DOCSLIB.ORG

Intel® Threading Building Blocks

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Intel® Threading Building Blocks CERN openlab-Intel MIC / Xeon Phi training Intel® Xeon Phi™ Product Family Intel® Threading Building Blocks Hans Pabst, April 12th 2013 Software and Services Group Intel Corporation Agenda Motivation and Introduction Finding Concurrency Generic Algorithms Task-based Programming Performance Tuning Concurrent Containers Synchronization Primitives Flow Graph 2 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Moore’s Law 90nm 2003 65nm 2005 45nm 2007 32nm 2009 22nm 2011 25 nm 14nm 15nm 2013 10nm Hi-K metal-gate 2015 3-D Tri-gate Shrink 3 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. The Problem Even when we integrate more of the system we still have spare transistors: multicore! Multi-threading: • Reduce or hide latency • Increase throughput Remember Pollack’s rule (Compute ∝ Area) • 4x area gives 2x performance in one core, but 4x performance in 4 cores • In same area multiple wimpy cores provide more compute than one big one How can we program all these cores? 4 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Programmability and Performance “Parallel hardware needs Programmability! parallel programming” Performance GHz Era Multicore Era Manycore Era Time 5 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Didn’t we solve this in the 1990s? Pthreads standard: IEEE 1003.1c-1995 OpenMP standard: 1997 Yes, but… • How to split up work? How to keep caches hot? • How to balance load between threads? • What about nested parallelism (call chain)? Programming with threads is HARD • Atomicity, ordering, and/vs. scalability • Data races, dead locks, etc. 6 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. A Family of Parallel Programming Models Developer Choice Intel® Cilk™ Plus Intel® Threading Domain-Specific Established Research and Building Blocks Libraries Standards Development C/C++ language Widely used C++ Intel® Integrated Message Passing Intel® Concurrent extensions to simplify template library for Performance Interface (MPI) Collections Primitives parallelism parallelism OpenMP* Offload Extensions Intel® Math Kernel Library Intel® Array Building Coarray Fortran Blocks Open sourced Open sourced OpenCL* Intel® SPMD Parallel Also an Intel product Also an Intel product Compiler Choice of high-performance parallel programming models - Libraries for pre-optimized and parallelized functionality - Intel® Cilk™ Plus and Intel® Threading Building Blocks supports composable parallelization of a wide variety of applications. - OpenCL* addresses the needs of customers in specific segments, and provides developers an additional choice to maximize their app performance - MPI supports distributed computation, combines with other models on nodes 7 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Intel® Parallel Studio XE 2013 Phase Productivity Tool Feature Benefit Analyze existing code Easier analysis and performance Advanced Intel® base and find Parallel heuristics, find compute hotspots and opportunities for Design Advisor XE check for parallelization strategies. parallelization. C/C++ and Fortran Advanced Application performance, scalability and Intel® Composer compilers, performance Build & quality for current multicore and future libraries, and parallel Debug XE many-core systems. models Memory & threading error Increases productivity and lowers cost, Advanced Intel® Inspector checking tool for higher by catching memory and threading Verify XE code reliability & quality defects early Removes guesswork, saves time, makes Performance Profiler to Advanced Intel® VTune™ it easier to find performance and optimize performance Tune scalability bottlenecks Combines ease of Amplifier XE and scalability use with deeper insights. 8 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Intel® Threading Building Blocks С++ Library for parallel programming • Takes care of managing multitasking Runtime library • Scalability to available number of threads Cross-platform • Windows, Linux, Mac OS* and others http://threadingbuildingblocks.org/ 9 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. License, and more… GPL with “runtime exception” You can use it commercially without the need to disclose your source code. • Drives C++ compiler features • Regular updates, and maintenance • Community preview features* • Commercially aligned releases • Premier support * You are always invited to provide feedback, to share your issues, and by the way help other this way. 10 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Intel® Threading Building Blocks Concurrent Containers Generic Parallel Algorithms Concurrent access, and a scalable Efficient scalable way to exploit the alternative to containers that are power of multi-core without having externally locked for thread-safety to start from scratch TBB Flow Graph Thread Local Storage Task scheduler Scalable implementation of thread-local data The engine that empowers parallel that supports infinite number of TLS algorithms that employs task- Synchronization Primitives stealing to maximize concurrency User-level and OS wrappers for mutual exclusion, ranging from atomic Miscellaneous