> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-boun...@ffmpeg.org> On Behalf Of
> Michael Niedermayer
> Sent: Mittwoch, 16. April 2025 23:25
> To: FFmpeg development discussions and patches <ffmpeg-
> de...@ffmpeg.org>
> Subject: Re: [FFmpeg-devel] [PATCH 1/3] avutil/dict2: Add
> AVDictionary2 with hash-based lookup
>
> Hi
>
> i like AI and ill reply with more comments about this elsewhere
Cool!
> but as i looked at the code, i had to reply here
[...]
> > +
>
> > + if (prev)
> > + return NULL; // 'prev' functionality not implemented
>
> not implemented ?
see below...
> > + // Get hash index
>
> > + hash = dict_hash(key, m->flags & AV_DICT2_MATCH_CASE);
>
> case sensitivity is supported by having the set funtiom insert with
> the
> case sensitivity that the get function later will use ?
>
>
> > + table_idx = hash % m->table_size;
> > +
> > + // Search in chain
> > + for (entry = m->entries[table_idx]; entry; entry = entry->next)
> {
> > + if ((m->flags & AV_DICT2_MATCH_CASE ?
> > + !strcmp(entry->key, key) :
> > + !av_strcasecmp(entry->key, key))) {
> > +
> > + // Found match
> > + de.key = entry->key;
> > + de.value = entry->value;
> > + return &de;
>
> tasty globals for thread saftey
To be fair, the "not implemented" and non-thread-safety goes on my account.
The AV_DICT2_MATCH_CASE hasn't come to my attention.
The initial problem was this: in your requirements text (that I had assembled
from two of your e-mails), there was a mention of AVL trees and it knew that
FFmpeg had an implementation for this (avtree) and tried to use that. And here
it struggled and took one attempt after another. Since I'm not familiar with the
concepts behind AVL trees, I couldn't help or understand what's missing, but it
became clear that something was missing - without deeper knowledge, I presume
that
those parts that the AI was looking for are exactly the ones you have just added
right?
While it was struggling with this, it tried several times to break out by
creating
an entirely new and hash-based implementation. Eventually I let it proceed and
the
results were pretty nice. A lot of work was then done on the benchmarks.
Benchmarks are beasty beasty beasts and I was never satisfied as I didn't
want to post any results that wouldn't stand up.
Getting a basic but performant implementation with benchmarks was the goal, and
then getting some feedback what people think about the code. It was meant as
a "do-what-you-want-with-it" submission. I never wanted to work on this - it
just came up in the right moment as a great subject for our little experiment.
Also, to be clear: I still have absolutely no ambitions to work on this. I think
you are on a very good track.
Or actually, it's not that I just think that, I even got first-hand evidence 😊
I have included AVMap in the benchmarks for the AIDictionary (let's call it
like that
to avoid confusion), spent a bit more time than intended, but I wanted results
to
solid and less synthetic than the ones you have added for AVMap.
You can find the results below as numbers.
Also, here's a chart:
https://softworkz.github.io/ffmpeg_output_apis/dictionary_benchmark_chart.html
(that's why I love AI, never would have created that manually)
My assessment of the results is as follows:
- AVMap and AIDictionary are very similar in their performance characteristics
- AIDictionary is slightly better in performance over a wide range, but those
few
percent don't play any role practically. And maybe you have still some
opportunities
for improvement And anyway - with so small margins, you can always easily
construct
a different test where the relations are turned. There is no test that is
ultimately
fair beyond all doubt. The smallest bits can turn the outcome - so IMO:
irrelevant
- What's not irrelevant is that AVMap has a number of punctual advantages by
much more
than just a few percent. Specifically, it's much better for cases with very
small
numbers of entries.
- It's also much faster when iterating over all entries (4.)
I've been a bit in doubt to be honest, somewhat irritated by the weird key
generation
(suppose you've seen what I meant, how much difference it makes whether
randomizing the
first two or 4th and 5th character).
But now from all those figures, I can only say: Well done so far!
