Intel Xeon Phi processor high performance programming / by Jim Jeffers, James Reinders, Avinash Sodani.

Includes bibliographical references and index.

Vendor-supplied metadata.

Machine generated contents note: ch. 1 Introduction -- Introduction to Many-Core Programming -- Trend: More Parallelism -- Why Intel� Xeon Phi["! Processors Are Needed -- Processors Versus Coprocessor -- Measuring Readiness for Highly Parallel Execution -- What About GPUs? -- Enjoy the Lack of Porting Needed but Still Tune! -- Transformation for Performance -- Hyper-Threading Versus Multithreading -- Programming Models -- Why We Could Skip To Section II Now -- For More Information -- ch. 2 Knights Landing Overview -- Overview -- Instruction Set -- Architecture Overview -- Motivation: Our Vision and Purpose -- Summary -- For More Information -- ch. 3 Programming MCDRAM and Cluster Modes -- Programming for Cluster Modes -- Programming for Memory Modes -- Query Memory Mode and MCDRAM Available -- SNC Performance Implications of Allocation and Threading -- How to Not Hard Code the NUMA Node Numbers -- Approaches to Determining What to Put in MCDRAM.

Note continued: Why Rebooting Is Required to Change Modes -- BIOS -- Summary -- For More Information -- ch. 4 Knights Landing Architecture -- Tile Architecture -- Cluster Modes -- Memory Interleaving -- Memory Modes -- Interactions of Cluster and Memory Modes -- Summary -- For More Information -- ch. 5 Intel Omni-Path Fabric -- Overview -- Performance and Scalability -- Transport Layer APIs -- Quality of Service -- Virtual Fabrics -- Unicast Address Resolution -- Multicast Address Resolution -- Summary -- For More Information -- ch. 6 [�]arch Optimization Advice -- Best Performance From 1, 2, or 4 Threads Per Core, Rarely 3 -- Memory Subsystem -- [�]arch Nuances (Tile) -- Direct Mapped MCDRAM Cache -- Advice: Use AVX-512 -- Summary -- For More Information -- ch. 7 Programming Overview for Knights Landing -- To Refactor, or Not to Refactor, That Is the Question -- Evolutionary Optimization of Applications -- Revolutionary Optimization of Applications.

Note continued: Know When to Hold'em and When to Fold'em -- For More Information -- ch. 8 Tasks and Threads -- OpenMP -- Fortran 2008 -- Intel TBB -- hStreams -- Summary -- For More Information -- ch. 9 Vectorization -- Why Vectorize? -- How to Vectorize -- Three Approaches to Achieving Vectorization -- Six-Step Vectorization Methodology -- Streaming Through Caches: Data Layout, Alignment, Prefetching, and so on -- Compiler Tips -- Compiler Options -- Compiler Directives -- Use Array Sections to Encourage Vectorization -- Look at What the Compiler Created: Assembly Code Inspection -- Numerical Result Variations with Vectorization -- Summary -- For More Information -- ch. 10 Vectorization Advisor -- Getting Started with Intel Advisor for Knights Landing -- Enabling and Improving AVX-512 Code with the Survey Report -- Memory Access Pattern Report -- AVX-512 Gather/Scatter Profiler -- Mask Utilization and FLOPS Profiler -- Advisor Roofline Report.

Note continued: Explore AVX-512 Code Characteristics Without AVX-512 Hardware -- Example -- Analysis of a Computational Chemistry Code -- Summary -- For More Information -- ch. 11 Vectorization with SDLT -- What Is SDLT? -- Getting Started -- SDLT Basics -- Example Normalizing 3d Points with SIMD -- What Is Wrong with AOS Memory Layout and SIMD? -- SIMD Prefers Unit-Stride Memory Accesses -- Alpha-Blended Overlay Reference -- Alpha-Blended Overlay With SDLT -- Additional Features -- Summary -- For More Information -- ch. 12 Vectorization with AVX-512 Intrinsics -- What Are Intrinsics? -- AVX-512 Overview -- Migrating From Knights Corner -- AVX-512 Detection -- Learning AVX-512 Instructions -- Learning AVX-512 Intrinsics -- Step-by-Step Example Using AVX-512 Intrinsics -- Results Using Our Intrinsics Code -- For More Information -- ch. 13 Performance Libraries -- Intel Performance Library Overview -- Intel Math Kernel Library Overview.

