git/sha1-lookup.c

130 lines
3.2 KiB
C
Raw Normal View History

sha1-lookup: more memory efficient search in sorted list of SHA-1 Currently, when looking for a packed object from the pack idx, a simple binary search is used. A conventional binary search loop looks like this: unsigned lo, hi; do { unsigned mi = (lo + hi) / 2; int cmp = "entry pointed at by mi" minus "target"; if (!cmp) return mi; "mi is the wanted one" if (cmp > 0) hi = mi; "mi is larger than target" else lo = mi+1; "mi is smaller than target" } while (lo < hi); "did not find what we wanted" The invariants are: - When entering the loop, 'lo' points at a slot that is never above the target (it could be at the target), 'hi' points at a slot that is guaranteed to be above the target (it can never be at the target). - We find a point 'mi' between 'lo' and 'hi' ('mi' could be the same as 'lo', but never can be as high as 'hi'), and check if 'mi' hits the target. There are three cases: - if it is a hit, we have found what we are looking for; - if it is strictly higher than the target, we set it to 'hi', and repeat the search. - if it is strictly lower than the target, we update 'lo' to one slot after it, because we allow 'lo' to be at the target and 'mi' is known to be below the target. If the loop exits, there is no matching entry. When choosing 'mi', we do not have to take the "middle" but anywhere in between 'lo' and 'hi', as long as lo <= mi < hi is satisfied. When we somehow know that the distance between the target and 'lo' is much shorter than the target and 'hi', we could pick 'mi' that is much closer to 'lo' than (hi+lo)/2, which a conventional binary search would pick. This patch takes advantage of the fact that the SHA-1 is a good hash function, and as long as there are enough entries in the table, we can expect uniform distribution. An entry that begins with for example "deadbeef..." is much likely to appear much later than in the midway of a reasonably populated table. In fact, it can be expected to be near 87% (222/256) from the top of the table. This is a work-in-progress and has switches to allow easier experiments and debugging. Exporting GIT_USE_LOOKUP environment variable enables this code. On my admittedly memory starved machine, with a partial KDE repository (3.0G pack with 95M idx): $ GIT_USE_LOOKUP=t git log -800 --stat HEAD >/dev/null 3.93user 0.16system 0:04.09elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+55588minor)pagefaults 0swaps Without the patch, the numbers are: $ git log -800 --stat HEAD >/dev/null 4.00user 0.15system 0:04.17elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+60258minor)pagefaults 0swaps In the same repository: $ GIT_USE_LOOKUP=t git log -2000 HEAD >/dev/null 0.12user 0.00system 0:00.12elapsed 97%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+4241minor)pagefaults 0swaps Without the patch, the numbers are: $ git log -2000 HEAD >/dev/null 0.05user 0.01system 0:00.07elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+8506minor)pagefaults 0swaps There isn't much time difference, but the number of minor faults seems to show that we are touching much smaller number of pages, which is expected. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2007-12-29 10:05:47 +00:00
#include "cache.h"
#include "sha1-lookup.h"
static uint32_t take2(const unsigned char *sha1)
{
return ((sha1[0] << 8) | sha1[1]);
}
/*
* Conventional binary search loop looks like this:
*
* do {
* int mi = lo + (hi - lo) / 2;
* int cmp = "entry pointed at by mi" minus "target";
* if (!cmp)
* return (mi is the wanted one)
* if (cmp > 0)
* hi = mi; "mi is larger than target"
* else
* lo = mi+1; "mi is smaller than target"
* } while (lo < hi);
*
* The invariants are:
*
* - When entering the loop, lo points at a slot that is never
* above the target (it could be at the target), hi points at a
* slot that is guaranteed to be above the target (it can never
* be at the target).
*
* - We find a point 'mi' between lo and hi (mi could be the same
* as lo, but never can be the same as hi), and check if it hits
* the target. There are three cases:
*
* - if it is a hit, we are happy.
*
* - if it is strictly higher than the target, we update hi with
* it.
*
* - if it is strictly lower than the target, we update lo to be
* one slot after it, because we allow lo to be at the target.
*
* When choosing 'mi', we do not have to take the "middle" but
* anywhere in between lo and hi, as long as lo <= mi < hi is
* satisfied. When we somehow know that the distance between the
* target and lo is much shorter than the target and hi, we could
* pick mi that is much closer to lo than the midway.
*/
/*
* The table should contain "nr" elements.
* The sha1 of element i (between 0 and nr - 1) should be returned
* by "fn(i, table)".