Best wishes,
sw
------------------------------
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 1 entries
1. Insertion Performance:
AVDictionary : avg 310 cycles (min: 256, max: 33236)
AVDictionary2: avg 466 cycles (min: 330, max: 33786) (150.3% of baseline)
AVMap : avg 214 cycles (min: 114, max: 38978) (69.0% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 41213 cycles (min: 36430, max: 3024922)
AVDictionary2: avg 81205 cycles (min: 70988, max: 3601472) (197.0% of
baseline)
AVMap : avg 48939 cycles (min: 39106, max: 2281900) (118.7% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 25610 cycles (min: 20798, max: 1442648)
AVDictionary2: avg 63738 cycles (min: 52730, max: 1698434) (248.9% of
baseline)
AVMap : avg 30859 cycles (min: 25826, max: 2865302) (120.5% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 42 cycles (min: 34, max: 119074)
AVDictionary2: avg 72 cycles (min: 64, max: 40280) (171.4% of baseline)
AVMap : avg 41 cycles (min: 34, max: 43098) (97.6% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 2 entries
1. Insertion Performance:
AVDictionary : avg 1111 cycles (min: 892, max: 45208)
AVDictionary2: avg 657 cycles (min: 514, max: 39998) (59.1% of baseline)
AVMap : avg 563 cycles (min: 388, max: 32666) (50.7% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 42563 cycles (min: 37992, max: 1444638)
AVDictionary2: avg 79977 cycles (min: 71940, max: 2278092) (187.9% of
baseline)
AVMap : avg 52504 cycles (min: 43040, max: 1694018) (123.4% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 28497 cycles (min: 23546, max: 1208568)
AVDictionary2: avg 63147 cycles (min: 52760, max: 2291256) (221.6% of
baseline)
AVMap : avg 34309 cycles (min: 29282, max: 1218274) (120.4% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 54 cycles (min: 42, max: 48548)
AVDictionary2: avg 78 cycles (min: 66, max: 122388) (144.4% of baseline)
AVMap : avg 55 cycles (min: 46, max: 43104) (101.9% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 3 entries
1. Insertion Performance:
AVDictionary : avg 1307 cycles (min: 938, max: 49588)
AVDictionary2: avg 1080 cycles (min: 720, max: 231430) (82.6% of baseline)
AVMap : avg 1068 cycles (min: 718, max: 145108) (81.7% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 43016 cycles (min: 35706, max: 20930252)
AVDictionary2: avg 78491 cycles (min: 71308, max: 2487020) (182.5% of
baseline)
AVMap : avg 52500 cycles (min: 44162, max: 6888954) (122.0% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 34417 cycles (min: 28256, max: 1679612)
AVDictionary2: avg 65345 cycles (min: 53772, max: 1619062) (189.9% of
baseline)
AVMap : avg 37534 cycles (min: 31524, max: 4348252) (109.1% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 57 cycles (min: 50, max: 32814)
AVDictionary2: avg 83 cycles (min: 70, max: 45314) (145.6% of baseline)
AVMap : avg 79 cycles (min: 54, max: 935208) (138.6% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 5 entries
1. Insertion Performance:
AVDictionary : avg 2253 cycles (min: 1724, max: 623386)
AVDictionary2: avg 1426 cycles (min: 1116, max: 98876) (63.3% of baseline)
AVMap : avg 1611 cycles (min: 1308, max: 138812) (71.5% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 46239 cycles (min: 38886, max: 2299626)
AVDictionary2: avg 78133 cycles (min: 71920, max: 5235922) (169.0% of
baseline)
AVMap : avg 56785 cycles (min: 48668, max: 1346058) (122.8% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 40070 cycles (min: 33774, max: 5272322)
AVDictionary2: avg 68731 cycles (min: 53924, max: 2461530) (171.5% of
baseline)
AVMap : avg 43399 cycles (min: 37238, max: 1579700) (108.3% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 74 cycles (min: 66, max: 39600)
AVDictionary2: avg 93 cycles (min: 80, max: 41166) (125.7% of baseline)
