freshlybakedca.ke
Git fork
Merge branch 'jk/fast-import-use-hashmap'
The custom hash function used by "git fast-import" has been
replaced with the one from hashmap.c, which gave us a nice
performance boost.
* jk/fast-import-use-hashmap:
fast-import: replace custom hash with hashmap.c
+36
-25
fast-import.c
+36
-25
fast-import.c
···
39
39
40
40
struct object_entry {
41
41
struct pack_idx_entry idx;
42
-
struct object_entry *next;
42
+
struct hashmap_entry ent;
43
43
uint32_t type : TYPE_BITS,
44
44
pack_id : PACK_ID_BITS,
45
45
depth : DEPTH_BITS;
46
46
};
47
+
48
+
static int object_entry_hashcmp(const void *map_data,
49
+
const struct hashmap_entry *eptr,
50
+
const struct hashmap_entry *entry_or_key,
51
+
const void *keydata)
52
+
{
53
+
const struct object_id *oid = keydata;
54
+
const struct object_entry *e1, *e2;
55
+
56
+
e1 = container_of(eptr, const struct object_entry, ent);
57
+
if (oid)
58
+
return oidcmp(&e1->idx.oid, oid);
59
+
60
+
e2 = container_of(entry_or_key, const struct object_entry, ent);
61
+
return oidcmp(&e1->idx.oid, &e2->idx.oid);
62
+
}
47
63
48
64
struct object_entry_pool {
49
65
struct object_entry_pool *next_pool;
···
178
194
/* Table of objects we've written. */
179
195
static unsigned int object_entry_alloc = 5000;
180
196
static struct object_entry_pool *blocks;
181
-
static struct object_entry *object_table[1 << 16];
197
+
static struct hashmap object_table;
182
198
static struct mark_set *marks;
183
199
static const char *export_marks_file;
184
200
static const char *import_marks_file;
···
455
471
456
472
static struct object_entry *find_object(struct object_id *oid)
457
473
{
458
-
unsigned int h = oid->hash[0] << 8 | oid->hash[1];
459
-
struct object_entry *e;
460
-
for (e = object_table[h]; e; e = e->next)
461
-
if (oideq(oid, &e->idx.oid))
462
-
return e;
463
-
return NULL;
474
+
return hashmap_get_entry_from_hash(&object_table, oidhash(oid), oid,
475
+
struct object_entry, ent);
464
476
}
465
477
466
478
static struct object_entry *insert_object(struct object_id *oid)
467
479
{
468
-
unsigned int h = oid->hash[0] << 8 | oid->hash[1];
469
-
struct object_entry *e = object_table[h];
480
+
struct object_entry *e;
481
+
unsigned int hash = oidhash(oid);
470
482
471
-
while (e) {
472
-
if (oideq(oid, &e->idx.oid))
473
-
return e;
474
-
e = e->next;
483
+
e = hashmap_get_entry_from_hash(&object_table, hash, oid,
484
+
struct object_entry, ent);
485
+
if (!e) {
486
+
e = new_object(oid);
487
+
e->idx.offset = 0;
488
+
hashmap_entry_init(&e->ent, hash);
489
+
hashmap_add(&object_table, &e->ent);
475
490
}
476
491
477
-
e = new_object(oid);
478
-
e->next = object_table[h];
479
-
e->idx.offset = 0;
480
-
object_table[h] = e;
481
492
return e;
482
493
}
483
494
484
495
static void invalidate_pack_id(unsigned int id)
485
496
{
486
-
unsigned int h;
487
497
unsigned long lu;
488
498
struct tag *t;
499
+
struct hashmap_iter iter;
500
+
struct object_entry *e;
489
501
490
-
for (h = 0; h < ARRAY_SIZE(object_table); h++) {
491
-
struct object_entry *e;
492
-
493
-
for (e = object_table[h]; e; e = e->next)
494
-
if (e->pack_id == id)
495
-
e->pack_id = MAX_PACK_ID;
502
+
hashmap_for_each_entry(&object_table, &iter, e, ent) {
503
+
if (e->pack_id == id)
504
+
e->pack_id = MAX_PACK_ID;
496
505
}
497
506
498
507
for (lu = 0; lu < branch_table_sz; lu++) {
···
3510
3519
branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3511
3520
avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3512
3521
marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3522
+
3523
+
hashmap_init(&object_table, object_entry_hashcmp, NULL, 0);
3513
3524
3514
3525
/*
3515
3526
* We don't parse most options until after we've seen the set of
+23
t/perf/p9300-fast-import-export.sh
+23
t/perf/p9300-fast-import-export.sh
···
1
+
#!/bin/sh
2
+
3
+
test_description='test fast-import and fast-export performance'
4
+
. ./perf-lib.sh
5
+
6
+
test_perf_default_repo
7
+
8
+
# Use --no-data here to produce a vastly smaller export file.
9
+
# This is much cheaper to work with but should still exercise
10
+
# fast-import pretty well (we'll still process all commits and
11
+
# trees, which account for 60% or more of objects in most repos).
12
+
#
13
+
# Use --reencode to avoid the default of aborting on non-utf8 commits,
14
+
# which lets this test run against a wider variety of sample repos.
15
+
test_perf 'export (no-blobs)' '
16
+
git fast-export --reencode=yes --no-data HEAD >export
17
+
'
18
+
19
+
test_perf 'import (no-blobs)' '
20
+
git fast-import --force <export
21
+
'
22
+
23
+
test_done