How Difficult is This?

Hi:

I've been programming VB, C#, php, Javascript, etc. for about 15 years now.
I'm working in C# with string searches and finding that it's slow as a sleeping hound dog.

I did some research and found this:
Fast string search

It turns out that basic C code blows the doors off of C# (like 20 to 1) with string searches.

So, how hard would it be for me to learn how to write stuff using strstr in C, given my experience? I don't know jack about pointers, memory leaks, garbage collection, etc.

About how much time can I expect to study to learn how to do it, all the way through compile and use?

Also, is there any way to access it from C# without C# turning it into a dog?

Why not try to implement something like Boyer-Moore in C# then?

Quote:

---

Originally Posted by MAtkins

About how much time can I expect to study to learn how to do it, all the way through compile and use?

Also, is there any way to access it from C# without C# turning it into a dog?

---

So, basically you're asking how long it will take you to learn C... With your experience, you should be able to pour through a good tutorial or book in a couple of days and be writing fairly simple but useful proggies within a week. With the right information, you should be able to write DLLs that can be accessed by C# within a couple of weeks.

Since C produces native machine code, C# can't turn it into a dog... it's far more likely you C language helpers will simply show C# up for the slow-mo puppy it is...

So you're gonna take the advice of a flawed 8 year-old benchmark? I hope you realise that the Java and C# datasets are twice as large as the C/C++ one. Sifting through less data is always going to be faster.

His results don't exactly hold true now anyway. My results for the C/C++ tests (AMD 64-bit dual-core 4GB ram / 32-bit program compiled with the VS2005 SP1 compiler):

Code:

---

the alg ids
cruntime=0,
naive, = 1
kmp = 2,
boyermoore = 3
C:\Users\Adrian\Downloads\fastsearch\fastsearch\fastsearch>fastsearch
duration for alg(0) test suite : 78 msecs
duration for alg(1) test suite : 125 msecs
duration for alg(2) test suite : 172 msecs
duration for alg(3) test suite : 62 msecs

C:\Users\Adrian\Downloads\fastsearch\fastsearch\fastsearch>fastsearch
duration for alg(0) test suite : 78 msecs
duration for alg(1) test suite : 124 msecs
duration for alg(2) test suite : 156 msecs
duration for alg(3) test suite : 63 msecs

C:\Users\Adrian\Downloads\fastsearch\fastsearch\fastsearch>fastsearch
duration for alg(0) test suite : 63 msecs
duration for alg(1) test suite : 124 msecs
duration for alg(2) test suite : 172 msecs
duration for alg(3) test suite : 62 msecs

---

>So, how hard would it be for me to learn how to write stuff using strstr in C, given my experience?
Well, it all really depends on you. Is this something you will have to do? If so there is a deadline and the pressure which will make you learn. If you have the enthusiasm for it, you wouldn't take too long at all.

Why do you have to do this in C, if suppose C# is much better in string handling than C.

ssharish

I would take that string test with a grain of salt.

Yes, C is faster than C#, at almost everything except coding it, but I have real trouble believing C# is that bad.

Also, the "test" string was heavily populated by very short strings, like "the" (323 times). That heavily favors strstr() and hurts the more elegant Boyer-Moore and etc., algorithms.

Why don't you post up a little sample of the data you have to search through, and the strings you need to find within that data, and let's test it.

Testing YOUR data, with different algorithms, (at least strstr() and B-M), (and here, languages of C# and C), is the only way to know for sure, what the facts are. There's no substitute for actual head to head testing on your data, and on your system (if possible).

thanks for the replies:

I've got a text file with 300K lines having 1 URL per line.
I want to append it with 40K new URLs - BUT - I don't want to duplicate any of the urls that are already in the original file.
I can do it 5 different ways but they're all dogs.
I'm basically looping through the 40K and in that looping through the 300K doing an IndexOf or == for each.
That would be 40,000 * 300,000 loops total - it takes all day!

I tried just grabbing the FileText for the 300K and looping through the 40K using an FileText.IndexOf(line) to see if the URL is there.
Really inefficient I think.

I've got an HMTL page with 2500 lines of HTML code in it.
I want to find out if "This Text" can be found in the HTML code.
I should add that a list of 'This Text' possibilities are stored in a text file.
So, I end up basically asking is "This Text A' in HTML if not then is 'This Text B' in HTMLif not then is 'This Text C' in HTML, etc.
I'm using for loops and IndexOf to do it - painfully slow.

