Current Academic Research

Strategic Research Partners

University	Research Projects
Northeastern	GPU accelerated biomedical imaging libraries in OpenCL Optimizing GPU compilation in OpenCL GPU architectural simulation Embedded CPU/DSP/GPU design exploration
Virginia Tech	OpenCL and the 13 Dwarfs: Assessment of Computational Motifs in Heterogeneous Computing Environments CU2CL: A CUDA-to-OpenCL Source-to-Source Translator for Multi- and Many-Core Architectures Performance, Power, and Programmability of Heterogeneous Computing Systems NSF Industry & University Cooperative Research Center (I/UCRC) Sponsorship for the Center for High-Performance Reconfigurable Computing
University of Texas - Austin	Framework for Automatic Generation of Max Power and di/dt Stressmarks for Multicore Processors Low Power Arithmetic Units and Compute Primitives
University of Wisconsin – Madison	Optimizing the performance and power of Graphics Processing Units (GPUs) and Accelerated Processing Units (APUs) Dynamic scheduling of multi-purpose accelerators (e.g., GPUs and reconfigurable hardware) in multicore systems Performance tools for heterogeneous architectures including GPUs and vectors Refining the gem5 simulation infrastructure for x86 and GPUs Reducing last-level cache power (& sometimes area)

Current Research

University	Research Projects
Boston University	Numerical embedding of large-scale communication networks for efficient packet forwarding. Study of the relationship between the topology and geometry of naturally arising network graphs.
Brigham Young University	Integrating FPGAs with CPUs and GPUs via fast on-chip data network
Carnegie Mellon University	Scalable and efficient memory subset and interconnect designs for heterogeneous systems
Clemson	Acceleration of spiking neural networks on single-GPU and multi-GPU systems Acceleration of anisotropic diffusion algorithms to remove image noise Molecular dynamics and bioinformatics
Edinburgh	Development of new programming techniques and solutions for future large scale HPC architectures HPC Service provision HPC training Development and optimisation of a wide range of applications across a wide range of architectures, including GPU accelerated systems.
Duke University	CPU/GPU chip architecture with cache-coherent shared virtual memory Extending pointer-based software to run efficiently on CPU/GPU chips
Harvard University	Acceleration of quantum chemistry calculations using GPUs OpenCL port of the Octopus TDDFT code Extending Theano optimizing expression compiler to leverage OpenCL for GPU computing, with application to machine learning, and computer vision.
Johns Hopkins University	Using GPU cores to speed Monte Carlo simulations of microeconomic models Christopher Carroll and a team of students are speeding up the performance of Simulated Method of Moments estimation of Dynamic Stochastic Optimization models of life cycle and buffer stock saving. For the latest public version of their code, click link below: http://econ.jhu.edu/people/ccarroll/SolvingMicroDSOPs/
Frankfurt Institute Advanced Studies	Locality aware task scheduling on large parallel many-core systems
National Chiao-Tung University – Taiwan	Development of rigorous HPC codes using GPU acceleration for lattice QCD, UrQMD, event reconstruction in particle physics, DGEMM and Linpack
New York University	Project One Engineering-scale high-order PDE solvers for hyperbolic and elliptic problems Tools and languages for GPU programming in OpenCL Project Two Hardware/software interaction: How much does the programmer need to make the best use of the hardware without sacrificing portability and development time? Real General purpose GPUs: Can GPUs execute non-GPU-intended software such as scripting language programs?"
Newcastle University	An object-oriented 2D hydraulic model for execution on the GPU Simulation of flooding in urban areas, using GPU to simulate flow at high resolution on a dynamically adaptive grid
Michigan Technological University	APU Parallel Software for Convex Optimization An efficient annotated parallel convex optimization solver and an associated highly optimized library of convex optimization functions for multi-core and GPU computing units.
MIT	Microphotonics Center Industry Consortium
Princeton University	Novel chip architectures for the data center Integration of GPUs into the cloud data center Prefetching and synchronization primitives in order to reduce or hide communication latencies between a CPU and GPU or between different functional units within the GPU. Our target is latency-sensitive (as opposed to bandwidth-oriented) computations. Regression-based design space exploration methods tailored to GPUs. Allow quick navigation of complicated hardware and software parameter spaces, with high-accuracy performance prediction. We are currently expanding our methods to handle a wide variety of GPU platforms.
Queen’s University	OpenCL development of a Finite-Difference Time-Domain Phononic Crystal Simulator
Rice University	Acceleration of discontinuous Galerkin based simulation codes for computational electromagnetics and fluid dynamics. Loo.py: Automatic generation of OpenCL kernels for numerical methods through high-level code transformations.
Stanford	Delite: A framework for building high-performance embedded DSLs targeting heterogeneous systems Framework support for domain-specific optimization, extensibility/interoperability across DSLs, and debugging Heterogeneous target code generation and resource management for systems including discrete & integrated GPUs Optimized DSL kernel generation for OpenCL targets Memory hierarchy and data path optimization to maximize performance on APUs
SUNY – Albany	Research into fundamental failure physics of lead-free solders Failure modeling in lead free solders for failure prediction and reliability simulation
Technical University Munich	Mapping genomic diversity through in silico full mutagenesis
Technical University Munich	Mapping genomic diversity through in silico full mutagenesis
Unicamp	Development of techniques and algorithms to speed up content-based image retrieval (CBIR) tasks on AMD GPUs
University of Athens	Performance Structures Dependability Analysis in Microprocessor Architectures Online Error Detection and Diagnosis in High-Performance Microprocessors
University of Bristol	OpenCL molecular mechanics-based drug docking code (BUDE)
University of British Columbia	Complex data structure supports on GPUs
University of California, Davis	OpenCL Rasterizer: This project is continuing work done at AMD during the summer, developing a complete software rendering pipeline in OpenCL. The continued work on this project is looking at pluging in alternative renderers, e.g. ray shading, and also at using the channel framework developed as part of Fusion.Next. Piko: This project is developing a general framework for task based execution on the GPU, which includes graphics and general purpose compute pipeines.
University of California – San Diego	Novel parallelization tools for multicore and next-generation GPUs New program analyses for discovering and planning for parallelization GPU acceleration for stereo to multiview video conversion
University of Delaware	Backend optimizations that reduce register pressure in GPUs
University of Massachusetts – Amherst	Real-time interactive fluid dynamics simulation of incompressible fluid design and fluid control applications Atmospheric mixing (local on-site weather prediction), jet and internal combustion engine performance and emissions, and vehicle drag reduction.
University of Montreal	Development of a strided n-dimensional array for GPU. https://github.com/inducer/compyte/wiki High-performance image synthesis, using both traditional programmable rasterization pipelines (e.g. shaders), as well as GPGPU APIs such as OpenCL
University of North Texas	Profiling Analysis of cache access in multi-core systems
University of Tennessee – Knoxville	Development of a dense linear algebra library similar to LAPACK but for heterogeneous/hybrid architectures, starting with current "Multicore+GPU" systems. Hybridization methodology where algorithms of interest are split into tasks of varying granularity and their execution scheduled over the available hardware components.
University of Toronto	On-chip sensors for adaptive power control Power aware cache based structure design Thermal modeling of next generation high-end processors
University of Virginia	Porting and Extending the Rodinia Benchmark Suite for OpenCL
University of Washington	GPU acceleration of 3D Modeling from photographs
University of Wisconsin-Madison	Designing and optimizing concurrent runtime environments for heterogeneous architectures