Note continued: Intel Data Analytics Library Overview -- Together: MKL and DAAL -- Intel Integrated Performance Primitives Library Overview -- Intel Performance Libraries and Intel Compilers -- Native (Direct) Library Usage -- Offloading to Knights Landing While Using a Library -- Precision Choices and Variations -- Performance Tip for Faster Dynamic Libraries -- For More Information -- ch. 14 Profiling and Timing -- Introduction to Knight Landing Tuning -- Event-Monitoring Registers -- Efficiency Metrics -- Potential Performance Issues -- Intel VTune Amplifier XE Product -- Performance Application Programming Interface -- MPI Analysis: ITAC -- HPCToolkit -- Tuning and Analysis Utilities -- Timing -- Summary -- For More Information -- ch. 15 MPI -- Internode Parallelism -- MPI on Knights Landing -- MPI Overview -- How to Run MPI Applications -- Analyzing MPI Application Runs -- Tuning of MPI Applications -- Heterogeneous Clusters -- Recent Trends in MPI Coding.

Note continued: Putting it all Together -- Summary -- For More Information -- ch. 16 PGAS Programming Models -- To Share or not to Share -- Why Use PGAS on Knights Landing? -- Programming with PGAS -- Performance Evaluation -- Beyond PGAS -- Summary -- For More Information -- ch. 17 Software-Defined Visualization -- Motivation for Software-Defined Visualization -- Software-Defined Visualization Architecture -- OpenSWR: OpenGL Raster-Graphics Software Rendering -- Embree: High-Performance Ray Tracing Kernel Library -- OSPRay: Scalable Ray Tracing Framework -- Summary -- Image Attributions -- For More Information -- ch. 18 Offload to Knights Landing -- Offload Programming Model-Using with Knights Landing -- Processors Versus Coprocessor -- Offload Model Considerations -- OpenMP Target Directives -- Concurrent Host and Target Execution -- Offload Over Fabric -- Summary -- For More Information -- ch. 19 Power Analysis -- Power Demand Gates Exascale -- Power 101.

Note continued: Hardware-Based Power Analysis Techniques -- Software-Based Knights Landing Power Analyzer -- ManyCore Platform Software Package Power Tools -- Running Average Power Limit -- Performance Profiling on Knights Landing -- Intel Remote Management Module -- Summary -- For More Information -- ch. 20 Optimizing Classical Molecular Dynamics in LAMMPS -- Molecular Dynamics -- LAMMPS -- Knights Landing Processors -- LAMMPS Optimizations -- Data Alignment -- Data Types and Layout -- Vectorization -- Neighbor List -- Long-Range Electrostatics -- MPI and OpenMP Parallelization -- Performance Results -- System, Build, and Run Configurations -- Workloads -- Organic Photovoltaic Molecules -- Hydrocarbon Mixtures -- Rhodopsin Protein in Solvated Lipid Bilayer -- Coarse Grain Liquid Crystal Simulation -- Coarse-Grain Water Simulation -- Summary -- Acknowledgment -- For More Information -- ch. 21 High Performance Seismic Simulations -- High-Order Seismic Simulations.

Note continued: Numerical Background -- Application Characteristics -- Intel Architecture as Compute Engine -- Highly-Efficient Small Matrix Kernels -- Sparse Matrix Kernel Generation and Sparse/Dense Kernel Selection -- Dense Matrix Kernel Generation: AVX2 -- Dense Matrix Kernel Generation: AVX-512 -- Kernel Performance Benchmarking -- Incorporating Knights Landing's Different Memory Subsystems -- Performance Evaluation -- Mount Merapi -- 1992 Landers -- Summary and Take-Aways -- For More Information -- ch. 22 Weather Research and Forecasting (WRF) -- WRF Overview -- WRF Execution Profile: Relatively Flat -- History of WRF on Intel Many-Core (Intel Xeon Phi Product Line) -- Our Early Experiences with WRF on Knights Landing -- Compiling WRF for Intel Xeon and Intel Xeon Phi Systems -- WRF CONUS12km Benchmark Performance -- MCDRAM Bandwidth -- Vectorization: Boost of AVX-512 Over AVX2 -- Core Scaling -- Summary -- For More Information -- ch. 23 N-Body simulation.

Note continued: Parallel Programming for Noncomputer Scientists -- Step-by-Step Improvements -- N-Body Simulation -- Optimization -- Initial Implementation (Optimization Step 0) -- Thread Parallelism (Optimization Step 1) -- Scalar Performance Tuning (Optimization Step 2) -- Vectorization with SOA (Optimization Step 3) -- Memory Traffic (Optimization Step 4) -- Impact of MCDRAM on Performance -- Summary -- For More Information -- ch. 24 Machine Learning -- Convolutional Neural Networks -- OverFeat-FAST Results -- For More Information -- ch. 25 Trinity Workloads -- Out of the Box Performance -- Optimizing MiniGhost OpenMP Performance -- Summary -- For More Information -- ch. 26 Quantum Chromodynamics -- LQCD -- The QPhiX Library and Code Generator -- Wilson-Dslash Operator -- Configuring the QPhiX Code Generator -- The Experimental Setup -- Results -- Conclusion -- For More Information.