Also, I want to extract a string that represents the first (or sometimes second) entire HTML form within the HTML.

I want to search for string Needle in string Haystack and get the INDEX of the first character of the Needle.

I know at least one program that can do stuff like this at least 20 times faster than I can in C#. I've tried as many techniques as I can find including using a MySQL database. All of them are dogs. They take for ever. What the other program can do in 5 seconds, my app takes minutes to do.

I'd really like to learn better ways of doing this kind of work.
I'm convinced that *much* better ways to do it exist.
I don't care if it's in C or C++ or C# or PHP for that matter.

I did get a c dll to be accessible from my C# app today.
I'm fighting with strings . . .

> I've got a text file with 300K lines having 1 URL per line.
Is it sorted?
If it were, then you can use a binary search to locate any string (or know it isn't there) in at most 19 comparisons.

Databases are going to be waaaayyyy too slow. You need to use some internal structure of your own for these problems.

URL File:
Read the 300k URLs into a fast insert/access data structure, like a binary tree. Read the 40k URLs, each time doing a lookup of that URL in the tree, and if it misses, insert the URL and append it to the file. GLib has some great data structures (including a balanced binary tree) and can be used with Windows (I'm a Linux guy though, so there may be better tools out there that I don't know of). Or you can roll your own if you want some extra practice. Balanced trees get tricky, but you might not need them in your case.

HTML:
Look at a generic XML parser. The only one I know of is Expat. There may bit of a learning curve, but it's pretty speedy and powerful.

Needle/haystack:
strstr with a simple pointer subtraction is the easiest way.

Quote:

---

I've got a text file with 300K lines having 1 URL per line.

---

I don't see any data posted up, AND I don't see a link on Swoopshare (or any other free file sharing site), either:
swoopshare - Save files on the web

So, you're probably lying! :p :p :D But why?

Quote:

---

I want to append it with 40K new URLs - BUT - I don't want to duplicate any of the urls that are already in the original file.

I can do it 5 different ways but they're all dogs.

---

Yes -- but why dogs - inquiring minds want to know! :D

Seriously, post up a little sample, and give it a test on your best C# program for time. We'll do the same with some C code, on those same test samples. Then, you'll have some facts to make a decision about.

The test doesn't have to be on a full size file, just a representative sample that takes maybe 15 to 60 seconds to run on C#, should be fine.

It's not uncommon for us to beat the pants off such samples, either by using C 's inherent speed, or by using a better algorithm, or both.

Otherwise my goons will have to escort you right up to Father O'Reilly for confession. I've got a kickback agreement with him. Figure I've cut off 10 years from purgatory this way.

:p :p

Adak: I'm not interested in testing yet. I *know* that what I've got doesn't work. It's not like I've got a 'one case' scenario here. I"ve gotta do it day in and day out.
Finding 40K of urls in a list or an HTML page w/ 2500 lines is nothing to accomplish. I'm looking for good programming techniques, once I think I've got something I'll be happy to test.

anduril462: Great stuff. Thanks. How do I learn more about binary trees? Or are there any better ideas for that job?
Most often, I've got to compare, not just the lines but the root url. For example:
http://cboard.cprogramming.com/newre...te=1&p=1001614
I need to not duplicate any URLs that include:
cboard.cprogramming.com

I don't think an xml parser will be that great. Generally, I'm looking for text and/or code anywhere within the html. Sometimes I strip the html tags first and then search just the text. The text I'm looking for may be part of the html code itself or text or a combination mixed together. A classic would be 'Powered By <a href='http://www.wordpress.com'>WordPress</a>. I may be looking for the text, the url or both. Currently, my list of search params is about 175 different patterns.

I've been working with c & strings all day. Pointers & strstr keep returning the actual text (by printing what is at the memory location) or the ASCII value, but not the actual location. I've looked high and low for it and can't find the code I need.
I'm looking for,
Idx = (ptr_Mem_Location_of_Needle - ptr_Mem_Location_of_Haystack)
correct? How do I get the actual memory location from the result of strstr?

This works:
I'm getting a warning about 'assignment of incompatible pointer type' though.
Can't figure out what a pointer to the location of a char should be I guess.

