Check out what’s in store for SIGGRAPH 2020, 19-23 July 2020 in Washington, D.C.!

Learn More
Exhibitor Sessions: Intel Exhibitor Session: From RenderMan 22.0® to Next Gen RenderMan XPU and Beyond: Role of Open Shading Language (OSL) with Intel® AVX512.
Event TypeExhibitor Sessions
Interest Areas
Arts & Design
Gaming & Interactive
New Technologies
Production & Animation
Research & Education
Primary Interest Areas
Production & Animation
Registration Levels
EX
XP
F
FP
S
B
E
TimeTuesday, 30 July 201911am - 12pm
LocationRoom 406B
DescriptionOscar-winning Open Shading Language (OSL) is the de-facto open-source standard for digital content creation. OSL has an industry-wide adoption exemplified by its use in renderers from Pixar's RenderMan®, Sony ImageWorks' Arnold, etc., and used in 100+ movies to date.

Intel has been leading the re-architecture of OSL to add SIMD to leverage, for instance, Intel® AVX512 in modern Intel® processors. Intel® SIMD OSL uses Single Program Multiple Data (SPMD), works with existing OSL shaders, and uses OpenMP explicit vectorizations on OSL library functions. The efforts can be broadly summarized in two steps: (i) introduction of vector LLVM IR generation (for JIT) during render-time optimization, and (ii) addition of a batched interface to the default single-point interface in OSL.

This talk focuses on our journey with Pixar on the SIMD OSL release in RenderMan22.5 and its adoption at Pixar Studio for rendering future movies. Since the effort's start in SIGGRAPH 2016, SIMD OSL has been improved to natively support AVX, AVX2, AVX512, and includes enhanced library features, debugging support, and an extensive test framework. These features are now integrated and available from OSL's GitHub repository. SIMD OSL integrated into RenderMan 22.5 includes parallel execution of an entire shading network on Intel® SIMD hardware. It dynamically schedules concurrent operations on 8 or 16 data points in a single CPU instruction depending on available CPU capabilities.

Our performance testing reveals OSL functions are 7x faster on average (up to 16x max) on wide SIMD hardware, such as Intel® AVX512 systems. Pixar has seen up to 2x faster full renders and faster time to first pixel on production assets, with 15% average render time speedups measured on real world production renders using SIMD OSL.

Finally, we will discuss the relationship between a renderer's ability to coherently execute shader networks and SIMD OSL, and how Pixar's next generation RenderMan XPU can further improve results with SIMD OSL.

Speakers: Steena Monteiro, Intel and Max Liani, Pixar

Contact Us