*/
int sha1_pos(const unsigned char *hash, void *table, size_t nr,
sha1_access_fn fn)
{
size_t hi = nr;
size_t lo = 0;
size_t mi = 0;
if (!nr)
return -1;
if (nr != 1) {
size_t lov, hiv, miv, ofs;
for (ofs = 0; ofs < the_hash_algo->rawsz - 2; ofs += 2) {
lov = take2(fn(0, table) + ofs);
hiv = take2(fn(nr - 1, table) + ofs);
miv = take2(hash + ofs);
if (miv < lov)
return -1;
if (hiv < miv)
msvc: avoid using minus operator on unsigned types MSVC complains about this with `-Wall`, which can be taken as a sign that this is indeed a real bug. The symptom is: C4146: unary minus operator applied to unsigned type, result still unsigned Let's avoid this warning in the minimal way, e.g. writing `-1 - <unsigned value>` instead of `-<unsigned value> - 1`. Note that the change in the `estimate_cache_size()` function is needed because MSVC considers the "return type" of the `sizeof()` operator to be `size_t`, i.e. unsigned, and therefore it cannot be negated using the unary minus operator. Even worse, that arithmetic is doing extra work, in vain. We want to calculate the entry extra cache size as the difference between the size of the `cache_entry` structure minus the size of the `ondisk_cache_entry` structure, padded to the appropriate alignment boundary. To that end, we start by assigning that difference to the `per_entry` variable, and then abuse the `len` parameter of the `align_padding_size()` macro to take the negative size of the ondisk entry size. Essentially, we try to avoid passing the already calculated difference to that macro by passing the operands of that difference instead, when the macro expects operands of an addition: #define align_padding_size(size, len) \ ((size + (len) + 8) & ~7) - (size + len) Currently, we pass A and -B to that macro instead of passing A - B and 0, where A - B is already stored in the `per_entry` variable, ready to be used. This is neither necessary, nor intuitive. Let's fix this, and have code that is both easier to read and that also does not trigger MSVC's warning. While at it, we take care of reporting overflows (which are unlikely, but hey, defensive programming is good!). We _also_ take pains of casting the unsigned value to signed: otherwise, the signed operand (i.e. the `-1`) would be cast to unsigned before doing the arithmetic. Helped-by: Denton Liu <liu.denton@gmail.com> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-04 15:09:26 +00:00
return index_pos_to_insert_pos(nr);
if (lov != hiv) {
/*
* At this point miv could be equal
* to hiv (but sha1 could still be higher);
* the invariant of (mi < hi) should be
* kept.
*/
mi = (nr - 1) * (miv - lov) / (hiv - lov);
if (lo <= mi && mi < hi)
break;
BUG("assertion failed in binary search");
}
}
}
do {
int cmp;
cmp = hashcmp(fn(mi, table), hash);
if (!cmp)
return mi;
if (cmp > 0)
hi = mi;
else
lo = mi + 1;
mi = lo + (hi - lo) / 2;
} while (lo < hi);
msvc: avoid using minus operator on unsigned types MSVC complains about this with `-Wall`, which can be taken as a sign that this is indeed a real bug. The symptom is: C4146: unary minus operator applied to unsigned type, result still unsigned Let's avoid this warning in the minimal way, e.g. writing `-1 - <unsigned value>` instead of `-<unsigned value> - 1`. Note that the change in the `estimate_cache_size()` function is needed because MSVC considers the "return type" of the `sizeof()` operator to be `size_t`, i.e. unsigned, and therefore it cannot be negated using the unary minus operator. Even worse, that arithmetic is doing extra work, in vain. We want to calculate the entry extra cache size as the difference between the size of the `cache_entry` structure minus the size of the `ondisk_cache_entry` structure, padded to the appropriate alignment boundary. To that end, we start by assigning that difference to the `per_entry` variable, and then abuse the `len` parameter of the `align_padding_size()` macro to take the negative size of the ondisk entry size. Essentially, we try to avoid passing the already calculated difference to that macro by passing the operands of that difference instead, when the macro expects operands of an addition: #define align_padding_size(size, len) \ ((size + (len) + 8) & ~7) - (size + len) Currently, we pass A and -B to that macro instead of passing A - B and 0, where A - B is already stored in the `per_entry` variable, ready to be used. This is neither necessary, nor intuitive. Let's fix this, and have code that is both easier to read and that also does not trigger MSVC's warning. While at it, we take care of reporting overflows (which are unlikely, but hey, defensive programming is good!). We _also_ take pains of casting the unsigned value to signed: otherwise, the signed operand (i.e. the `-1`) would be cast to unsigned before doing the arithmetic. Helped-by: Denton Liu <liu.denton@gmail.com> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-10-04 15:09:26 +00:00
return index_pos_to_insert_pos(lo);
}
int bsearch_hash(const unsigned char *sha1, const uint32_t *fanout_nbo,
const unsigned char *table, size_t stride, uint32_t *result)
{
uint32_t hi, lo;
hi = ntohl(fanout_nbo[*sha1]);
lo = ((*sha1 == 0x0) ? 0 : ntohl(fanout_nbo[*sha1 - 1]));
while (lo < hi) {
unsigned mi = lo + (hi - lo) / 2;
int cmp = hashcmp(table + mi * stride, sha1);
if (!cmp) {
if (result)
*result = mi;
return 1;
}
if (cmp > 0)
hi = mi;
else
lo = mi + 1;
}
if (result)
*result = lo;
return 0;
}