Skip to main content
Single instruction, multiple data

Single instruction, multiple data

Search complete. 11 mentions across 5 episodes found for "Single instruction, multiple data".

Nov 19, 2026

Shadow Fox InfiniteGUEST
79:55
So it's about what that Zen 2 core could do and how well Rockstar used the CPU intrinsics for it.
Shadow Fox InfiniteGUEST
80:06
They would just have to be just as smart using Neon on Switch 2 and taking those SIMD tasks and moving them over to there.
Shadow Fox InfiniteGUEST
80:14
Neon is newer than Zen 2's AVX.
Shadow Fox InfiniteGUEST
80:18
So there's a little bit of benefit there should they choose to use it.
Continue

Continue

SIMD City

Nov 14 · 7 Mentions

Kris BrandowHOST
5:26
So that's 8-3-0-0-9.
Kris BrandowHOST
5:28
And this is a proposal to add the SIMD and SIMD slash Arch SIMD packages to the standard library to add portable and architecture-specific SIMD.
Kris BrandowHOST
5:39
This is a proposal from Austin, the great BDFL for Go, or I guess BDFL for now for Go, the Go tech lead, if you will.
Kris BrandowHOST
5:50
And so this is a proposal to move the SIMD functionality that's currently behind, I think, an experimental flag at the moment into the standard library, or at least a subset of that functionality into the standard library.
Kris BrandowHOST
6:04
What I think is really interesting and cool about this is that they have a very specific way that they're approaching this API design, where the same name must mean the same thing, even across the different SIMD types that you have.
Kris BrandowHOST
6:18
And the structure of the names needs to be consistent as well.
Kris BrandowHOST
6:21
So even using like the same letters or the same letter sequence needs to mean the same thing across the pack, which means you wind up with a very consistent API at the end of the day.
Kris BrandowHOST
6:31
The interesting thing about this is there is an open question around vectors specifically about whether a vector should have an equal method.
Bob FarzinGUEST
51:47
Isn't that true? Yeah.
Mike MullHOST
51:48
Yeah, so like when I was mentioning earlier that people were porting codes to the Cray, that was one of the big reasons was, you know, they had a sophisticated version of what we would now call SIMD.
Mike MullHOST
52:01
So, you know, you could, if you had loops in your Fortran code that were going element by element in a matrix, you could now do like 16 of those calculations in one clock cycle on the Cray.
Erik AkerHOST
52:16
The other
Sudeep BhojaGUEST
11:35
It had a runtime.
Sudeep BhojaGUEST
11:37
So we were able to prove out the whole stack using these basic fundamental building block elements that did matrix vector multiply, SIMD operations, reshapes, data movement, and so on.
Vik SekarHOST
11:52
Yeah.
Vik SekarHOST
11:52
What was the first version of this chip called? What did you call it?
Ryan HuntGUEST
9:52
On the website of things, we've just been very preoccupied with, I think, all the features I sort of mentioned before, like shipping exception handling, and GC, and just different feet.
Ryan HuntGUEST
10:06
Oh, SIMD.
Ryan HuntGUEST
10:09
different things that were sort of like higher on the priority list and components were in so much flux that we kind of just took like a well let's let it bake a bit before we kind of come back to evaluate it but about a year ago I was just kind of looking over like some of the future WASM features that we'd like and it They're definitely important, but I kind of got the feeling that actually components are probably the thing that we'd want to look at next.
Ryan HuntGUEST
10:36
About the same time, too, is when I was really kind of on a kick of like, why don't more people use Wasm? And I was doing a lot of that sort of user research.

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.