SW acceleration and parallelization

Instigate provides vast spectrum of parallel programming services. Having huge experience with FPGA/ASIC, NVIDIA CUDA, ATI Stream, AMD NUMA architecture programming and more, we offer thorough consulting and R&D services to cover the whole work-flow, starting from problem classification and corresponding platform assortment to development, optimization and testing. Due to the experienced developers team and in-house methodologies/techniques, we offer our Customers reliable and efficient solutions.

Methodology

• Application exploration, problem classification
• Target platform selection
• Target platform specific decomposition
• Implementation
• Target platform specific optimization
• Verification

Massive parallel reconfigurable architecture programming

• Experience in NVIDIA CUDA and ATI Stream
• Up to 200x acceleration over CPU
• Wide range of applications
• Financial forecasting
• Video encoding/decoding
• Image processing
• E-mail Spam filter
• Linear/Quadratic solver
• Basic math functions

• Research of reference implementation
• Parallelization of reference implementation
• Profiling and identification of active part
• HW-SW decomposition and co-design

• Complex real-life designs parallelization and acceleration
• FFTs (with 64 - 256 M transform size)
• Wi-Fi 802.11 g (PHY layer, 54 Mb/s)
• System Level modeling
• Model level parallelization
• Target platform virtual design
• Simulation and bottlenecks detection
• Platform based optimization
• Export to architecture specific proprietary language

Multi-core NUMA architecture CPU programming

• Efficient use of multi-core/cpu Non-Uniform-Memory-Architecture capabilities
• Reducing the number of processing units competing for access to a shared memory bus
• Algorithm specific scheduling on available computing units
• Custom memory distribution considering neighborhood between cores and memory units
• Used NUMA C library (libnuma)
• Example: accelerating TLM simulation engine using NUMA benefits

Overview

Whitepaper