Code:

---

int main()
{
        char str1[]="Testing a string of text";
        char str2[]="esting";
        char *ptr1;
        char *ptr2;       
        int idx;
       
        ptr1 = &str1;         
    ptr2 = strstr(str1,str2);
        idx = ptr2 - ptr1;
        printf( "\nptr1:%d ptr2:%d Idx=%d\n", ptr1,ptr2,idx);

  return 0;
}

---

This doesn't work. I can't figure out how to get the strings into the indexOf function:

Code:

---

#include <stdio.h>
#include <string.h>
int indexOf(char *str1, char *str2)
{
        char *ptr1;
        char *ptr2;
        int idx;
    ptr1 = &str1;         
    ptr2 = strstr(str1,str2);
        idx = ptr2 - ptr1;
        printf( "\nptr1:%d ptr2:%d Idx=%d\n", ptr1,ptr2,idx);
        return idx;
}
int main()
{
        char str1[]="Testing a string of text";
        char str2[]="esting";       
        int idx;
       
        idx = indexOf(str1,str2);
        printf( "\nIdx=%d\n", idx);

  return 0;
}

---

This works now (I removed the &) but I see 2 potential problems:
1. How do I use something like this with UTF-8 encoding?
2. Can I throw a 2500 line (probably more like 500) HTML file at it?

Code:

---

#include <stdio.h>
#include <string.h>
int indexOf(char *str1, char *str2)
{
        char *ptr1;
        char *ptr2;
        int idx;
    ptr1 = str1;         
    ptr2 = strstr(str1,str2);
        idx = ptr2 - ptr1;
        if(idx < 0) idx = -1;
        printf( "\nptr1:%d ptr2:%d Idx=%d\n", ptr1,ptr2,idx);       
        return idx;
}
int main()
{
        char str1[]="Testing a string of text";
        char str2[]="Festing";       
        int idx;
       
        idx = indexOf(str1,str2);
        printf( "\nIdx=%d\n", idx);

  return 0;
}

---

well, I asked the girl who wrote the 'fast' app how she did it . . .
Assembly Language through Delphi.
That will only handly ASCII though won't it?
I thought I understood that strstr was about as close to Assembly as you could get.
Yes? No? Ideas?

Most C compilers these days support inline assembly... so that shouldn't stop you.

Most of the standard C99 functions are likely to be optimized using either inline or pure assembly. We don't actually know what's "under the hood" but given that it's nearly impossible to match the library's performance with C code, I'm guessing it's a safe assumption.

And yes... if her code is written for 8bit char it's not going to support Unicode... But that's also true of the C library functions. If you note there are two versions of strstr... strstr() and wcsstr() ... one is in string.h the other in wchar.h... one is for ascii and the other for unicode.

Do you have example data files you could upload somewhere, as Adak suggested?

Your example programs do not show us the problem, only your constrained thinking of what you think the answer should look like.

OK, np.

Here is a master list of URLs:
swoopshare - Download TestMasterList.txt


Here is a test search list:
swoopshare - Download TestSearchList.txt

Here is a test html page:
swoopshare - Download TestHTMLPage.htm

generally, I need make sure the base domain isn't duped, not just the whole url.

One thing I might look for would be:
powered by ShrinkTheWeb. (notice that ShrinkTheWeb is under an a tag).
Sometimes I'd want the URL, sometimes just to know the TEXT says 'powered by ShrinkTheWeb'

I hope this helps.
As I mentioned, just a long list (sometimes 300K) compared with (usually) a shorter list and I need to remove the dupes before adding the shorter list to the longer one.

Right now I'm studying Binary Trees as a possible solution for the URL duping problem.
I wonder if strstr would be faster or a Binary Tree?

I need basic string search capabilities that will work with UTF-8
.indexOf(char *Haystack, char *Needle) <- would return the index of Needle within Haystack

.indexOfCI(char *Haystack, char *Needle) <- same thing but case insensitive

.subStr(char *Str, int nStart) <- creates a substring of Str but shaves everything off before the INDEX of nStar

.subStr(char *Str, int nStart, nLength) <- same thing but specifies the total length of the string returned

.strReplaceFirst(char *Str, char *strSrch, char *strReplace) <- replaces first case of strSrch w/ strReplace in string Str

.strReplace(char *Str, char *strSrch, char *strReplace) <- replaces all cases of strSrch w/ strReplace in string Str

I'm working on these but I have to admit the whole 'pointer' thing is kicking my butt.
I can't seem to find clear docs on nomenclature. I'm getting better though.
I'm really surprised these funcs aren't in a readily available library already (or are they?).
Of course the only reason I want them in C is because I need them to be FAST

The duplicate checking part is what I'm concentrating on.

<< Big Evil Grin >> :p :p :p

Given your two data files, I can do the search of the 2nd list in the first list in about 0.25 seconds.

