I'm enjoying the talk. Really glad they went the extra mile and gave a lavalier mic to some guy in the audience with a sore throat. Really simulating the authentic in-person conference experience. Makes me want to wash my hands.
It's funny coming back to this. I was in the crowd. I remember thinking the exact same thing it got really annoying. It was the front two rows of the auditorium from what I remember lol.
Fantastic talk. A nice amount of breadth and great cross-references to other materials (books, presentations, articles) for those who are interested. Note: It was extremely hard to listen to this talk due to the excessive throat clearing of the participants. Maybe some audio processing could easily clean this kinda stuff up?!
I love Timur's talks but this mic keeps picking up three things instead of just one: Timur's voice, Timur's reverberated voice reflected in the room, and that one super annoying coughing guy.
Am I misunderstanding something, but at 1:26:26 doesn't b[i+1] += c[i] do an out-of-bounds write in the example on the left? I know the example is about vectorization and this is unrelated, but I found it an odd that the example would have that line.
I believe so. The common case is when ‘size’ is actually the size of an array that it being traversed, and thie assume attribute can allow the compiler to emit fewer bound-checks, and iterate over multiple items simultaneously- which leads to SIMD. This is discussed in slide 106
Timur's habit is to talk too fast about many diffenet things that may not be needed necessarily, making his talks not very advantageous to me. Most of the times they are confusing more than being simple, clear and ready to be understood.
I'm enjoying the talk.
Really glad they went the extra mile and gave a lavalier mic to some guy in the audience with a sore throat. Really simulating the authentic in-person conference experience. Makes me want to wash my hands.
alright I got 46 minutes in and the guy in the audience started talking to himself in between clearing his throat. I'm done.
@@adodge2 😂😂😂
It's funny coming back to this. I was in the crowd. I remember thinking the exact same thing it got really annoying. It was the front two rows of the auditorium from what I remember lol.
Fantastic talk. A nice amount of breadth and great cross-references to other materials (books, presentations, articles) for those who are interested. Note: It was extremely hard to listen to this talk due to the excessive throat clearing of the participants. Maybe some audio processing could easily clean this kinda stuff up?!
I love Timur's talks but this mic keeps picking up three things instead of just one: Timur's voice, Timur's reverberated voice reflected in the room, and that one super annoying coughing guy.
the noise is annoying.
Download the video then suppress the noise using AI filter
Should not have read the comments, did not notice the background noises till they were brought up.
Loves from Turkey, thx
Am I misunderstanding something, but at 1:26:26 doesn't b[i+1] += c[i] do an out-of-bounds write in the example on the left? I know the example is about vectorization and this is unrelated, but I found it an odd that the example would have that line.
[[assume false]] :) And have fun.
why does the audio sound so bad?
sounds like the in camera audio track was left in
First example and code starts at 44 min.
I wonder, does something like [[assume size % 32 ==0 ]] help promote SIMD optimizations?
I believe so. The common case is when ‘size’ is actually the size of an array that it being traversed, and thie assume attribute can allow the compiler to emit fewer bound-checks, and iterate over multiple items simultaneously- which leads to SIMD. This is discussed in slide 106
It might. On clang (for arm64-v8a) I've noticed even a fixed value of `N=32` does not always lead to autovectorisation, but `size % 64 == 0` would do.
Code starts at 45.00
1:24:40 Jeff Garland’s talk: ua-cam.com/video/hlgCeWC9jxI/v-deo.html
1:24:40 Eduardo Madrid’s talk from 2022 about SWAR: ua-cam.com/video/B4VxpvFX9YY/v-deo.html
is there anywhere this has been uploaded without the horrific perpetual background coughing and footstamping?
That damn cough!
Timur's habit is to talk too fast about many diffenet things that may not be needed necessarily, making his talks not very advantageous to me. Most of the times they are confusing more than being simple, clear and ready to be understood.
it's 2023. who goes to a conference with a cough, ffs?
bad sound recording
That's one hell of annoying crowd to present to.