static inline ssize_t eytzinger0_find_le(void *base, size_t nr, size_t size,
- eytzinger_cmp_fn cmp, const void *search)
+ cmp_func_t cmp, const void *search)
{
unsigned i, n = 0;
do {
i = n;
- n = eytzinger0_child(i, cmp(search, base + i * size, size) >= 0);
+ n = eytzinger0_child(i, cmp(search, base + i * size) >= 0);
} while (n < nr);
if (n & 1) {
#define eytzinger0_find(base, nr, size, _cmp, search) \
({ \
- void *_base = (base); \
- void *_search = (search); \
- size_t _nr = (nr); \
- size_t _size = (size); \
- size_t _i = 0; \
+ void *_base = (base); \
+ const void *_search = (search); \
+ size_t _nr = (nr); \
+ size_t _size = (size); \
+ size_t _i = 0; \
int _res; \
\
while (_i < _nr && \
- (_res = _cmp(_search, _base + _i * _size, _size))) \
+ (_res = _cmp(_search, _base + _i * _size))) \
_i = eytzinger0_child(_i, _res > 0); \
_i; \
})
-void eytzinger0_sort(void *, size_t, size_t,
- int (*cmp_func)(const void *, const void *, size_t),
- void (*swap_func)(void *, void *, size_t));
+void eytzinger0_sort_r(void *, size_t, size_t,
+ cmp_r_func_t, swap_r_func_t, const void *);
+void eytzinger0_sort(void *, size_t, size_t, cmp_func_t, swap_func_t);
-#endif /* _EYTZINGER_H */
+#endif /* _LINUX_EYTZINGER_H */