Meet Us at SC26

This week, we are at SC26 in Chicago. We present three research posters on tensor contractions for multicore CPUs, AWS Trainium2 and AMD’s XDNA2. The posters are on display from Tuesday to Thursday, November 17–19, in the Skyline Ballroom North Pre-Function. Come and talk to us, especially at the poster reception on Tuesday from 5:15 to 7 pm.

TEIR: A Tiled Execution IR for Tensor Contractions

TEIR poster (PDF)

TEIR is an intermediate representation (IR) that describes a tensor contraction as loops around operations on small tiles. Our MLIR-based CPU compiler derives tiling, loop order and parallelization from a few hardware parameters without auto-tuning and maps the tile operations to LIBXSMM kernels. For 24 row-major TCCG contractions, the generated code is on average 1.5–1.9× faster than the next-best backend on CPUs from AMD, NVIDIA and Apple.

Tensor Contractions on AWS Trainium2

Trainium2 poster (PDF)

Each NeuronCore of Trainium2 multiplies matrices on a 128×128 systolic array. Fast kernels must fill the array’s rows and columns and hide data movement behind computation. Our Trainium2 compiler lowers a contraction to TEIR, picks the tiling with a performance model based on measurements and generates a kernel for the Neuron Kernel Interface. On two NeuronCores, the generated kernels reach up to 99% of the FP16 peak for matrix multiplications and up to 93% for four TCCG contractions.

VLIW Scheduling: Timing Is All You Need

XDNA2 poster (PDF)

The XDNA2 neural processing unit (NPU) reaches peak performance only if its VLIW cores issue a BFP16 matrix multiply-accumulate every cycle. Double-buffering the required 2×2 block of output tiles needs eight accumulator registers, but a core has only five. Our hand-scheduled kernel time-multiplexes the four tiles onto two registers and achieves 96% of peak on one core with L1-resident data. For a large matrix multiplication, the whole NPU reaches 21.4 BFP16 TFLOPS and saturates its DRAM bandwidth.

Riptide: Tsunamis in Your Browser

Riptide showing a simulated tsunami off Japan, with earthquake settings, a wave-height legend, and controls for virtual tide gauges.

Riptide is a finite-volume tsunami solver that runs as an interactive web application. Tsunami simulations typically run on compute clusters and require specialized software and careful setup. Riptide lowers this barrier: it runs directly in the browser, and a few clicks are enough to explore tsunami scenarios and build an intuition for how earthquakes can generate tsunamis.

Earthquakes can be placed along subduction zones on a 3D globe, and the resulting tsunami can be observed as it propagates within the selected region. Five presets based on historical earthquakes (Tōhoku 2011, Sumatra-Andaman 2004, Valdivia 1960, Maule 2010 and Alaska 1964) provide ready-made starting points, but users can also select their own region. For a closer analysis, they can place virtual tide gauges and follow the recorded water-level time series while the simulation runs.

Riptide’s simulation core is written in C++ and uses an f-wave solver for the shallow water equations. It is compiled to WebAssembly and runs in a background thread. The globe and the simulated waves are rendered with WebGL2, and the interface is built with React. The simulation runs on bathymetry from the GEBCO 2026 global grid, with fault geometry (depth, strike, dip) taken from the USGS Slab2 subduction model. From the chosen magnitude, the app estimates rupture dimensions using scaling laws and derives the average slip from the seismic moment. The Okada model then uses these parameters to compute the initial vertical displacement of the seafloor.

Riptide was developed as a student project by Mika Brückner, Yannik Köllmann and Jan Vogt as part of the Tsunami Lab in the summer term of 2026.

One XDNA2 Kernel, Many Tensor Shapes

Earlier this year, we published Hello XDNA!, our guide to programming the neural processing units in AMD Ryzen AI chips. We have now updated its XDNA2 example so that key dimensions are supplied at runtime instead of fixed in the assembly code. The same assembly implementation can therefore handle a range of supported tensor shapes.

New step-by-step diagrams make the design easier to follow. They show how the kernel divides the calculation, shares its limited on-chip storage, and overlaps arithmetic with data movement. We also replaced the earlier fixed-size instruction schedule after discovering a subtle conflict at the accumulator write port that could silently discard a write.

Beyond the kernel example, the update expands and corrects our inferred instruction-set tables, fixes the BFP16 format description, improves the assembly listings, and addresses smaller errors and terminology issues throughout the site. Read the updated XDNA guide, check out the full changelog, or explore the sources and benchmarks.


Latest Posts