AVMap : avg 97 cycles (min: 78, max: 133814) (131.1% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 10 entries
1. Insertion Performance:
AVDictionary : avg 5019 cycles (min: 3798, max: 352176)
AVDictionary2: avg 2699 cycles (min: 2040, max: 247852) (53.8% of baseline)
AVMap : avg 3933 cycles (min: 2766, max: 165834) (78.4% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 58752 cycles (min: 48994, max: 3076670)
AVDictionary2: avg 82146 cycles (min: 72202, max: 6867008) (139.8% of
baseline)
AVMap : avg 66605 cycles (min: 55006, max: 3153026) (113.4% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 74699 cycles (min: 59334, max: 3707194)
AVDictionary2: avg 89139 cycles (min: 66064, max: 25436082) (119.3% of
baseline)
AVMap : avg 54566 cycles (min: 42618, max: 10240104) (73.0% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 129 cycles (min: 102, max: 199994)
AVDictionary2: avg 132 cycles (min: 112, max: 142572) (102.3% of baseline)
AVMap : avg 154 cycles (min: 126, max: 44478) (119.4% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 25 entries
1. Insertion Performance:
AVDictionary : avg 12823 cycles (min: 10154, max: 944296)
AVDictionary2: avg 7146 cycles (min: 5632, max: 464294) (55.7% of baseline)
AVMap : avg 8176 cycles (min: 6256, max: 350452) (63.8% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 89609 cycles (min: 77974, max: 2995522)
AVDictionary2: avg 82302 cycles (min: 73160, max: 3007334) (91.8% of
baseline)
AVMap : avg 79883 cycles (min: 68174, max: 3140628) (89.1% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 123457 cycles (min: 110558, max: 1610438)
AVDictionary2: avg 90002 cycles (min: 77210, max: 1325144) (72.9% of
baseline)
AVMap : avg 64624 cycles (min: 56714, max: 2473494) (52.3% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 260 cycles (min: 208, max: 600242)
AVDictionary2: avg 243 cycles (min: 192, max: 798670) (93.5% of baseline)
AVMap : avg 356 cycles (min: 282, max: 392158) (136.9% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 50 entries
1. Insertion Performance:
AVDictionary : avg 29978 cycles (min: 24328, max: 1474502)
AVDictionary2: avg 16420 cycles (min: 12988, max: 284426) (54.8% of baseline)
AVMap : avg 23558 cycles (min: 11908, max: 970974) (78.6% of baseline)
2. Lookup Performance (100% existing keys, 100000 runs):
AVDictionary : avg 155116 cycles (min: 135390, max: 9012880)
AVDictionary2: avg 82615 cycles (min: 73602, max: 2761130) (53.3% of
baseline)
AVMap : avg 90493 cycles (min: 77622, max: 3824524) (58.3% of
baseline)
3. Lookup Performance (50% existing keys, 100000 runs):
AVDictionary : avg 251623 cycles (min: 206464, max: 12732728)
AVDictionary2: avg 112692 cycles (min: 88980, max: 3814908) (44.8% of
baseline)
AVMap : avg 91424 cycles (min: 74244, max: 2877784) (36.3% of
baseline)
4. Iteration Performance (100000 runs):
AVDictionary : avg 519 cycles (min: 400, max: 1152432)
AVDictionary2: avg 439 cycles (min: 380, max: 217098) (84.6% of baseline)
AVMap : avg 661 cycles (min: 558, max: 810242) (127.4% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 100 entries
1. Insertion Performance:
AVDictionary : avg 81115 cycles (min: 64952, max: 847136)
AVDictionary2: avg 40204 cycles (min: 27924, max: 401170) (49.6% of baseline)
AVMap : avg 50393 cycles (min: 23934, max: 298484) (62.1% of baseline)
2. Lookup Performance (100% existing keys, 10000 runs):
AVDictionary : avg 313071 cycles (min: 264542, max: 7147818)
AVDictionary2: avg 82780 cycles (min: 74604, max: 1326828) (26.4% of
baseline)
AVMap : avg 102494 cycles (min: 90724, max: 1213110) (32.7% of
baseline)
3. Lookup Performance (50% existing keys, 10000 runs):
AVDictionary : avg 474283 cycles (min: 403718, max: 2530658)
AVDictionary2: avg 126364 cycles (min: 101840, max: 1848498) (26.6% of
baseline)
AVMap : avg 118218 cycles (min: 95340, max: 1516344) (24.9% of
