hyPACK-2013 Mode-2 : Prog. on Computing Systems with GPGPUs

Practical approaches to write efficient programs on GPGPUs focusing on certain class of data-parallel application kernels in the area of Numerical Computations (Numerical Linear Algebra) & Solution of Partial Differential equations; and String Search Algorithms are discussed in Mode-2 programme. Programming approaches on Heterogeneous Computing based on OpenCL, and mixed programming on CPUs & GPUs) and an overview of High-Performance Computing (GPU) GPU Cluster will be discussed. Industry experts may participate and discuss trends on practical programming aspects of performance issues of heterogeneous computing.

Participants will get an opportunity to walk-through and execute some of the programs designed for Mode-1, Mode-2, Mode-3 and Mode-4 of this workshop. To understand scalability and performance of selective scientific and engineering or commercial applications, minor or substantive modification of the hyPACK-2013 software programs may be required. Efforts are on to include State-of-the-Art Multi-Core Processor Systems as well as GPU based Servers in hyPACK-2013 Laboratory Sessions in order to understand performance issues for large-scale application kernels.

Different programming paradigms on host CPU i.e. MPI, OpenMP, Pthreads and CUDA enabled NVIDIA GPUs & Heterogeneous Programming on GPGPUs (OpenCL) are integrated and several example programs for Dense/Sparse Matrix Computations are included.

• GPU APIs (Past /Present) - An Overview : HLSL, Cg, Sh, Brook; CUDA, Brook+, OpenCL programming
• An Overview of GPU Computing - Architecture - CUDA enabled NVIDIA GPUs
• GPU Computing - CUDA Software Development Kit, and Application Programming Interface (API);
• GPGPU : AMD-APP SDK - Hardware Threading Architecture - SDK
• GPGPU - AMD-APP SDK - Multi-threaded Data Parallel Computations - SIMD features
• An Overview of Open Standard language (OpenCL) Programming - Heterogeneous computing
• An Overview of NVIDIA CUDA (Low Level / High Level) APIs & CUBLAS Libraries- Performance issues
• NVIDIA CUDA - Debugging & Data Parallel Primitive Library; CUDA enabled NVIDIA GPU Libraries
• An Overview of basic data rearrangement operations on GPU (Read/Write Access; Data Re-ordering for multi-dimensional data,
• An Overview of HPC GPU Cluster & Programming Paradigms (MPI, Pthreads, OpenMP & CUDA/OpenCL)
• Heterogeneous Computing - Mixed Mode of Programming - GPU Computing and CPU
• Algorithms on Numerical Computations & Effective Ways to Parallelize applications & Performance Issues
• Programming based on CUDA enabled NVIDIA GPUs, OpenCL with Intel TBB, OpenMP, & MPI

• Programming exercises for Numerical Computations based on CUDA enabled NVIDIA, AMD-APP OpenCL Programming
• Tuning & Performance - Use of CUDA enabled NVIDIA GPU Libraries; Memory Bandwidth, Data-access optimization Test Codes
• Numerical Computations (Dense Matrix Computations, Sparse Matrix Computations); Solution of Partial Differential Equations; String Search Algorithms - OpenCL & CUDA enabled NVIDIA GPUs
• Heterogeneous Computing - Mixed Mode of Programming - GPU Computing (CUDA enabled NVIDIA GPUs) & CPU (OpenMP, Intel TBB, MPI) for Scientific Computing Kernels
• Algorithms on Numerical Computations & Effective Ways to Parallelize Applications & Performance Issues
• Tuning & Performance - Selective Application Benchmarks CUDA enabled NVIDIA GPUs & AMD-APP SDK - OpenCL Programming
• Demonstration of Selective Applications using Directives and Compiler Analysis - CUDA enabled NVIDIA GPUs (Fortran & C- lang.)
• Demonstration of Numerical Linear Algebra Kernels on CUDA enabled NVIDIA GPUs & OpenCL enabled AMD-APP Tech.
• Demonstration of Intgrated Numerical Linear Algebra Kernels on Dense Matrix Computations (Open Source Software) on CUDA enabled NVIDIA GPUs & OpenCL enabled AMD-APP Tech.
• Demonstration of Application Perspective - CUDA enabled NVIDIA GPUs; OpenCL & AMD-APP GPUs - SDKs
• GPU Accelerator Programming Model - Compiler Optimizations; Performance Tuning - NVIDIA GPUs - Data Parallelism & Algorithms