Skip to main content
LLVM

LLVM

Computer programWikipedia

Search complete. 67 mentions across 23 episodes found for "LLVM".

Nov 19, 2026

Jeremy DorfmanGUEST
11:46
Catapult is used for HLS to RTL synthesis, and TAFL is used to glue the HLS together, both pre-synthesis as a C model and post-synthesis as an RTL design.
Jeremy DorfmanGUEST
11:57
Are use of HLS allowed for novel use of the LLVM sanitizers, especially address sanitizer and memory sanitizer, to uncover issues where the hardware would have misbehaved or had some undefined behavior? Because our C model was available throughout the project's development, we were able to use our video quality corpus of many thousands of videos, along with many thousands of servers, to identify and fix a number of perceptual quality issues pre-Silicon.
Jeremy DorfmanGUEST
12:22
Perhaps most importantly, HLS's agility made evaluating and implementing encoding tools much faster compared to a conventional HDL design.
Jeremy DorfmanGUEST
12:31
Notably, we were able to add a few very late features that would have otherwise been prohibitive.
Jason TupnerHOST
25:57
Team does not write anything itself, and the account says so.
Jason TupnerHOST
26:01
The tree imports 119 third-party components, 92 under contrib, 22 under sys contrib, five under crypto, among them LLVM, OpenSSH, OpenSSL, OpenZFS, XZ, and the TimeZone database.
Jason TupnerHOST
26:16
The committer's guide prescribes how each arrives on a vendor branch where it is crucial to verify the source code you are importing comes from a trustworthy source.
Jason TupnerHOST
26:27
and is merged from there into the tree, after which is built with the kernel, tested with the kernel, released on the kernel's day, and patched through the same advisory as kernel.
Jason TupnerHOST
27:30
Counted on the 21st of October, 2026 from the Debian cloud image manifest and the Trixie package index from FreeBSD.org administration to contributors article, the release engineering 15.1 tree and two commits from the public log.
Jason TupnerHOST
27:47
The limit, one tree still carries 119 imports.
Jason TupnerHOST
27:51
LLVM, OpenSSH, OpenZFS and XSED come from outside when XSED ships a bad release, the vendor branch receives it like anyone else.
Jason TupnerHOST
28:00
What changes is that one team integrates it and the same team answers for it afterwards.
Dan AcristiniiGUEST
3:01
And they're basically compiling your app and the runtime.
Dan AcristiniiGUEST
3:05
I think they're going to AST, and then they're using LLVM to pass that into microcontroller code.
Dan AcristiniiGUEST
3:11
So this is my high-level understanding of it.
Dan AcristiniiGUEST
3:15
The thing is, it works both for microcontrollers and for Wasm.
Josh StroscheinHOST
0:09
I've never actually created them before.
Josh StroscheinHOST
0:10
LLVM seems like a great place to start.
Josh StroscheinHOST
0:12
It has an architecture.
Josh StroscheinHOST
0:13
There's a project out there, OLVM.
Josh StroscheinHOST
0:21
This episode is going to be a bit different.
Josh StroscheinHOST
0:23
I'm trading seats with my colleague and co-host, Blas Kajusner.
Josh StroscheinHOST
0:26
We're covering obfuscation and deobfuscation, from intimidating first impressions to building obfuscators in LLVM and also covering agent-driven tooling.
Josh StroscheinHOST
0:35
We'll also talk about delivering the syntactical supremacy course at Black Hat this past summer and how I used it to structure my research.
Chris TitusHOST
4:56
Like most large companies, they didn't want the deals of the tech stack for unreleased products to become public sooner than necessary.
Chris TitusHOST
5:04
Their largest contribution was LLVM.
Chris TitusHOST
5:06
And even goes down to say, if their contribution is sponsored by at all, they showed up as sponsored by FreeBSD developer consulting company.
Chris TitusHOST
5:17
Sony was very big on not getting credit.
TomHOST
76:58
RDNA 5 shows up in Linux drivers and then something called Siren Halo is rumored.
TomHOST
77:04
Quoting from Pharonix First, since the beginning of the year, there has been an AMD GPU LLVM backend patches for GFX 13 as AMD's next generation graphics IP.
TomHOST
77:15
This is most likely for RDNA 5 graphics.
TomHOST
77:18
Now, out today are the gfx 13 patches finally surfacing for the amd gpu linux kernel driver graphics driver so yes first part of the story the solid part amd has submitted over 12,600 lines of gfx 13 driver code to linux with two targets gfx 13.0.1 and 13.0.4 amd hasn't officially called it rdna 5 but gfx 12 was rdna 4 So it seems like a pretty safe guess.
Frank KruegerHOST
27:38
Imagine it from the developer's perspective, everything they want to change, they have two runtimes they have to keep in check.
Frank KruegerHOST
27:44
And it's actually more than that, because there's the LLVM mono runtime, and then there's the normal monoruntime.
Frank KruegerHOST
27:52
And so they're having to keep these things all in sync.
Frank KruegerHOST
27:54
So whenever they want to do something crazy, like say move async into the runtime, then they have to do it to multiple runtimes and keep it all straight.
Frank KruegerHOST
28:21
I do have a critique to it, but I want to say overall super cool, obviously, because we've been waiting for this since basically .NET 5.
James MontemagnoHOST
28:30
What's your critique?
Frank KruegerHOST
28:32
My critique is I love LLVM, the mono runtime that's LLVM.
Frank KruegerHOST
28:37
So I have a little app, if I haven't mentioned it 20 times yet on this podcast, iCircuit, whose whole, the whole app is basically math, doing math as fast as it possibly can.
Geoff NielsonHOST
0:24
Chris Lattner has built coding infrastructure for all of them that you've probably never heard of.
Geoff NielsonHOST
0:29
But if you're a full-stack developer, then LLVM, Swift, and Clang are probably all familiar to you.
Geoff NielsonHOST
0:36
Now, Chris is taking on what he calls an epic industry-scale problem creating a new programming language to break down siloed environments and cover all of compute.
Geoff NielsonHOST
0:47
In a moment when AI is upending all of software development, Chris is the rare guy who doesn't just have a vision for the future, he's actually building it.

