Performance optimization of AVL tree comparator functions

MFV:
commit ee36c709c3
Author: Gvozden Neskovic <neskovic@gmail.com>
Date:   Sat Aug 27 20:12:53 2016 +0200

    perf: 2.75x faster ddt_entry_compare()
        First 256bits of ddt_key_t is a block checksum, which are expected
    to be close to random data. Hence, on average, comparison only needs to
    look at first few bytes of the keys. To reduce number of conditional
    jump instructions, the result is computed as: sign(memcmp(k1, k2)).

    Sign of an integer 'a' can be obtained as: `(0 < a) - (a < 0)` := {-1, 0, 1} ,
    which is computed efficiently.  Synthetic performance evaluation of
    original and new algorithm over 1G random keys on 2.6GHz Intel(R) Xeon(R)
    CPU E5-2660 v3:

    old     6.85789 s
    new     2.49089 s

    perf: 2.8x faster vdev_queue_offset_compare() and vdev_queue_timestamp_compare()
        Compute the result directly instead of using conditionals

    perf: zfs_range_compare()
        Speedup between 1.1x - 2.5x, depending on compiler version and
    optimization level.

    perf: spa_error_entry_compare()
        `bcmp()` is not suitable for comparator use. Use `memcmp()` instead.

    perf: 2.8x faster metaslab_compare() and metaslab_rangesize_compare()
    perf: 2.8x faster zil_bp_compare()
    perf: 2.8x faster mze_compare()
    perf: faster dbuf_compare()
    perf: faster compares in spa_misc
    perf: 2.8x faster layout_hash_compare()
    perf: 2.8x faster space_reftree_compare()
    perf: libzfs: faster avl tree comparators
    perf: guid_compare()
    perf: dsl_deadlist_compare()
    perf: perm_set_compare()
    perf: 2x faster range_tree_seg_compare()
    perf: faster unique_compare()
    perf: faster vdev_cache _compare()
    perf: faster vdev_uberblock_compare()
    perf: faster fuid _compare()
    perf: faster zfs_znode_hold_compare()

    Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
    Signed-off-by: Richard Elling <richard.elling@gmail.com>
    Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
    Closes #5033

cddl/contrib/opensolaris/lib/libzfs/common/libzfs_dataset.c

@@ -787,15 +787,13 @@ typedef struct mnttab_node {
 static int
 libzfs_mnttab_cache_compare(const void *arg1, const void *arg2)
 {
-	const mnttab_node_t *mtn1 = arg1;
-	const mnttab_node_t *mtn2 = arg2;
+	const mnttab_node_t *mtn1 = (const mnttab_node_t *)arg1;
+	const mnttab_node_t *mtn2 = (const mnttab_node_t *)arg2;
 	int rv;
 
 	rv = strcmp(mtn1->mtn_mt.mnt_special, mtn2->mtn_mt.mnt_special);
 
-	if (rv == 0)
-		return (0);
-	return (rv > 0 ? 1 : -1);
+	return (AVL_ISIGN(rv));
 }
 
 void

cddl/contrib/opensolaris/lib/libzfs/common/libzfs_iter.c

@@ -272,12 +272,7 @@ zfs_snapshot_compare(const void *larg, const void *rarg)
 	lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
 	rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
 
-	if (lcreate < rcreate)
-		return (-1);
-	else if (lcreate > rcreate)
-		return (+1);
-	else
-		return (0);
+	return (AVL_CMP(lcreate, rcreate));
 }
 
 int

cddl/contrib/opensolaris/lib/libzfs/common/libzfs_sendrecv.c

@@ -489,15 +489,10 @@ typedef struct fsavl_node {
 static int
 fsavl_compare(const void *arg1, const void *arg2)
 {
-	const fsavl_node_t *fn1 = arg1;
-	const fsavl_node_t *fn2 = arg2;
+	const fsavl_node_t *fn1 = (const fsavl_node_t *)arg1;
+	const fsavl_node_t *fn2 = (const fsavl_node_t *)arg2;
 
-	if (fn1->fn_guid > fn2->fn_guid)
-		return (+1);
-	else if (fn1->fn_guid < fn2->fn_guid)
-		return (-1);
-	else
-		return (0);
+	return (AVL_CMP(fn1->fn_guid, fn2->fn_guid));
 }
 
