Theresa Foley

At NVIDIA. Languages, compilers, and APIs for real-time. DMs open for folks looking for advice/mentoring. She/her 🏳️‍⚧️🏳️‍🌈. Personal acct: @TessFactor.

[Theresa Foley @TangentVector]

The Slang project is seeking experienced GPU/graphics/AI compiler programmers who want to be part of the development of an ecosystem for AI-powered real-time graphics. My DMs are open.

[NVIDIA AI Developer @NVIDIAAIDev]

New #NVIDIAResearch paper: SLANG.D: Fast, Modular and Differentiable Shader Programming: shows how a single language serves as a unified platform for real-time, inverse, and differentiable rendering.

Collaboration with @MIT, @UCSanDiego, & @UW.
🧵 1/2

https://t.co/5DHzQldMMV

[NVIDIA AI Developer @NVIDIAAIDev]

New #NVIDIAResearch paper: SLANG.D: Fast, Modular and Differentiable Shader Programming: shows how a single language serves as a unified platform for real-time, inverse, and differentiable rendering.

Collaboration with @MIT, @UCSanDiego, & @UW.
🧵 1/2

https://t.co/5DHzQldMMV

[Yong He @csyonghe]

Bringing autodiff to shaders is a challenging task. It takes years of effort to design the language that integrates differentiation as a first-class citizen, allowing autodiff to work seamlessly with custom types, arbitrary control flow, generics and dynamic dispatch.

[Yong He @csyonghe]

See our blog post here: https://t.co/1tjloYKEzz

[Yong He @csyonghe]

Slang is an open-source, cross-platform shading language that targets D3D, Vulkan, GLSL, CUDA and C++. Today, it is fully differentiable, which means you can autodiff your existing shader code!

[Theresa, Who Smiles at Death @TessFactor]

It is traditional for programmers to measure our ages in hexadecimal. Today I became 2B, so it only seemed appropriate…

[Theresa, Who Smiles at Death @TessFactor]

Be the cyberpunk dystopia you want to see in the world.

[Theresa Foley @TangentVector]

@aaronlefohn This is a great talk by @csyonghe, showing how key design decisions Yong made in his Ph.D. work on Slang (advised by @kayvonf) allow the compiler to provide services to real-time ray tracing applications that enable high performance with clean and modular code.

[Aaron Lefohn @aaronlefohn]

(1/5) Shading language and real-time rendering innovation are inextricably tied together.

In this new talk, Slang researcher, Yong He, describes the details of new language features in Slang, and how they are used by real-time path tracing researchers.

https://t.co/nYXYa3iSHu

[Aaron Lefohn @aaronlefohn]

We've been co-researching shading language and real-time rendering innovation for years.

@TangentVector's recent talk describes the Slang journey from research to production.

[Theresa Foley @TangentVector]

Anybody who is excited about the possibilities of Slang can contact me by email/DM. Collaborators are welcome, and also we are hiring: https://t.co/D5SyTfrr52

[Theresa Foley @TangentVector]

Folks who would like to learn more about the Slang shading language and our experiences deploying it at NVIDIA can check out the recording of my talk at the LLVM WiCT meetup: https://t.co/MC2OOz04yl

[Aaron Lefohn @aaronlefohn]

(1/10) Slides + Recording of Petrik Clarberg’s GDC/GTC real-time path tracing research talk are now available online #GDC2022 #GTC22

https://t.co/CNAlWkIwlP

[Theresa Foley @TangentVector]

@tom_forsyth The renewed conversations around dxc and DXIL are also relevant context for me.
The dxc and Slang projects started at similar times, one based on clang/LLVM and the other not. The trade-offs involved in such decisions are nuanced.

[Theresa Foley @TangentVector]

I see people debating use of clang/LLVM for GPU/graphics compilers again, but few recognizing the “nobody ever got fired for buying IBM” factor.

[Theresa Foley @TangentVector]

@tom_forsyth We are in subtweet-of-a-subtweet territory, and many of us are stuck vaguely alluding to compilers we have worked on/with at various companies, because we aren’t allowed talk specifics. Part of what I saw was talking about the Mesa stack and its “NIR.”

[Theresa Foley @TangentVector]

@GrandmasterTru The perception of safety can lead people to ignore the technical reasons why an idea may be less favorable (like the “unstable IR” issue). Part of my point is that people aren’t necessarily wrong to do so, given the political aspects of growing a project.

[Theresa Foley @TangentVector]

@GrandmasterTru LLVM being seen as a “safe” bet is part of why SPIR happened despite warnings from people like me about the instability issue.
I suspect dxc was built on top of clang in part because it was seen as “safe,” and yet here we are as an industry discussing a re-do on that effort.

[Theresa Foley @TangentVector]

I can attest to the flip side being challenging. If you are sure that the “safe” choice is not the right one for your project, do not expect immediate buy-in and be prepared to do the hard work of explaining and justifying your choice.

[Theresa Foley @TangentVector]

It helps that both LLVM and clang are great technologies with strong communities. They are seldom an outright *bad* choice when they get used.