39 MINS LATER

Chris LattnerGUEST
39:49
And so what Qualcomm wants is it wants choice.
Chris LattnerGUEST
39:51
It wants to neutralize the software landscape so that chips can thrive and compete on their own merits so that people do have choice and go forward with this.
Chris LattnerGUEST
40:00
To make an analogy, so when I was at Apple, I was building this LLVM compiler technology that also helped make the entire world work and pretty much everybody builds on top of.
Chris LattnerGUEST
40:09
It's the same thing.
Kostya SerebryanyGUEST
2:44
It is a part of two different compilers.
Kostya SerebryanyGUEST
2:49
This is now in released versions of both LLVM and GCC compilers.
Kostya SerebryanyGUEST
2:55
The compiler model instruments all loads and stores and does some tricky stuff with the with the stack and globals.
Kostya SerebryanyGUEST
3:08
The second part of the tool is the runtime library, which is basically a complete malloc replacement.

7 MINS LATER

Kostya SerebryanyGUEST
9:46
Let's see what we have about the second tool.
Kostya SerebryanyGUEST
9:50
It is also using compile time instrumentation.
Kostya SerebryanyGUEST
9:54
And it is also a part of both compilers, both major open source compilers, LLVM and GCC.
Kostya SerebryanyGUEST
10:05
The major part of the tool is, again, runtime library, which is, again, a malloc replacement.
Steve KlabnikHOST
61:54
It's only passing like 99% of the test suite or whatever.
Steve KlabnikHOST
61:59
And it's also using Cranelift instead of LLVM.
Steve KlabnikHOST
62:01
So it's like about 30% slower in terms of the generated code, but it's like way faster compiling stuff.
Steve KlabnikHOST
62:06
It's like got those trade-offs.

13 more episodes mention LLVM.

Create an account to see the whole feed, search across every transcript, and follow the entities you care about.

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.