Code:

---

$ time ./a.out
597756 calls to comparison to sort 44337 records
Hits=91, misses=122
3216 calls to comparison to search 44337 records

real        0m0.219s
user        0m0.188s
sys        0m0.020s

---

That's a 1.6GHhz laptop and no attempt at optimisation.

I'm getting results similar to yours Salem, just using a sort and binary search, with string arrays. ***

Unless there's a problem with the actual file sizes that were uploaded, this looks to be an algorithm fix, much more than anything else.

MAtkins, can you confirm we have the right files? Have you tried using a binary search function to limit the comparisons that have to be made?

*** On my spiffy new Pelles C IDE and compiler! :D :D

Thanks, Common Tater, for the tip on Pelles C.

Quote:

---

Originally Posted by Adak

*** On my spiffy new Pelles C IDE and compiler! :D :D

Thanks, Common Tater, for the tip on Pelles C.

---

It may not happen all that often, but at least some of the time I actually do know what I'm talking about :devil:

It's a pretty cool tool... Glad you like it.

First a note on the binary search:
===============================
The heart of the program is the binary search algorithm - which is just the old "guess the number I'm thinking of" game, you probably played at some time, as a child.

Guess a number between 1 and 10:
5 (always start in the middle)

Too low: New "floor" is 6 (lowest possible valid number)
8 (6 + 10)/2 is the new "middle"

Too high: New "ceiling" is 7 (highest possible valid number)
6 (6 + 7)/2, rounded down to nearest whole number (integer).

Too low
Has to be: 7

Only 3 comparisons had to be made, instead of possibly 9. As the number of comparisons increase, binary search ("binary" because it constantly "cuts" the possible answer space, in half), just gets better and better.

The only requirement to remember, is that the binary search has to be made, on sorted data. It won't work, otherwise.

Program note:
====================
The fastest string sorter I know uses the system call:
system("sort master.txt >master1.txt");

You may not be able to use that system call. If so, we'll set you up with a fast programmed sorter, no problem.

Code:

---

#include <stdio.h>
#include <time.h>
#include <string.h>
#include <stdlib.h>

#define MSIZE 50000
#define SSIZE 215

int checkIt(char s[SSIZE][200], char **m, int i, int low, int high);

int main() {
  int i=0, isdup, mcount=0, scount=0, low, high, dupcount=0;
  char s[SSIZE][200];
  char **m;
  clock_t start, stop;
  FILE *fp;
 
  printf("\n\n");
  start = clock();
  system("sort master.txt >master1.txt");
  m=malloc(MSIZE * sizeof(char *));

  for(i=0;i<MSIZE; i++) {
    m[i] = malloc(200);
    if(m[i] == NULL) {
      printf("Error allocating memory for m[]\n");
      return 1;
    }
  }

  if((fp=fopen("master1.txt","rt")) == NULL) {
    printf("Error opening master file - terminating\n");
    return 1;
  }

  while((fgets(m[mcount], 200, fp)) != NULL) {
    ++mcount;
    if(mcount == MSIZE-1)
      break;
  }
  fclose(fp);
  if((fp=fopen("slave.txt", "rt")) == NULL) {
    printf("Error opening slave file -- terminating\n");
    return 1;
  }
  while((fgets(s[scount], 200, fp)) != NULL) {
    ++scount;
    if(scount == SSIZE-1)
      break;
  }
  fclose(fp);
  low = high = 0;
  for(i=0;i<scount; i++) {
    isdup = checkIt(s, m, i, 0, MSIZE-1);
    if(isdup) {
      dupcount++;
      //printf("Check:\n%s\n%s\n", m[isdup], s[i]);
      //getchar();
    }
  }

  stop = clock();
  printf("\nElapsed Time: %lf \n",(double)(stop-start)/CLOCKS_PER_SEC);
  printf("%d Duplicates were found in the master list\n", ++dupcount);

  for(i=0;i<MSIZE; i++) {
    free(m[i]);
  }
  free(m);
  return 0;
}
/* this is the binary search function. The low and high parameters
  aren't used, but could be, for an even faster search, by
  further decreasing the search space.
*/
int checkIt(char s[SSIZE][200], char **m, int i, int low, int high) {
  unsigned int mid;

  while(low <= high) {
    mid = (low + high)/2;
    if((strcmp(m[mid], s[i])) > 0)
      high = mid-1;
    else if((strcmp(m[mid], s[i])) < 0)
      low = mid+1;
    else
      return mid;  //duplicate found
  }
  return 0;
}

---

This is rough and untested code. Be sure to test it thoroughly - very thoroughly, before using it. You assume all risks by running it.