Threads operations to several flavors of mutexes Thread-safe timers OS API wrappers and condition variables Memory Allocation Per-thread scalable memory manager and false-sharing free allocators 11 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Agenda Motivation and Introduction Finding Concurrency Generic Algorithms Task-based Programming Performance Tuning Concurrent Containers Synchronization Primitives Flow Graph 12 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Finding Concurrency • Design patterns – An encoded expertise to capture that “quality without a name” that distinguishes truly excellent designs – A small number of patterns can support a wide range of applications • Parallel pattern: commonly occurring combination of task distribution and data access Identify and use parallel patterns! Examples: reduction, or pipeline 13 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Parallel Patterns • Stencil • Superscalar • Reduce sequence • Speculative • Pack & selection Expand • Partition • Map • Nest • Scan and Recurrence • Gather/scatter • Search & Match • Pipeline * http://software.intel.com/sites/products/documentation/hpc/composerxe/en -us/2011Update/tbbxe/Design_Patterns.pdf http://www.cs.cmu.edu/afs/cs.cmu.edu/Web/People/guyb/papers/Ble90.pdf 14 4/12/2013 Copyright© 2012, Intel Corporation. All rights reserved. *Other brands and names are the property of their respective owners. Reference Details a pattern language for parallel algorithm design Examples in MPI, OpenMP and Java are given Represents the author's hypothesis for how programmers think about parallel programming Patterns for Parallel Programming, Timothy G. Mattson, Beverly A. Sanders, Berna L. Massingill, Addison-Wesley, 2005, ISBN 0321228111 15 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Finding Concurrency From serial code* • Use profiler, such as Intel® VTune™ Amplifier XE • Identify hotspots in an application • Examine the code in hotspots • Determine whether the tasks within the hotspots can be executed independently From a design document • Examine the design components, services, etc. • Find components that contain independent operations * Note, to start from serial code may never lead to the best known parallel algorithm. 16 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Intel® Advisor XE • Tool for what-if analysis – Modeling: use code annotations to introduce parallelism – Evaluation: estimate the effect e.g. the speedup – GUI-driven assistant (5 steps) • Productivity and Safety – Parallel correctness is checked based on a correct program – Non-intrusive API • It’s not auto-parallelization • It’s not modifying the code 17 Copyright© 2012, Intel Corporation. All rights reserved. 4/12/2013 *Other brands and names are the property of their respective owners. Example: Quicksort with Intel Advisor template<typename I> void serial_qsort(I begin, I end) { typedef typename std::iterator_traits<I>::value_type T; if (begin != end) { const I pivot = end - 1; const I middle = std::partition(begin, pivot, std::bind2nd(std::less<T>(), *pivot)); std::swap(*pivot, *middle); ANNOTATE_SITE_BEGIN(Parallel Region); ANNOTATE_TASK_BEGIN(Left Partition); serial_qsort(begin, middle), ANNOTATE_TASK_END(Left Partition); ANNOTATE_TASK_BEGIN(Right Partition); serial_qsort(middle
Load more

Recommended publications
  • Other Apis What’S Wrong with Openmp?
    Threaded Programming Other APIs What’s wrong with OpenMP? • OpenMP is designed for programs where you want a fixed number of threads, and you always want the threads to be consuming CPU cycles. – cannot arbitrarily start/stop threads – cannot put threads to sleep and wake them up later • OpenMP is good for programs where each thread is doing (more-or-less) the same thing. • Although OpenMP supports C++, it’s not especially OO friendly – though it is gradually getting better. • OpenMP doesn’t support other popular base languages – e.g. Java, Python What’s wrong with OpenMP? (cont.) Can do this Can do this Can’t do this Threaded programming APIs • Essential features – a way to create threads – a way to wait for a thread to finish its work – a mechanism to support thread private data – some basic synchronisation methods – at least a mutex lock, or atomic operations • Optional features – support for tasks – more synchronisation methods – e.g. condition variables, barriers,... – higher levels of abstraction – e.g. parallel loops, reductions What are the alternatives? • POSIX threads • C++ threads • Intel TBB • Cilk • OpenCL • Java (not an exhaustive list!) POSIX threads • POSIX threads (or Pthreads) is a standard library for shared memory programming without directives. – Part of the ANSI/IEEE 1003.1 standard (1996) • Interface is a C library – no standard Fortran interface – can be used with C++, but not OO friendly • Widely available – even for Windows – typically installed as part of OS – code is pretty portable • Lots of low-level control over behaviour of threads • Lacks a proper memory consistency model Thread forking #include <pthread.h> int pthread_create( pthread_t *thread, const pthread_attr_t *attr, void*(*start_routine, void*), void *arg) • Creates a new thread: – first argument returns a pointer to a thread descriptor.