 /*

cddl/contrib/opensolaris/lib/libzpool/common/sys/zfs_context.h

@@ -670,6 +670,9 @@ extern zoneid_t getzoneid(void);
 #define	root_mount_wait()	do { } while (0)
 #define	root_mounted()		(1)
 
+#define	noinline	__attribute__((noinline))
+#define	likely(x)	__builtin_expect((x), 1)
+
 struct file {
 	void *dummy;
 };

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_send.c

@@ -1906,14 +1906,10 @@ typedef struct guid_map_entry {
 static int
 guid_compare(const void *arg1, const void *arg2)
 {
-	const guid_map_entry_t *gmep1 = arg1;
-	const guid_map_entry_t *gmep2 = arg2;
+	const guid_map_entry_t *gmep1 = (const guid_map_entry_t *)arg1;
+	const guid_map_entry_t *gmep2 = (const guid_map_entry_t *)arg2;
 
-	if (gmep1->guid < gmep2->guid)
-		return (-1);
-	else if (gmep1->guid > gmep2->guid)
-		return (1);
-	return (0);
+	return (AVL_CMP(gmep1->guid, gmep2->guid));
 }
 
 static void

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dnode.c

@@ -78,19 +78,13 @@ dbuf_compare(const void *x1, const void *x2)
 	const dmu_buf_impl_t *d1 = x1;
 	const dmu_buf_impl_t *d2 = x2;
 
-	if (d1->db_level < d2->db_level) {
-		return (-1);
-	}
-	if (d1->db_level > d2->db_level) {
-		return (1);
-	}
+	int cmp = AVL_CMP(d1->db_level, d2->db_level);
+	if (likely(cmp))
+		return (cmp);
 
-	if (d1->db_blkid < d2->db_blkid) {
-		return (-1);
-	}
-	if (d1->db_blkid > d2->db_blkid) {
-		return (1);
-	}
+	cmp = AVL_CMP(d1->db_blkid, d2->db_blkid);
+	if (likely(cmp))
+		return (cmp);
 
 	if (d1->db_state == DB_SEARCH) {
 		ASSERT3S(d2->db_state, !=, DB_SEARCH);
@@ -100,13 +94,7 @@ dbuf_compare(const void *x1, const void *x2)
 		return (1);
 	}
 
-	if ((uintptr_t)d1 < (uintptr_t)d2) {
-		return (-1);
-	}
-	if ((uintptr_t)d1 > (uintptr_t)d2) {
-		return (1);
-	}
-	return (0);
+	return (AVL_PCMP(d1, d2));
 }
 
 /* ARGSUSED */

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_deadlist.c

@@ -55,15 +55,10 @@
 static int
 dsl_deadlist_compare(const void *arg1, const void *arg2)
 {
-	const dsl_deadlist_entry_t *dle1 = arg1;
-	const dsl_deadlist_entry_t *dle2 = arg2;
+	const dsl_deadlist_entry_t *dle1 = (const dsl_deadlist_entry_t *)arg1;
+	const dsl_deadlist_entry_t *dle2 = (const dsl_deadlist_entry_t *)arg2;
 
-	if (dle1->dle_mintxg < dle2->dle_mintxg)
-		return (-1);
-	else if (dle1->dle_mintxg > dle2->dle_mintxg)
-		return (+1);
-	else
-		return (0);
+	return (AVL_CMP(dle1->dle_mintxg, dle2->dle_mintxg));
 }
 
 static void

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_deleg.c

@@ -384,14 +384,13 @@ typedef struct perm_set {
 static int
 perm_set_compare(const void *arg1, const void *arg2)
 {
-	const perm_set_t *node1 = arg1;
-	const perm_set_t *node2 = arg2;
+	const perm_set_t *node1 = (const perm_set_t *)arg1;
+	const perm_set_t *node2 = (const perm_set_t *)arg2;
 	int val;
 
 	val = strcmp(node1->p_setname, node2->p_setname);
-	if (val == 0)
-		return (0);
-	return (val > 0 ? 1 : -1);
+
+	return (AVL_ISIGN(val));
 }
 
 /*

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c

@@ -541,8 +541,8 @@ metaslab_class_expandable_space(metaslab_class_t *mc)
 static int
 metaslab_compare(const void *x1, const void *x2)
 {
-	const metaslab_t *m1 = x1;
-	const metaslab_t *m2 = x2;
+	const metaslab_t *m1 = (const metaslab_t *)x1;
+	const metaslab_t *m2 = (const metaslab_t *)x2;
 
 	int sort1 = 0;
 	int sort2 = 0;
@@ -568,22 +568,13 @@ metaslab_compare(const void *x1, const void *x2)
 	if (sort1 > sort2)
 		return (1);
 
