I have a string which is generated by linux uuid generation code (libc):
1b4e28ba-2fa1-11d2-883f-b9a761bde3fb
I need to replace some of the characters in this string:
- with _
2 with f
4 with x
I am generating the 200 UUIDs using a loop.
So for every uuid I need to replace using a custom function, so that function must be maximum optimised to do so, how can I attain that?
I suppose you're using char[] str
char *c;
for(c = str; *c != '\0'; ++c){
if( *c == '-' ) *c = '_';
else if( *c == '2' ) *c = 'f';
else if( *c == '4' ) *c = 'x';
}
switch version
char *c;
for(c = str; *c != '\0'; ++c){
switch(*c){
case '-': *c = '_'; break;
case '2': *c = 'f'; break;
case '4': *c = 'x'; break;
}
}
switch as I said in comment above instead of if/else if would be more readable, but it likely compiles to the same or very similar code.
-O3 in your compiler argument?Is something as trivial as this what you want?
void my_replace(char* str)
{
while (*str) {
switch (*str) {
case '-':
*str = '_';
break;
case '2':
*str = 'f';
break;
case '4':
*str = 'x';
break;
default:
break;
}
++str;
}
}
It is really fast and simple. I really can't see how you can make it faster.
EDIT: I am aware of some optimizations in certain string operations, but I do not see how they can be applicable here. For example, in the case of memcpy, one may copy 4 or more bytes at a time, depending on the processor. In case of comparing strings that is properly aligned, comparing as integers may be possible and more efficient. I just can't see an applicable technique.
memcpy, strnlen, and other functions. In fact, several libc implementations do employ optimisations like this.C library functions may be optimized and significantly faster than a hand-coded iteration.
char* uuid; // = ...
// size_t uuid_len; // = ... length of uuid
char* ptr = strpbrk(uuid, "-24");
while (ptr)
{
switch(*ptr)
{
case '-':
*ptr = '_';
break;
case '2':
*ptr = 'f';
break;
case '4':
*ptr = 'x';
break;
}
// if (ptr-uuid == uuid_len) break;
ptr = strpbrk(ptr+1, "-24");
}
Edit: Took out the range checking, on the basis of the example here that doesn't seem necessary.
Edit: So I decided to test out the 3 algorithms here to see which is faster. I had a loop of 100000 strings, on a vintage 2006 Mac Pro, compiling with gcc, -O3. I took the average of 1000 runs and did 5 cycles.
@johnchen by a hair with an average time of 7.85ms.
@YongweiWu just behind with an average time with 7.89ms. The difference looks significant; going in and doing a proper statistical test is not going to happen tonight unfortunately. :)
...and strpbrk a distant third at 32ms. (Glad I qualified all my optimization claims with 'might', 'may' etc....)
Edit: There is a big difference with Clang--j @ WY's algorithms take 10ms under Clang (looks tied between them), mine is unchanged.
strpbrk restarts at the next character, it does not start over at the beginning.