baseline)
4. Iteration Performance (10000 runs):
AVDictionary : avg 893 cycles (min: 772, max: 41262)
AVDictionary2: avg 778 cycles (min: 724, max: 44356) (87.1% of baseline)
AVMap : avg 1168 cycles (min: 1088, max: 45592) (130.8% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 250 entries
1. Insertion Performance:
AVDictionary : avg 333125 cycles (min: 258744, max: 1325610)
AVDictionary2: avg 159356 cycles (min: 95324, max: 537758) (47.8% of
baseline)
AVMap : avg 186784 cycles (min: 78274, max: 585006) (56.1% of
baseline)
2. Lookup Performance (100% existing keys, 10000 runs):
AVDictionary : avg 748016 cycles (min: 683050, max: 3483690)
AVDictionary2: avg 128249 cycles (min: 87802, max: 1590854) (17.1% of
baseline)
AVMap : avg 140113 cycles (min: 122018, max: 1288682) (18.7% of
baseline)
3. Lookup Performance (50% existing keys, 10000 runs):
AVDictionary : avg 1030383 cycles (min: 948024, max: 3080186)
AVDictionary2: avg 154383 cycles (min: 129396, max: 1409748) (15.0% of
baseline)
AVMap : avg 141779 cycles (min: 124046, max: 1447820) (13.8% of
baseline)
4. Iteration Performance (10000 runs):
AVDictionary : avg 2690 cycles (min: 1988, max: 367178)
AVDictionary2: avg 2521 cycles (min: 1828, max: 1178014) (93.7% of baseline)
AVMap : avg 3374 cycles (min: 2686, max: 1377088) (125.4% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 500 entries
1. Insertion Performance:
AVDictionary : avg 1011189 cycles (min: 831082, max: 3192930)
AVDictionary2: avg 374018 cycles (min: 243866, max: 1257420) (37.0% of
baseline)
AVMap : avg 525409 cycles (min: 171142, max: 1671800) (52.0% of
baseline)
2. Lookup Performance (100% existing keys, 10000 runs):
AVDictionary : avg 1479133 cycles (min: 1297470, max: 15430274)
AVDictionary2: avg 182932 cycles (min: 140122, max: 1818982) (12.4% of
baseline)
AVMap : avg 184390 cycles (min: 155302, max: 1593508) (12.5% of
baseline)
3. Lookup Performance (50% existing keys, 10000 runs):
AVDictionary : avg 2033581 cycles (min: 1858034, max: 7074306)
AVDictionary2: avg 176236 cycles (min: 148652, max: 1795112) (8.7% of
baseline)
AVMap : avg 174391 cycles (min: 153552, max: 1727248) (8.6% of
baseline)
4. Iteration Performance (10000 runs):
AVDictionary : avg 4934 cycles (min: 3708, max: 322528)
AVDictionary2: avg 4885 cycles (min: 3660, max: 758670) (99.0% of baseline)
AVMap : avg 6480 cycles (min: 5322, max: 930392) (131.3% of baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 1000 entries
1. Insertion Performance:
AVDictionary : avg 3898727 cycles (min: 3289026, max: 5660258)
AVDictionary2: avg 928904 cycles (min: 718546, max: 1384310) (23.8% of
baseline)
AVMap : avg 778864 cycles (min: 432600, max: 1539546) (20.0% of
baseline)
2. Lookup Performance (100% existing keys, 5000 runs):
AVDictionary : avg 3152981 cycles (min: 2739356, max: 10072370)
AVDictionary2: avg 249594 cycles (min: 183798, max: 2771614) (7.9% of
baseline)
AVMap : avg 224534 cycles (min: 186610, max: 1681348) (7.1% of
baseline)
3. Lookup Performance (50% existing keys, 5000 runs):
AVDictionary : avg 3675493 cycles (min: 3391216, max: 15415892)
AVDictionary2: avg 186528 cycles (min: 156582, max: 1320576) (5.1% of
baseline)
AVMap : avg 193777 cycles (min: 170188, max: 1368988) (5.3% of
baseline)
4. Iteration Performance (5000 runs):
AVDictionary : avg 10146 cycles (min: 7962, max: 931870)
AVDictionary2: avg 10030 cycles (min: 7852, max: 955820) (98.9% of baseline)
AVMap : avg 12658 cycles (min: 10646, max: 826786) (124.8% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 2500 entries
1. Insertion Performance:
AVDictionary : avg 20761246 cycles (min: 19053182, max: 24746272)
AVDictionary2: avg 1966098 cycles (min: 1619400, max: 3111606) (9.5% of
baseline)
AVMap : avg 1857357 cycles (min: 1191456, max: 3066144) (8.9% of
baseline)
