aboutsummaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorMattias Andrée <maandree@kth.se>2016-05-04 21:04:52 +0200
committerMattias Andrée <maandree@kth.se>2016-05-04 21:04:52 +0200
commitfbace74784b115d24441d2a67b932a96011d7012 (patch)
tree9adaaccb875eccf53ca415777f626bae8d9a5942 /src
parentOptimise znot (diff)
downloadlibzahl-fbace74784b115d24441d2a67b932a96011d7012.tar.gz
libzahl-fbace74784b115d24441d2a67b932a96011d7012.tar.bz2
libzahl-fbace74784b115d24441d2a67b932a96011d7012.tar.xz
Optimise zlsh
Signed-off-by: Mattias Andrée <maandree@kth.se>
Diffstat (limited to 'src')
-rw-r--r--src/internals.h47
-rw-r--r--src/zlsh.c7
2 files changed, 44 insertions, 10 deletions
diff --git a/src/internals.h b/src/internals.h
index a475ada..0690ad3 100644
--- a/src/internals.h
+++ b/src/internals.h
@@ -4,6 +4,7 @@
#include <errno.h>
#include <stdlib.h>
#include <string.h>
+#include <unistd.h>
/* clang pretends to be GCC... */
#if defined(__GNUC__) && defined(__clang__)
@@ -103,7 +104,6 @@ extern void *libzahl_temp_allocation;
#define zpositive1(a, b) (zpositive(a) + zpositive(b) > 0)
#define zpositive2(a, b) (zsignum(a) + zsignum(b) == 2)
#define zzero2(a, b) (!(zsignum(a) | zsignum(b)))
-#define zmemmove(d, s, n) memmove((d), (s), (n) * sizeof(zahl_char_t))
#define zmemcpy(d, s, n) libzahl_memcpy(d, s, n)
#define zmemset(a, v, n) libzahl_memset(a, v, n)
@@ -245,12 +245,8 @@ static inline void
zswap_tainted_unsigned(z_t a, z_t b)
{
z_t t;
- t->used = b->used;
- b->used = a->used;
- a->used = t->used;
- t->chars = b->chars;
- b->chars = a->chars;
- a->chars = t->chars;
+ SWAP(a, b, t, used);
+ SWAP(b, a, t, chars);
}
static inline void
@@ -398,3 +394,40 @@ zfree_temp(z_t a)
a__[i__] = OP(b__[i__]); \
} \
} while (0)
+
+static inline void
+zmemcpyb(register zahl_char_t *restrict d, register const zahl_char_t *restrict s, size_t n_)
+{
+ ssize_t i, n = (ssize_t)n_;
+ switch (n & 3) {
+ case 3:
+ d[n - 1] = s[n - 1];
+ d[n - 2] = s[n - 2];
+ d[n - 3] = s[n - 3];
+ break;
+ case 2:
+ d[n - 1] = s[n - 1];
+ d[n - 2] = s[n - 2];
+ break;
+ case 1:
+ d[n - 1] = s[n - 1];
+ break;
+ default:
+ break;
+ }
+ for (i = n & ~3; (i -= 4) >= 0;) {
+ d[i + 3] = s[i + 3];
+ d[i + 2] = s[i + 2];
+ d[i + 1] = s[i + 1];
+ d[i + 0] = s[i + 0];
+ }
+}
+
+static inline void
+zmemmove(register zahl_char_t *d, register const zahl_char_t *s, size_t n)
+{
+ if (d < s)
+ zmemcpy(d, s, n);
+ else
+ zmemcpyb(d, s, n);
+}
diff --git a/src/zlsh.c b/src/zlsh.c
index 1c9fd8f..59a05bf 100644
--- a/src/zlsh.c
+++ b/src/zlsh.c
@@ -18,10 +18,11 @@ zlsh(z_t a, z_t b, size_t bits)
cbits = BITS_PER_CHAR - bits;
ENSURE_SIZE(a, b->used + chars + 1);
- if (likely(a == b))
- zmemmove(a->chars + chars, b->chars, b->used);
- else
+ if (likely(a == b)) {
+ zmemcpyb(a->chars + chars, b->chars, b->used);
+ } else {
zmemcpy(a->chars + chars, b->chars, b->used);
+ }
zmemset(a->chars, 0, chars);
a->used = b->used + chars;