Why didn't you use qsort() and bsearch(), both in stdlib.h ?

I bulked up the data to 2M lines in the master file, and 10K lines in the search list.

Code:

---

$ wc -l *large.txt
  2169075 TestMasterList_large.txt
    10905 TestSearchList_large.txt
  2179980 total
$ gcc -std=c99 -g -Wall -O2 zarniwoop.c
$ ./a.out TestMasterList_large.txt TestSearchList_large.txt
Read all files: Elapsed time = 2.211

Sort master list for bsearch: Elapsed time = 3.076
38724037 calls to comparison to sort 2169075 records

Search update list in master: Elapsed time = 0.010
Hits=52, misses=10853
229535 calls to comparison to search 10905 records

Search 10% (always hits) list in master: Elapsed time = 0.684
Hits=216907, misses=0
4351053 calls to comparison to search 216907 records

---

If the master list were deliberately kept in a sorted state, that would basically half the time it takes.

In any event, the actual lookup time is basically unmeasurable noise compared to everything else. The big factor that isn't going away is reading the file in the first place.

Boy, I don't know if that's gonna be fast enough.
It took .046 seconds to run through 172984 records against 78036 records.
SHEESH! (did you hear the sarcasm?) That's scary man!!

I blinked and it was DONE!
Thanks:o)
Now I've gotta figure out how to make it do what I need (which should be cake).

I had that wrong, I named my master file Master1.txt - need to look more closely at the code.
So, I found a file w/ 1481807 records & a slave w/ 189526 records.
It took Adak's proc 5.616 seconds to run it.

I couldn't compile zarniwoop so I couldn't test that one.

Well, I found out how to compile zarniwoop w/ gcc.
I ran it from cmd line like:
zarniwoop 2 Master.txt Slave.txt

It threw a pile of errors.
Did I do that wrong?

@Salem, I was trying for the most basic code, so if MAtkins wanted to do this in C#, with his own code, he could, and learn something, too.

I learned binary search in BASIC, and just never got into the habit of using binsearch(). Half the time, I forget it's there.

For faster input he could use one of the newer SSD drives -- they're screamers! After that, (or a large buffer), you are quite right - the IO piper must be paid his due.

That program is very rough, MAtkins. Be careful that you don't run outside the limits of the array, or (low + high)/2 doesn't overflow the range of unsigned int's, on your system.

What I've done in the past, is set a good sized array (always a power of 2 and 16), and then used a BIG loop to empty and refill the array, over and over, as needed, until all the data is handled.

Of course, you can't always just make a HUGE array and expect to handle every GIANT file of data, in one fill up.

Glad you like it. Salem's version would be great for you to study. As you can see by his increased data size, it scales quite well.

Are you suggesting that in the vast bloat of C#, there is nothing to do a simple sort and search on some user defined data?

The key to this problem is picking the right algorithms and data structures.
Not attempting some simplistic "brute force" approach in a faster language.

I'm sold on doing it in C. I can pretty easily make a dll that I can access from C#.
I'm studying the code in both files.
It's pretty straightforward I think so far bu really brilliant stuff though.

So, tell me about those SSD drives . . .
I'm running Win 7 64 bit on an Alienware Aurora - overclocked at I think 3.67 Ghz Quad core - 6 Gig RAM.
It's got a radiator in it.

I didn't mention when I ran your proc I was simultaneously running 50 treads that are scraping the Internet. It ran without a blink.

Salem, I understand and agree.
I did get zarniwoop to work (I think).
I'm still not sure how to call it.
I'm typing in the command line now:
zarniwoop Master.txt Slave.txt

Is C faster than C# or not?
Partly, I'm thinking, I've got it working in C and I can access that from C# real easy now.
Why not in C?
I realized (and did from the start) that the algorithm would probably contribute the most if I could find a technique that was far superior to what I was doing.
With your help, that's done.

If I'm doing this right, zarniwoop is going through 1.48M > 260K in less than 3 seconds total.
That's with my box humming because it's running 50 threads to scrape the Internet and dogging my HD.

Bravo! That's unbelievable.
I'm still studying your code . . .

Salem:
Yea, it's doing it - with my 50 threads still running it did 1.48M > 260K in just a smidge under 2 seconds.
WHEW! - that would have taken my old proc all day:)
Thanks!

