FFmpeg devs boast of another 100x leap thanks to handwritten assembly code

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by Aardwolf on Sunday July 20, 2025 @05:31PM (#44629483)

The article somtimes says 100x, other times it says 100% speed boost. E.g. it says "boosts the app’s ‘rangedetect8_avx512’ performance by 100.73%." but the screenshot shows 100.73x.

100x would be a 9900% speed boost, while a 100% speed boost would mean it's 2x as fast.

Which one is it?

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by ethan_smith on Sunday July 20, 2025 @07:08PM (#44630230)

It's definitely 100x (or 100.73x) as shown in the screenshot, which represents a 9973% speedup - the article text incorrectly uses percentage notation in some places.

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by andersmurphy on Monday July 21, 2025 @03:27AM (#44632610)

The article probably contains traces of blue.

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by VladVladikoff on Monday July 21, 2025 @08:45AM (#44634475)

What does this mean?

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by kaitak on Monday July 21, 2025 @10:13AM (#44635342)

https://news.ycombinator.com/item?id=44627072

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by MadnessASAP on Sunday July 20, 2025 @05:46PM (#44629619)

100x to the single function 100% (2x) to the whole filter

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by pizlonator on Sunday July 20, 2025 @05:33PM (#44629498)

The ffmpeg folks are claiming 100x not 100%. Article probably has a typo

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by k_roy on Sunday July 20, 2025 @08:20PM (#44630658)

That would be quite the percentage difference with 100x

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by torginus on Sunday July 20, 2025 @06:17PM (#44629882)

I'd guess the function operates of 8 bit values judging from the name. If the previous implementation was scalar, a double-pumped AVX512 implementation can process 128 elements at a time, making the 100x speedup plausible.

[ Parent | on HN ]

Re:FFmpeg devs boast of another 100x leap thanks to handwrit... by ErrorNoBrain on Sunday July 20, 2025 @06:31PM (#44629984)

it doesnt matter

it's faster

thats what matters

[ Parent | on HN ]