-	if (m1->ms_weight < m2->ms_weight)
-		return (1);
-	if (m1->ms_weight > m2->ms_weight)
-		return (-1);
+	int cmp = AVL_CMP(m2->ms_weight, m1->ms_weight);
+	if (likely(cmp))
+		return (cmp);
 
-	/*
-	 * If the weights are identical, use the offset to force uniqueness.
-	 */
-	if (m1->ms_start < m2->ms_start)
-		return (-1);
-	if (m1->ms_start > m2->ms_start)
-		return (1);
+	IMPLY(AVL_CMP(m1->ms_start, m2->ms_start) == 0, m1 == m2);
 
-	ASSERT3P(m1, ==, m2);
-
-	return (0);
+	return (AVL_CMP(m1->ms_start, m2->ms_start));
 }
 
 /*
@@ -1195,18 +1186,14 @@ metaslab_rangesize_compare(const void *x1, const void *x2)
 	uint64_t rs_size1 = r1->rs_end - r1->rs_start;
 	uint64_t rs_size2 = r2->rs_end - r2->rs_start;
 
-	if (rs_size1 < rs_size2)
-		return (-1);
-	if (rs_size1 > rs_size2)
-		return (1);
+	int cmp = AVL_CMP(rs_size1, rs_size2);
+	if (likely(cmp))
+		return (cmp);
 
 	if (r1->rs_start < r2->rs_start)
 		return (-1);
 
-	if (r1->rs_start > r2->rs_start)
-		return (1);
-
-	return (0);
+	return (AVL_CMP(r1->rs_start, r2->rs_start));
 }
 
 /*

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sa.c

@@ -242,31 +242,23 @@ sa_cache_fini(void)
 static int
 layout_num_compare(const void *arg1, const void *arg2)
 {
-	const sa_lot_t *node1 = arg1;
-	const sa_lot_t *node2 = arg2;
+	const sa_lot_t *node1 = (const sa_lot_t *)arg1;
+	const sa_lot_t *node2 = (const sa_lot_t *)arg2;
 
-	if (node1->lot_num > node2->lot_num)
-		return (1);
-	else if (node1->lot_num < node2->lot_num)
-		return (-1);
-	return (0);
+	return (AVL_CMP(node1->lot_num, node2->lot_num));
 }
 
 static int
 layout_hash_compare(const void *arg1, const void *arg2)
 {
-	const sa_lot_t *node1 = arg1;
-	const sa_lot_t *node2 = arg2;
+	const sa_lot_t *node1 = (const sa_lot_t *)arg1;
+	const sa_lot_t *node2 = (const sa_lot_t *)arg2;
 
-	if (node1->lot_hash > node2->lot_hash)
-		return (1);
-	if (node1->lot_hash < node2->lot_hash)
-		return (-1);
-	if (node1->lot_instance > node2->lot_instance)
-		return (1);
-	if (node1->lot_instance < node2->lot_instance)
-		return (-1);
-	return (0);
+	int cmp = AVL_CMP(node1->lot_hash, node2->lot_hash);
+	if (likely(cmp))
+		return (cmp);
+
+	return (AVL_CMP(node1->lot_instance, node2->lot_instance));
 }
 
 boolean_t

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c

@@ -905,19 +905,14 @@ spa_change_guid(spa_t *spa)
 static int
 spa_error_entry_compare(const void *a, const void *b)
 {
-	spa_error_entry_t *sa = (spa_error_entry_t *)a;
-	spa_error_entry_t *sb = (spa_error_entry_t *)b;
+	const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
+	const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
 	int ret;
 
-	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
+	ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
 	    sizeof (zbookmark_phys_t));
 
-	if (ret < 0)
-		return (-1);
-	else if (ret > 0)
-		return (1);
-	else
-		return (0);
+	return (AVL_ISIGN(ret));
 }
 
 /*

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c

@@ -1016,18 +1016,13 @@ typedef struct spa_aux {
 	int		aux_count;
 } spa_aux_t;
 
-static int
+static inline int
 spa_aux_compare(const void *a, const void *b)
 {
-	const spa_aux_t *sa = a;
-	const spa_aux_t *sb = b;
+	const spa_aux_t *sa = (const spa_aux_t *)a;
+	const spa_aux_t *sb = (const spa_aux_t *)b;
 
