@matheusbsbs / matheus-bagliano / commits / 38d4debb6d

pack-objects: don't traverse objects unnecessarily

This brings back some of the performance lost in optimizing recency order inside pack objects. We were doing extreme amounts of object re-traversal: for the 2.14 million objects in the Linux kernel repository, we were calling add_to_write_order() over 1.03 billion times (a 0.2% hit rate, making 99.8% of of these calls extraneous). Two optimizations take place here- we can start our objects array iteration from a known point where we left off before we started trying to find our tags, and we don't need to do the deep dives required by add_family_to_write_order() if the object has already been marked as filled. These two optimizations bring some pretty spectacular results via `perf stat`: task-clock: 83373 ms --> 43800 ms (50% faster) cycles: 221,633,461,676 --> 116,307,209,986 (47% fewer) instructions: 149,299,179,939 --> 122,998,800,184 (18% fewer) Helped-by: Ramsay Jones (format string fix in "die" message) Signed-off-by: Dan McGee <dpmcgee@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>

Dan McGee committed Oct 18, 2011 at 00:21 UTC 38d4debb6d180ca53fcb12b8115e81fd4c5262d0

1 file changed +12 -6

builtin/pack-objects.c

+12 -6

@@ -520,7 +520,7 @@ static void add_family_to_write_order(struct object_entry **wo,

520

521 static struct object_entry **compute_write_order(void)

522 {

523 - unsigned int i, wo_end;

523 + unsigned int i, wo_end, last_untagged;

524

525 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));

526

@@ -551,7 +551,7 @@ static struct object_entry **compute_write_order(void)

551 for_each_tag_ref(mark_tagged, NULL);

552

553 /*

554 - * Give the commits in the original recency order until

554 + * Give the objects in the original recency order until

555 * we see a tagged tip.

556 */

557 for (i = wo_end = 0; i < nr_objects; i++) {

@@ -559,6 +559,7 @@ static struct object_entry **compute_write_order(void)

559 break;

560 add_to_write_order(wo, &wo_end, &objects[i]);

561 }

562 + last_untagged = i;

563

564 /*

565 * Then fill all the tagged tips.

@@ -571,7 +572,7 @@ static struct object_entry **compute_write_order(void)

572 /*

573 * And then all remaining commits and tags.

574 */

574 - for (i = 0; i < nr_objects; i++) {

575 + for (i = last_untagged; i < nr_objects; i++) {

576 if (objects[i].type != OBJ_COMMIT &&

577 objects[i].type != OBJ_TAG)

578 continue;

@@ -581,7 +582,7 @@ static struct object_entry **compute_write_order(void)

582 /*

583 * And then all the trees.

584 */

584 - for (i = 0; i < nr_objects; i++) {

585 + for (i = last_untagged; i < nr_objects; i++) {

586 if (objects[i].type != OBJ_TREE)

587 continue;

588 add_to_write_order(wo, &wo_end, &objects[i]);

@@ -590,8 +591,13 @@ static struct object_entry **compute_write_order(void)

591 /*

592 * Finally all the rest in really tight order

593 */

593 - for (i = 0; i < nr_objects; i++)

594 - add_family_to_write_order(wo, &wo_end, &objects[i]);

594 + for (i = last_untagged; i < nr_objects; i++) {

595 + if (!objects[i].filled)

596 + add_family_to_write_order(wo, &wo_end, &objects[i]);

597 + }

598 +

599 + if (wo_end != nr_objects)

600 + die("ordered %u objects, expected %"PRIu32, wo_end, nr_objects);

601

602 return wo;

603 }