2. Lookup Performance (100% existing keys, 1000 runs):
AVDictionary : avg 2950138 cycles (min: 2741008, max: 5653494)
AVDictionary2: avg 234341 cycles (min: 202266, max: 795188) (7.9% of
baseline)
AVMap : avg 222170 cycles (min: 202556, max: 759470) (7.5% of
baseline)
3. Lookup Performance (50% existing keys, 1000 runs):
AVDictionary : avg 8563196 cycles (min: 7522276, max: 16255690)
AVDictionary2: avg 247513 cycles (min: 180328, max: 1408638) (2.9% of
baseline)
AVMap : avg 255538 cycles (min: 210374, max: 1343230) (3.0% of
baseline)
4. Iteration Performance (1000 runs):
AVDictionary : avg 29115 cycles (min: 24030, max: 960446)
AVDictionary2: avg 43865 cycles (min: 34684, max: 1158070) (150.7% of
baseline)
AVMap : avg 31546 cycles (min: 27504, max: 326618) (108.3% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 5000 entries
1. Insertion Performance:
AVDictionary : avg 84455803 cycles (min: 78127970, max: 94263882)
AVDictionary2: avg 4225330 cycles (min: 3606828, max: 5460052) (5.0% of
baseline)
AVMap : avg 3455363 cycles (min: 2522392, max: 5799720) (4.1% of
baseline)
2. Lookup Performance (100% existing keys, 200 runs):
AVDictionary : avg 2963695 cycles (min: 2748164, max: 4507726)
AVDictionary2: avg 250247 cycles (min: 215836, max: 1117026) (8.4% of
baseline)
AVMap : avg 238958 cycles (min: 221464, max: 454684) (8.1% of
baseline)
3. Lookup Performance (50% existing keys, 200 runs):
AVDictionary : avg 17699182 cycles (min: 15498304, max: 29633318)
AVDictionary2: avg 329159 cycles (min: 190404, max: 1297984) (1.9% of
baseline)
AVMap : avg 362946 cycles (min: 246870, max: 1585834) (2.1% of
baseline)
4. Iteration Performance (200 runs):
AVDictionary : avg 53846 cycles (min: 47884, max: 463870)
AVDictionary2: avg 92514 cycles (min: 80262, max: 328678) (171.8% of
baseline)
AVMap : avg 61905 cycles (min: 54010, max: 765550) (115.0% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 10000 entries
1. Insertion Performance:
AVDictionary : avg 362830859 cycles (min: 353381220, max: 383073726)
AVDictionary2: avg 8866988 cycles (min: 7779818, max: 13506320) (2.4% of
baseline)
AVMap : avg 7838390 cycles (min: 5678236, max: 11821630) (2.2% of
baseline)
2. Lookup Performance (100% existing keys, 200 runs):
AVDictionary : avg 3117775 cycles (min: 2739644, max: 4955702)
AVDictionary2: avg 277893 cycles (min: 222796, max: 1665114) (8.9% of
baseline)
AVMap : avg 291541 cycles (min: 247616, max: 1194956) (9.4% of
baseline)
3. Lookup Performance (50% existing keys, 200 runs):
AVDictionary : avg 39118446 cycles (min: 35471378, max: 45295782)
AVDictionary2: avg 432935 cycles (min: 296184, max: 1383836) (1.1% of
baseline)
AVMap : avg 514603 cycles (min: 373354, max: 1077976) (1.3% of
baseline)
4. Iteration Performance (200 runs):
AVDictionary : avg 132250 cycles (min: 95844, max: 1254784)
AVDictionary2: avg 259621 cycles (min: 210856, max: 635688) (196.3% of
baseline)
AVMap : avg 124136 cycles (min: 106810, max: 956594) (93.9% of
baseline)
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 25000 entries
1. Insertion Performance:
AVDictionary : avg 2399516038 cycles (min: 2397701224, max: 2401330852)
AVDictionary2: avg 27057229 cycles (min: 26615140, max: 27499318) (1.1% of
baseline)
AVMap : avg 38317427 cycles (min: 37231696, max: 39403158) (1.6% of
baseline)
2. Lookup Performance (100% existing keys, 500 runs):
AVDictionary : avg 2935091 cycles (min: 2737462, max: 5343036)
AVDictionary2: avg 244981 cycles (min: 211942, max: 1225956) (8.3% of
baseline)
AVMap : avg 299864 cycles (min: 272888, max: 1218488) (10.2% of
baseline)
3. Lookup Performance (50% existing keys, 500 runs):
AVDictionary : avg 92590831 cycles (min: 86713864, max: 109873738)
AVDictionary2: avg 403363 cycles (min: 291794, max: 1639414) (0.4% of
baseline)
AVMap : avg 728032 cycles (min: 503912, max: 1638330) (0.8% of