-	if (sa->aux_guid < sb->aux_guid)
-		return (-1);
-	else if (sa->aux_guid > sb->aux_guid)
-		return (1);
-	else
-		return (0);
+	return (AVL_CMP(sa->aux_guid, sb->aux_guid));
 }
 
 void
@@ -2053,11 +2048,8 @@ spa_name_compare(const void *a1, const void *a2)
 	int s;
 
 	s = strcmp(s1->spa_name, s2->spa_name);
-	if (s > 0)
-		return (1);
-	if (s < 0)
-		return (-1);
-	return (0);
+
+	return (AVL_ISIGN(s));
 }
 
 int

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/space_reftree.c

@@ -54,20 +54,14 @@
 static int
 space_reftree_compare(const void *x1, const void *x2)
 {
-	const space_ref_t *sr1 = x1;
-	const space_ref_t *sr2 = x2;
+	const space_ref_t *sr1 = (const space_ref_t *)x1;
+	const space_ref_t *sr2 = (const space_ref_t *)x2;
 
-	if (sr1->sr_offset < sr2->sr_offset)
-		return (-1);
-	if (sr1->sr_offset > sr2->sr_offset)
-		return (1);
+	int cmp = AVL_CMP(sr1->sr_offset, sr2->sr_offset);
+	if (likely(cmp))
+		return (cmp);
 
-	if (sr1 < sr2)
-		return (-1);
-	if (sr1 > sr2)
-		return (1);
-
-	return (0);
+	return (AVL_PCMP(sr1, sr2));
 }
 
 void

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zfs_context.h

@@ -146,4 +146,7 @@ extern struct mtx zfs_debug_mtx;
 
 #define	sys_shutdown	rebooting
 
+#define	noinline	__attribute__((noinline))
+#define	likely(x)	__builtin_expect((x), 1)
+
 #endif	/* _SYS_ZFS_CONTEXT_H */

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/unique.c

@@ -42,14 +42,10 @@ typedef struct unique {
 static int
 unique_compare(const void *a, const void *b)
 {
-	const unique_t *una = a;
-	const unique_t *unb = b;
+	const unique_t *una = (const unique_t *)a;
+	const unique_t *unb = (const unique_t *)b;
 
-	if (una->un_value < unb->un_value)
-		return (-1);
-	if (una->un_value > unb->un_value)
-		return (+1);
-	return (0);
+	return (AVL_CMP(una->un_value, unb->un_value));
 }
 
 void

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_label.c

@@ -1047,19 +1047,13 @@ vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size)
  * among uberblocks with equal txg, choose the one with the latest timestamp.
  */
 static int
-vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
+vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
 {
-	if (ub1->ub_txg < ub2->ub_txg)
-		return (-1);
-	if (ub1->ub_txg > ub2->ub_txg)
-		return (1);
+	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
+	if (likely(cmp))
+		return (cmp);
 
-	if (ub1->ub_timestamp < ub2->ub_timestamp)
-		return (-1);
-	if (ub1->ub_timestamp > ub2->ub_timestamp)
-		return (1);
-
-	return (0);
+	return (AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp));
 }
 
 struct ubl_cbdata {

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c

@@ -311,20 +311,15 @@ sysctl_zfs_async_write_active_max_dirty_percent(SYSCTL_HANDLER_ARGS)
 int
 vdev_queue_offset_compare(const void *x1, const void *x2)
 {
-	const zio_t *z1 = x1;
-	const zio_t *z2 = x2;
+	const zio_t *z1 = (const zio_t *)x1;
+	const zio_t *z2 = (const zio_t *)x2;
 
-	if (z1->io_offset < z2->io_offset)
-		return (-1);
-	if (z1->io_offset > z2->io_offset)
-		return (1);
+	int cmp = AVL_CMP(z1->io_offset, z2->io_offset);
 
-	if (z1 < z2)
-		return (-1);
-	if (z1 > z2)
-		return (1);
+	if (likely(cmp))
+		return (cmp);
 
-	return (0);
+	return (AVL_PCMP(z1, z2));
 }
 
 static inline avl_tree_t *

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zap_micro.c

@@ -281,15 +281,11 @@ mze_compare(const void *arg1, const void *arg2)
 	const mzap_ent_t *mze1 = arg1;
 	const mzap_ent_t *mze2 = arg2;
 
