@matheusbsbs / matheus-bagliano / commits / 43fe901b71
Fork 0

compat: Add simplified merge sort implementation from glibc

qsort in Windows 2000 (and various other C libraries) is a Quicksort with the usual O(n^2) worst case. Unfortunately, sorting Git trees seems to get very close to that worst case quite often: $ /git/gitbad runstatus # On branch master qsort, nmemb = 30842 done, 237838087 comparisons. This patch adds a simplified version of the merge sort that is glibc's qsort(3). As a merge sort, this needs a temporary array equal in size to the array that is to be sorted, but has a worst-case performance of O(n log n). The complexity that was removed is: * Doing direct stores for word-size and -aligned data. * Falling back to quicksort if the allocation required to perform the merge sort would likely push the machine into swap. Even with these simplifications, this seems to outperform the Windows qsort(3) implementation, even in Windows XP (where it is "fixed" and doesn't trigger O(n^2) complexity on trees). [jes: moved into compat/qsort.c, as per Johannes Sixt's suggestion] [bcd: removed gcc-ism, thanks to Edgar Toernig. renamed make variable per Junio's comment.] Signed-off-by: Brian Downing <bdowning@lavos.net> Signed-off-by: Steffen Prohaska <prohaska@zib.de> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>

Brian Downing committed Feb 5, 2008 at 15:10 UTC 43fe901b71f81cdff142048dfc27e492c445d225
3 files changed +76
Makefile
+8
@@ -137,6 +137,10 @@ all::
137 # Define THREADED_DELTA_SEARCH if you have pthreads and wish to exploit
138 # parallel delta searching when packing objects.
139 #
140 +# Define INTERNAL_QSORT to use Git's implementation of qsort(), which
141 +# is a simplified version of the merge sort used in glibc. This is
142 +# recommended if Git triggers O(n^2) behavior in your platform's qsort().
143 +#
144
145 GIT-VERSION-FILE: .FORCE-GIT-VERSION-FILE
146 @$(SHELL_PATH) ./GIT-VERSION-GEN
@@ -722,6 +726,10 @@ ifdef NO_MEMMEM
726 COMPAT_CFLAGS += -DNO_MEMMEM
727 COMPAT_OBJS += compat/memmem.o
728 endif
729 +ifdef INTERNAL_QSORT
730 + COMPAT_CFLAGS += -DINTERNAL_QSORT
731 + COMPAT_OBJS += compat/qsort.o
732 +endif
733
734 ifdef THREADED_DELTA_SEARCH
735 BASIC_CFLAGS += -DTHREADED_DELTA_SEARCH
compat/qsort.c new
+62
@@ -0,0 +1,62 @@
1 +#include "../git-compat-util.h"
2 +
3 +/*
4 + * A merge sort implementation, simplified from the qsort implementation
5 + * by Mike Haertel, which is a part of the GNU C Library.
6 + */
7 +
8 +static void msort_with_tmp(void *b, size_t n, size_t s,
9 + int (*cmp)(const void *, const void *),
10 + char *t)
11 +{
12 + char *tmp;
13 + char *b1, *b2;
14 + size_t n1, n2;
15 +
16 + if (n <= 1)
17 + return;
18 +
19 + n1 = n / 2;
20 + n2 = n - n1;
21 + b1 = b;
22 + b2 = (char *)b + (n1 * s);
23 +
24 + msort_with_tmp(b1, n1, s, cmp, t);
25 + msort_with_tmp(b2, n2, s, cmp, t);
26 +
27 + tmp = t;
28 +
29 + while (n1 > 0 && n2 > 0) {
30 + if (cmp(b1, b2) <= 0) {
31 + memcpy(tmp, b1, s);
32 + tmp += s;
33 + b1 += s;
34 + --n1;
35 + } else {
36 + memcpy(tmp, b2, s);
37 + tmp += s;
38 + b2 += s;
39 + --n2;
40 + }
41 + }
42 + if (n1 > 0)
43 + memcpy(tmp, b1, n1 * s);
44 + memcpy(b, t, (n - n2) * s);
45 +}
46 +
47 +void git_qsort(void *b, size_t n, size_t s,
48 + int (*cmp)(const void *, const void *))
49 +{
50 + const size_t size = n * s;
51 + char buf[1024];
52 +
53 + if (size < sizeof(buf)) {
54 + /* The temporary array fits on the small on-stack buffer. */
55 + msort_with_tmp(b, n, s, cmp, buf);
56 + } else {
57 + /* It's somewhat large, so malloc it. */
58 + char *tmp = malloc(size);
59 + msort_with_tmp(b, n, s, cmp, tmp);
60 + free(tmp);
61 + }
62 +}
git-compat-util.h
+6
@@ -426,4 +426,10 @@ static inline int strtol_i(char const *s, int base, int *result)
426 return 0;
427 }
428
429 +#ifdef INTERNAL_QSORT
430 +void git_qsort(void *base, size_t nmemb, size_t size,
431 + int(*compar)(const void *, const void *));
432 +#define qsort git_qsort
433 +#endif
434 +
435 #endif