baseline)
4. Iteration Performance (500 runs):
AVDictionary : avg 367308 cycles (min: 257078, max: 1456362)
AVDictionary2: avg 873216 cycles (min: 686270, max: 3019018) (237.7% of
baseline)
AVMap : avg 328124 cycles (min: 274246, max: 2826626) (89.3% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 50000 entries
1. Insertion Performance:
AVDictionary : avg 10073876706 cycles (min: 10045243966, max: 10102509446)
AVDictionary2: avg 67559563 cycles (min: 60919964, max: 74199162) (0.7% of
baseline)
AVMap : avg 101557773 cycles (min: 101546978, max: 101568568) (1.0%
of baseline)
2. Lookup Performance (100% existing keys, 100 runs):
AVDictionary : avg 3062505 cycles (min: 2741326, max: 4254850)
AVDictionary2: avg 284400 cycles (min: 226482, max: 725182) (9.3% of
baseline)
AVMap : avg 349076 cycles (min: 289712, max: 1353800) (11.4% of
baseline)
3. Lookup Performance (50% existing keys, 100 runs):
AVDictionary : avg 191193034 cycles (min: 178624738, max: 217993696)
AVDictionary2: avg 443757 cycles (min: 397302, max: 1303032) (0.2% of
baseline)
AVMap : avg 957559 cycles (min: 862608, max: 2635630) (0.5% of
baseline)
4. Iteration Performance (100 runs):
AVDictionary : avg 1485354 cycles (min: 814616, max: 3650096)
AVDictionary2: avg 2917238 cycles (min: 2007364, max: 8844446) (196.4% of
baseline)
AVMap : avg 777566 cycles (min: 684692, max: 1939322) (52.3% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 100000 entries
1. Insertion Performance:
AVDictionary : avg 45446026787 cycles (min: 45101533372, max: 45790520202)
AVDictionary2: avg 150192824 cycles (min: 144429420, max: 155956228) (0.3%
of baseline)
AVMap : avg 225821733 cycles (min: 212536504, max: 239106962) (0.5%
of baseline)
2. Lookup Performance (100% existing keys, 100 runs):
AVDictionary : avg 3085317 cycles (min: 2747490, max: 4587106)
AVDictionary2: avg 284027 cycles (min: 228512, max: 836238) (9.2% of
baseline)
AVMap : avg 375552 cycles (min: 309292, max: 733184) (12.2% of
baseline)
3. Lookup Performance (50% existing keys, 100 runs):
AVDictionary : avg 519644568 cycles (min: 425058486, max: 713126436)
AVDictionary2: avg 492125 cycles (min: 419314, max: 2092476) (0.1% of
baseline)
AVMap : avg 1208417 cycles (min: 1036400, max: 2008580) (0.2% of
baseline)
4. Iteration Performance (100 runs):
AVDictionary : avg 3133421 cycles (min: 2194592, max: 7405204)
AVDictionary2: avg 7557450 cycles (min: 6160044, max: 12146984) (241.2% of
baseline)
AVMap : avg 1613565 cycles (min: 1431512, max: 3524188) (51.5% of
baseline)
Benchmark completed successfully
Benchmarking AVDictionary vs AVDictionary2 vs AVMap with 200000 entries
1. Insertion Performance:
AVDictionary : avg 281602757428 cycles (min: 281064682286, max: 282140832570)
AVDictionary2: avg 327146262 cycles (min: 326223736, max: 328068788) (0.1%
of baseline)
AVMap : avg 538823074 cycles (min: 510914178, max: 566731970) (0.2%
of baseline)
2. Lookup Performance (100% existing keys, 100 runs):
AVDictionary : avg 3069747 cycles (min: 2743932, max: 4913570)
AVDictionary2: avg 282565 cycles (min: 230628, max: 959528) (9.2% of
baseline)
AVMap : avg 400410 cycles (min: 326608, max: 2402842) (13.0% of
baseline)
3. Lookup Performance (50% existing keys, 100 runs):
AVDictionary : avg 1557460485 cycles (min: 1494701820, max: 1737170544)
AVDictionary2: avg 471980 cycles (min: 437314, max: 720342) (0.0% of
baseline)
AVMap : avg 1347415 cycles (min: 1230254, max: 1943424) (0.1% of
baseline)
4. Iteration Performance (100 runs):
AVDictionary : avg 6116946 cycles (min: 5399890, max: 9638810)
AVDictionary2: avg 16990515 cycles (min: 15183402, max: 26774720) (277.8% of
baseline)
AVMap : avg 3142539 cycles (min: 2862462, max: 5237466) (51.4% of
baseline)
Benchmark completed successfully
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
