Re: [FFmpeg-devel] AVDictionary vs. AVSet (AVDictionary2 approximation)

["softworkz ." <softworkz-at-hotmail.com@ffmpeg.org>]

> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces@ffmpeg.org> On Behalf Of
> Michael Niedermayer
> Sent: Montag, 14. April 2025 13:33
> To: FFmpeg development discussions and patches <ffmpeg-
> devel@ffmpeg.org>
> Subject: [FFmpeg-devel] AVDictionary vs. AVSet (AVDictionary2
> approximation)
> 
> Hi
> 
> I just posted a AVSet implementation i wrote in the last 2 days (yes
> thats
> why i did dissapear for the last 2 days)
> 
> My plan was to use that AVSet as basis for AVDictionary2 in case
> benchmarks indicate that its worth it, so is it ?
> 
> with 3 entries (100000 runs)
> AVDictionary    0.040sec
> AVSet           0.027sec
> 
> with 5 entries (100000 runs)
> AVDictionary    0.065sec
> AVSet           0.042sec
> 
> with 10 entries (100000 runs)
> AVDictionary    0.193sec
> AVSet           0.087sec
> 
> with 100 entries (100000 runs)
> AVDictionary    8.7  sec
> AVSet           1.4  sec
> 
> with 1000 entries (1000 runs)
> AVDictionary    8.0   sec
> AVSet           0.240 sec
> 
> with 10000 entries (10 runs)
> AVDictionary    7.2   sec
> AVSet           0.042 sec
> 
> 
> I was a bit surprised for the 3 and 5 entry case, maybe my benchmark
> is buggy or
> AVSet is, but then AVDictionary is pretty bad with memory allocations
> 
> AVDictionary needs to strdup every key and value, needs to allocate
> the AVDictionary itself and reallocs the entry array each time
> thats 10 memory allocation related calls for adding 3 entries
> 
> while AVSet allocates the AVSet and then uses av_fast_realloc() for
> the array
> and theres nothing else, the key/value goes in that array too
> 
> 
> bechmark code used is below:
> 
> 
> #if 0
>     for (int runs = 0; runs < 100000; runs++) {
>         AVSet *set = av_set_new(strcmp, NULL, NULL);
>         for(int pass = 0; pass < 2; pass++) {
>             unsigned r = 5;
>             for(int i=0; i<100; i++) {
>                 r = r*123 + 7;
>                 char str[2*7] = "TESTXXTESTXX";
>                 str[4] = r;
>                 str[5] = r>>8;
>                 if(pass == 0) {
>                     av_set_add(set, str, 2*7, 0);
>                 } else {
>                     av_set_get(set, NULL, str, NULL);
>                 }
>             }
>         }
>         av_set_free(&set);
>     }
> #else
>     for (int runs = 0; runs < 100000; runs++) {
>         AVDictionary *dict = NULL;
>         for(int pass = 0; pass < 2; pass++) {
>             unsigned r = 5;
>             for(int i=0; i<100; i++) {
>                 r = r*123 + 7;
>                 char str[7] = "TEST";
>                 str[4] = r;
>                 str[5] = r>>8;
>                 if(pass == 0) {
>                     av_dict_set(&dict, str, str, 0);
>                 } else {
>                     av_dict_get(dict, str, NULL, 0);
>                 }
>             }
>         }
>         av_dict_free(&dict);
>     }
> #endif
> 
> 
> --

Hi Michael,


what's not quite realistic is that all keys are starting with the same 4 characters. This affects the lookups of course - and probably (maybe) not equally for both sides.

Doesn't the code create duplicate keys (at least when it gets > 65536 it will for sure) ?

So, I think, the keys should be completely random (all chars).

I would also check whether the lookups are successful (just to be sure).


Best,
sw


_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".