-	if (mze1->mze_hash > mze2->mze_hash)
-		return (+1);
-	if (mze1->mze_hash < mze2->mze_hash)
-		return (-1);
-	if (mze1->mze_cd > mze2->mze_cd)
-		return (+1);
-	if (mze1->mze_cd < mze2->mze_cd)
-		return (-1);
-	return (0);
+	int cmp = AVL_CMP(mze1->mze_hash, mze2->mze_hash);
+	if (likely(cmp))
+		return (cmp);
+
+	return (AVL_CMP(mze1->mze_cd, mze2->mze_cd));
 }
 
 static int

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_fuid.c

@@ -71,14 +71,10 @@ static char *nulldomain = "";
 static int
 idx_compare(const void *arg1, const void *arg2)
 {
-	const fuid_domain_t *node1 = arg1;
-	const fuid_domain_t *node2 = arg2;
+	const fuid_domain_t *node1 = (const fuid_domain_t *)arg1;
+	const fuid_domain_t *node2 = (const fuid_domain_t *)arg2;
 
-	if (node1->f_idx < node2->f_idx)
-		return (-1);
-	else if (node1->f_idx > node2->f_idx)
-		return (1);
-	return (0);
+	return (AVL_CMP(node1->f_idx, node2->f_idx));
 }
 
 /*
@@ -87,14 +83,13 @@ idx_compare(const void *arg1, const void *arg2)
 static int
 domain_compare(const void *arg1, const void *arg2)
 {
-	const fuid_domain_t *node1 = arg1;
-	const fuid_domain_t *node2 = arg2;
+	const fuid_domain_t *node1 = (const fuid_domain_t *)arg1;
+	const fuid_domain_t *node2 = (const fuid_domain_t *)arg2;
 	int val;
 
 	val = strcmp(node1->f_ksid->kd_name, node2->f_ksid->kd_name);
-	if (val == 0)
-		return (0);
-	return (val > 0 ? 1 : -1);
+
+	return (AVL_ISIGN(val));
 }
 
 void

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_rlock.c

@@ -594,12 +594,8 @@ zfs_range_reduce(rl_t *rl, uint64_t off, uint64_t len)
 int
 zfs_range_compare(const void *arg1, const void *arg2)
 {
-	const rl_t *rl1 = arg1;
-	const rl_t *rl2 = arg2;
+	const rl_t *rl1 = (const rl_t *)arg1;
+	const rl_t *rl2 = (const rl_t *)arg2;
 
-	if (rl1->r_off > rl2->r_off)
-		return (1);
-	if (rl1->r_off < rl2->r_off)
-		return (-1);
-	return (0);
+	return (AVL_CMP(rl1->r_off, rl2->r_off));
 }

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c

@@ -131,17 +131,11 @@ zil_bp_compare(const void *x1, const void *x2)
 	const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva;
 	const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva;
 
-	if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2))
-		return (-1);
-	if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2))
-		return (1);
+	int cmp = AVL_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
+	if (likely(cmp))
+		return (cmp);
 
-	if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2))
-		return (-1);
-	if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2))
-		return (1);
-
-	return (0);
+	return (AVL_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2)));
 }
 
 static void
@@ -503,12 +497,7 @@ zil_lwb_vdev_compare(const void *x1, const void *x2)
 	const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev;
 	const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev;
 
-	if (v1 < v2)
-		return (-1);
-	if (v1 > v2)
-		return (1);
-
-	return (0);
+	return (AVL_CMP(v1, v2));
 }
 
 static lwb_t *
@@ -1626,12 +1615,7 @@ zil_aitx_compare(const void *x1, const void *x2)
 	const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid;
 	const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid;
 
-	if (o1 < o2)
-		return (-1);
-	if (o1 > o2)
-		return (1);
-
-	return (0);
+	return (AVL_CMP(o1, o2));
 }
 
 /*

sys/cddl/contrib/opensolaris/uts/common/sys/avl.h

@@ -106,6 +106,14 @@ extern "C" {
  */
 
 
+/*
+ * AVL comparator helpers
+ */
+#define	AVL_ISIGN(a)	(((a) > 0) - ((a) < 0))
+#define	AVL_CMP(a, b)	(((a) > (b)) - ((a) < (b)))
+#define	AVL_PCMP(a, b)	\
+	(((uintptr_t)(a) > (uintptr_t)(b)) - ((uintptr_t)(a) < (uintptr_t)(b)))
+
 /*
  * Type used for the root of the AVL tree.
  */