Salem:
I'm studying your code.
Why do you call bsearch the second time against 10% of the master file?
Also, with rand() you could end up searching the same rec twice yes?

Why do sometimes you dim function args as const?

SSD drives are apparently, our future:

Intel’s SSD Plans : Intel's X25-M Solid State Drive Reviewed

VERY fast, and no moving parts. Pricey, but it will come down more, no doubt.

What's the purpose of all these url's you're gathering?

@Salem: I know absolutely nothing about C#, except that it needs dot net to run. Never looked into it. Yes, brute force and the most basic but efficient algorithm, was my goal. Data structures might depend on what else he's doing.

Youza - 250Mb/s is fast. My drives read at 120/s sustained.
My main drive is 650 Gig though, they'll have to make 'em bigger.

I'm doing Internet data polling for clients, mostly for SEO work.

MAtkins... have a browse around this site... Techgage ... these guys really like SSDs and Rob Williams is a pretty good reviewer...

Well, I'm stumped on my first real edit.
I'm appending files:
TestOut1.txt with the base_url
TestOut2.txt with the full_url

I'm loading both from the same loop but in the middle of it Windows throws me an error saying the app quit.

When I look at the output files TestOut1.txt has less recs than TestOut2.txt.
TestOut2.txt looks like it quit right in the middle of a URL.

I can't tell if I ran out of memory or . . .
Can you point me in a direction so I have a clue as to how to figure out what the problem is?

I tried to attach my test files but they're pretty big . . .

> Why do you call bsearch the second time against 10% of the master file?
> Also, with rand() you could end up searching the same rec twice yes?
The whole last section is just a bit of bench-marking to see how long searching takes.
0.01 seconds for the first result is noise, so I wanted a bigger sample.

Reading the whole of the 2nd file is sub-optimal as well, since there is no need to store anything. But it was quick and easy to re-use the same code. It will save some short-term RAM use, but it won't do much to change the slowness of the file reading.

> Why do sometimes you dim function args as const?
- To stop me changing things by accident
- To give the compiler a little extra something to play with at optimisation time. Certain things which are known to be const can be optimised better.

> fprintf(fp1, strcat(db[i].base_url,"\n"));
Use
fprintf(fp1, "%s\n", db[i].base_url);

1. strcat is attempting to append a character to a string where the exact length has been allocated. This is a buffer overflow.
2. Any % characters that appear in this string will cause fprintf to look for additional (and non-existent) additional parameters.


> result[numRecords].full_url = r.full_url;
> result[numRecords].base_url = r.base_url;
> numRecords++;
At this point, you need to set your isUnique member to false as well.
realloc won't zero this memory for you.

Man, how do you guys learn this stuff!?
I had figured out that it had something to do with the base_url and not about writing the file at all.
So, fprintf works like printf then? That makes sense.

Boy, I'm feeling DUMB here -) OK, how do I set an int to false? Just make it 0? or NULL or . . .

Zero (0) will do it.

K, I did that.
For the 1.48M > 260K it took 3.4 seconds total, including writing to the 2 files.
That's just unbelievable.

Now if I can figure out how to skate through a text file that fast . . . :)

The person who wrote the article is an idiot.
Specialised algorithms like Boyer Moore a specifically designed for searching under different conditions than naieve algorithms are designed for. E.g. Longer strings to search for, more limited alphabet, high number of prefix but not whole word matches, searching through very long datasets etc. It's apples and oranges.
Using an algorithm like that to search for "the" would be like using quicksort on 5 items. Bubble sort is much faster in that case.

Surely it would have been fairer to put as much effort into optimising the C++ code as it was to whoever optimised the assembly version.

The benchmark is flawed in other ways too. Things like the switch statement having the strstr variant as the first case. Hmm, assuming the compiler doesn't optimise the switch to a jump-table, which one will that one make fastest?! Caching effects are not taken into consideration. I could go on...

Quote:

---

Originally Posted by MAtkins

So, tell me about those SSD drives . . .
I'm running Win 7 64 bit on an Alienware Aurora - overclocked at I think 3.67 Ghz Quad core - 6 Gig RAM.
It's got a radiator in it.

---

Ooh nice! Just got a new machine at work yesterday. Win 7 64 bit, 2.9GHz 8 core, 12 Gig RAM, and a couple of 10000RPM raptor 512 Gig hard drives.

One guy at work had an SSD in his old machine, and oh yeah is that fast!