- added str_hash.[ch] for fast evaluation of str_len() and str_hash()

- added a tester/benchmark str-test.c, it is not compiled by default

diff --git a/lib/Makefile b/lib/Makefile
index 01f7963428dd7b65620c7e76ccc9c80e3d6d99a4..2d96bb1d90bc6aea1928b80121c9efe214ddd3b6 100644 (file)
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -8,7 +8,7 @@ SHLIB_OBJS=alloc.o alloc_str.o ctmatch.o db.o fastbuf.o fb-file.o fb-mem.o lists
        prime.o random.o realloc.o regex.o timer.o url.o wildmatch.o \
        wordsplit.o str_ctype.o str_upper.o bucket.o conf.o object.o sorter.o \
        finger.o proctitle.o ipaccess.o profile.o bitsig.o randomkey.o \
        prime.o random.o realloc.o regex.o timer.o url.o wildmatch.o \
        wordsplit.o str_ctype.o str_upper.o bucket.o conf.o object.o sorter.o \
        finger.o proctitle.o ipaccess.o profile.o bitsig.o randomkey.o \

-       hash-string.o hash-istring.o custom.o base224.o
+       hash-string.o hash-istring.o custom.o base224.o str_hash.o

 
 obj/lib/libsh.a: $(addprefix obj/lib/,$(SHLIB_OBJS))
 
 
 obj/lib/libsh.a: $(addprefix obj/lib/,$(SHLIB_OBJS))
 


@@ -20,3 +20,4 @@ obj/lib/sort-test: obj/lib/sort-test.o obj/lib/libsh.a
 obj/lib/lfs-test: obj/lib/lfs-test.o obj/lib/libsh.a
 obj/lib/regex-test: obj/lib/regex-test.o obj/lib/libsh.a
 obj/lib/hash-test: obj/lib/hash-test.o obj/lib/libsh.a
 obj/lib/lfs-test: obj/lib/lfs-test.o obj/lib/libsh.a
 obj/lib/regex-test: obj/lib/regex-test.o obj/lib/libsh.a
 obj/lib/hash-test: obj/lib/hash-test.o obj/lib/libsh.a

+obj/lib/str-test: obj/lib/str-test.o obj/lib/libsh.a

diff --git a/lib/str-test.c b/lib/str-test.c
new file mode 100644 (file)
index 0000000..adac0fb
--- /dev/null
+++ b/lib/str-test.c
@@ -0,0 +1,113 @@
+/*
+ *     Checking the correctness of str_len() and str_hash() and proving, that
+ *     it is faster than the classical version ;-)
+ */
+
+#include "lib/str_hash.h"
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/time.h>
+
+/* It will be divided by (10 + strlen()).  */
+#define TEST_TIME      1000000
+
+static void
+random_string(char *str, int len)
+{
+       int i;
+       for (i=0; i<len; i++)
+               str[i] = random() % 255 + 1;
+       str[len] = 0;
+}
+
+static uns
+elapsed_time(void)
+{
+       static struct timeval last_tv, tv;
+       uns elapsed;
+       gettimeofday(&tv, NULL);
+       elapsed = (tv.tv_sec - last_tv.tv_sec) * 1000000 + (tv.tv_usec - last_tv.tv_usec);
+       last_tv = tv;
+       return elapsed;
+}
+
+int
+main(void)
+{
+       char *strings[] = {
+               "",
+               "a",
+               "aa",
+               "aaa",
+               "aaaa",
+               "aaaaa",
+               "aaaaaa",
+               "aaaaaaa",
+               "aaaaaaaa",
+               "aaaaaaaaa",
+               "aaaaaaaaaa",
+               "\200aaaa",
+               "\200",
+               "\200\200",
+               "\200\200\200",
+               "\200\200\200\200",
+               "\200\200\200\200\200",
+               "kelapS treboR",
+               "Robert Spalek",
+               "uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu",
+               "********************************",
+               "****************************************************************",
+               NULL
+       };
+       int lengths[] = {
+               0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
+               11, 12, 13, 14, 15, 16, 17, 18, 19, 20,
+               30, 40, 50, 60, 70, 80, 90, 100,
+               200, 300, 400, 500, 600, 700, 800, 900, 1000,
+               2000, 4000, 8000, 16000, 32000, 64000,
+               -1
+       };
+       int i;
+       for (i=0; strings[i]; i++)
+               if (strlen(strings[i]) != str_len(strings[i]))
+                       die("Internal error on string %d", i);
+       printf("%d strings tested OK\n", i);
+       for (i=0; strings[i]; i++)
+               printf("hash %2d = %08x\n", i, str_hash(strings[i]));
+       for (i=0; lengths[i] >= 0; i++)
+       {
+               char str[lengths[i] + 1];
+               uns count = TEST_TIME / (lengths[i] + 10);
+               uns el1 = 0, el2 = 0, elh = 0;
+               uns tot1 = 0, tot2 = 0, hash = 0;
+               uns j;
+               for (j=0; j<count; j++)
+               {
+                       random_string(str, lengths[i]);
+                       elapsed_time();
+                       /* Avoid "optimizing" by gcc, since the functions are
+                        * attributed as ((const)).  */
+                       tot1 += strlen(str);
+                       el1 += elapsed_time();
+                       tot2 += str_len(str);
+                       el2 += elapsed_time();
+                       hash ^= str_hash(str);
+                       elh += elapsed_time();
+               }
+               if (tot1 != tot2)
+                       die("Internal error during test %d", i);
+               printf("Test %d: strlen = %d, passes = %d, classical = %d usec, speedup = %.4f\n",
+                       i, lengths[i], count, el1, (el1 + 0.) / el2);
+               printf("\t\t total hash = %08x, hash time = %d usec\n", hash, elh);
+       }
+/*
+       printf("test1: %d\n", hash_modify(10000000, 10000000, 99777555));
+       printf("test1: %d, %d\n", i, hash_modify(i, lengths[i-2], 99777333));
+       printf("test1: %d, %d\n", i, hash_modify(lengths[i-2], i, 99777333));
+       printf("test1: %d,%d,%d->%d\n", i, i*3-2, i*i, hash_modify(4587, i*3-2, i*i));
+       printf("test1: %d\n", hash_modify(lengths[5], 345, i));
+*/
+       return 0;
+}

