@FelipeNorrisLee "In an ideal world would we need branch prediction? Are there algorithms that can only run efficiently through branch prediction?" I'm not a security person, and I've only done performance-aware programming for a few years now, so I don't have the experience to say anything conclusively. But I think I know enough to suggest that only a very small branch predictor would be needed at worst. What you can do instead of branching is masking, like you do on the GPU. On the CPU side, a similar concept is also sometimes called branchless programming. I have very limited experience with it, but I believe branchless programming typically refers to scalar code, and the optimal thing would be if virtually all code was vectorized/SIMD. In fairly performant code, I typically only use scalar code for loop controls or for other setup that is not the meat of the application. Today, simple loop controls (like an index in a for-loop or a while-is-non-zero) become effectively free, because they run parallel to the SIMD and is extremely easy for the branch predictor (because it'll never miss during the hot loop), and I would guess that there could be algorithms where that ability would still be great to keep. Regardless though, the CPU's job is to be general-purpose, so I would be very surprised if we ever saw the branch predictor going away, even if we got a software revolution that made most programmers much more performance-aware.
I was gonna keep it nice but here's ur constructive criticism. I love Casey, but you bored me to death with this video. You may as well repost @ThePrimeTimeagen. You're talking to a guy with 30 years of mastery and you say "wow awesome, so here's this question from one of my viewers" you talked mostly about how you needed X premium and then let Casey handle everything else?
@@ChristIsBreathi was honestly just about to argue with you, but i don't think my opinion will matter to you. nobody is forcing you to watch and your feedback wasn't very nice. it's not his job to teach us. the people he brings on are the experts
@@ChristIsBreath You're right that the interviewer should have been more prepared and came with some specific technical questions. But I still learned a few things.
LOL me beating you to the punch, @ (th-cam.com/video/IXiPWa9RWgc/w-d-xo.html) "No, this is great".10/10 content peepz, truly #31337 individuals, for real for real
Incredible knowledge here from Casey. Thanks for the video!
My favourite Channel on youtube!, thank you, love your content!
You should put "Casey Muratori" in your description or on screen so it appears when you search that term.
I'm glad you noticed! I had all of that in there and it disappeared. Added it back now. Thanks!
Such a great video.
Thank you so much for sharing information in such an informative way.
thanks Casey!
@FelipeNorrisLee "In an ideal world would we need branch prediction? Are there algorithms that can only run efficiently through branch prediction?" I'm not a security person, and I've only done performance-aware programming for a few years now, so I don't have the experience to say anything conclusively. But I think I know enough to suggest that only a very small branch predictor would be needed at worst. What you can do instead of branching is masking, like you do on the GPU. On the CPU side, a similar concept is also sometimes called branchless programming. I have very limited experience with it, but I believe branchless programming typically refers to scalar code, and the optimal thing would be if virtually all code was vectorized/SIMD. In fairly performant code, I typically only use scalar code for loop controls or for other setup that is not the meat of the application. Today, simple loop controls (like an index in a for-loop or a while-is-non-zero) become effectively free, because they run parallel to the SIMD and is extremely easy for the branch predictor (because it'll never miss during the hot loop), and I would guess that there could be algorithms where that ability would still be great to keep. Regardless though, the CPU's job is to be general-purpose, so I would be very surprised if we ever saw the branch predictor going away, even if we got a software revolution that made most programmers much more performance-aware.
Starts properly at 4:29
Removing comments hey?
I was gonna keep it nice but here's ur constructive criticism. I love Casey, but you bored me to death with this video. You may as well repost @ThePrimeTimeagen.
You're talking to a guy with 30 years of mastery and you say "wow awesome, so here's this question from one of my viewers" you talked mostly about how you needed X premium and then let Casey handle everything else?
th-cam.com/video/xCBrtopAG80/w-d-xo.html
@@ChristIsBreath "and then let Casey handle everything else" I fail to see the problem.
@@ChristIsBreathi was honestly just about to argue with you, but i don't think my opinion will matter to you. nobody is forcing you to watch and your feedback wasn't very nice. it's not his job to teach us. the people he brings on are the experts
@@ChristIsBreath You're right that the interviewer should have been more prepared and came with some specific technical questions. But I still learned a few things.
LOL me beating you to the punch, @ (th-cam.com/video/IXiPWa9RWgc/w-d-xo.html) "No, this is great".10/10 content peepz, truly #31337 individuals, for real for real
And thankyou