diff --git a/lib/str_hash.c b/lib/str_hash.c
new file mode 100644 (file)
index 0000000..4431015
--- /dev/null
+++ b/lib/str_hash.c
@@ -0,0 +1,93 @@
+/*
+ *     Hyper-super-meta-alt-control-shift extra fast str_len() and str_hash()
+ *     routines
+ *
+ *     It is always at least as fast as the classical strlen() routine and for
+ *     strings longer than 100 characters, it is substantially faster.
+ *
+ *     (c) 2002, Robert Spalek <robert@ucw.cz>
+ */
+
+#include "lib/lib.h"
+#include "lib/str_hash.h"
+
+/* The number of bits the hash (in the function str_hash()) is rotated after
+ * every pass by.  It should be prime with the word size.  */
+#define        SHIFT_BITS      5
+
+/* A bit-mask which clears higher bytes than a given threshold.  */
+static uns mask_higher_bits[sizeof(uns)];
+
+static void CONSTRUCTOR
+str_hash_init(void)
+{
+       uns i, j;
+       char *str;
+       for (i=0; i<sizeof(uns); i++)
+       {
+               str = (char *) mask_higher_bits + i;
+               for (j=0; j<i; j++)
+                       str[j] = -1;
+               for (j=i; j<sizeof(uns); j++)
+                       str[j] = 0;
+       }
+}
+
+static inline uns str_len_uns(uns x) __attribute__((const));
+
+static inline uns
+str_len_uns(uns x)
+{
+       const uns sub = ((uns) -1) / 0xff;
+       const uns and = sub * 0x80;
+       uns a, i;
+       char *bytes;
+       a = (x ^ (x - sub)) & and;
+       /* 
+        * x_2 = x - 0x01010101;
+        * x_3 = x ^ x_2;
+        * a = x_3 & 0x80808080;
+        *
+        * If none byte of x is in {0, 0x80}, then the highest bit of each byte
+        * of x_2 is the same as of x.  Hence x_3 has all these highest bits
+        * cleared.  If a == 0, then we are sure there is no zero byte in x.
+        */
+       if (!a)
+               return sizeof(uns);
+       bytes = (char *) &x;
+       for (i=0; i<sizeof(uns) && bytes[i]; i++);
+       return i;
+}
+
+uns
+str_len(const char *str)
+{
+       const uns *u = (const uns *) str;
+       uns len = 0;
+       while (1)
+       {
+               uns l = str_len_uns(*u++);
+               len += l;
+               if (l < sizeof(uns))
+                       return len;
+       }
+}
+
+uns
+str_hash(const char *str)
+{
+       const uns *u = (const uns *) str;
+       uns hash = 0;
+       while (1)
+       {
+               uns last_len = str_len_uns(*u);
+               hash = ROL(hash, SHIFT_BITS);
+               if (last_len < sizeof(uns))
+               {
+                       uns tmp = *u & mask_higher_bits[last_len];
+                       hash ^= tmp;
+                       return hash;
+               }
+               hash ^= *u++;
+       }
+}

diff --git a/lib/str_hash.h b/lib/str_hash.h
new file mode 100644 (file)
index 0000000..2c1c0d7
--- /dev/null
+++ b/lib/str_hash.h
@@ -0,0 +1,14 @@
+/*
+ *     Hyper-super-meta-alt-control-shift extra fast str_len() and str_hash()
+ *     routines
+ *
+ *     (c) 2002, Robert Spalek <robert@ucw.cz>
+ */
+
+#include "lib/lib.h"
+
+/* An equivalent of the Intel's rol instruction.  */
+#define        ROL(x, bits)    (((x) << (bits)) | ((x) >> (sizeof(uns)*8 - (bits))))
+
+uns str_len(const char *str) __attribute__((const));
+uns str_hash(const char *str) __attribute__((const));