    [Show full text]
  • Beyond BIOS Developing with the Unified Extensible Firmware Interface
    Digital Edition Digital Editions of selected Intel Press books are in addition to and complement the printed books. Click the icon to access information on other essential books for Developers and IT Professionals Visit our website at www.intel.com/intelpress Beyond BIOS Developing with the Unified Extensible Firmware Interface Second Edition Vincent Zimmer Michael Rothman Suresh Marisetty Copyright © 2010 Intel Corporation. All rights reserved. ISBN 13 978-1-934053-29-4 This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold with the understanding that the publisher is not engaged in professional services. If professional advice or other expert assistance is required, the services of a competent professional person should be sought. Intel Corporation may have patents or pending patent applications, trademarks, copyrights, or other intellectual property rights that relate to the presented subject matter. The furnishing of documents and other materials and information does not provide any license, express or implied, by estoppel or otherwise, to any such patents, trademarks, copyrights, or other intellectual property rights. Intel may make changes to specifications, product descriptions, and plans at any time, without notice. Fictitious names of companies, products, people, characters, and/or data mentioned herein are not intended to represent any real individual, company, product, or event. Intel products are not intended for use in medical, life saving, life sustaining, critical control or safety systems, or in nuclear facility applications. Intel, the Intel logo, Celeron, Intel Centrino, Intel NetBurst, Intel Xeon, Itanium, Pentium, MMX, and VTune are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.
    [Show full text]
  • Intel Advisor for Dgpu Intel® Advisor Workflows
    Profile DPC++ and GPU workload performance Intel® VTune™ Profiler, Advisor Vladimir Tsymbal, Technical Consulting Engineer, Intel, IAGS Agenda • Introduction to GPU programming model • Overview of GPU Analysis in Intel® VTune Profiler • Offload Performance Tuning • GPU Compute/Media Hotspots • A DPC++ Code Sample Analysis Demo • Using Intel® Advisor to increase performance • Offload Advisor discrete GPUs • GPU Roofline for discrete GPUs Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. 2 Intel GPUs and Programming Model Gen9 Application Workloads • Most common Optimized Middleware & Frameworks in mobile, desktop and Intel oneAPI Product workstations Intel® Media SDK Direct Direct API-Based Gen11 Programming Programming Programming • Data Parallel Mobile OpenCL platforms with C API C++ Libraries Ice Lake CPU Gen12 Low-Level Hardware Interface • Intel Xe-LP GPU • Tiger Lake CPU Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. 3 GPU Application Analysis GPU Compute/Media Hotspots • Visibility into both host and GPU sides • HW-events based performance tuning methodology • Provides overtime and aggregated views GPU In-kernel Profiling • GPU source/instruction level profiling • SW instrumentation • Two modes: Basic Block latency and memory access latency Identify GPU occupancy and which kernel to profile. Tune a kernel on a fine grain level Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. 4 GPU Analysis: Aggregated and Overtime Views Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others.
    [Show full text]
  • Getting Started with Oneapi
    ONEAPI SINGLE PROGRAMMING MODEL TO DELIVER CROSS-ARCHITECTURE PERFORMANCE Getting started with oneAPI March 2020 How oneAPIaddresses our Heterogeneous World? DIVERSE WORKLOADS DEMAND DIVERSEARCHITECTURES The future is a diverse mix of scalar, vector, matrix, andspatial architectures deployed in CPU, GPU, AI, FPGA, and other accelerators. Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. Getting started withoneAPI 4 *Other names and brands may be claimed as the property of others. CHALLENGE: PROGRAMMING IN A HETEROGENEOUSWORLD ▷ Diverse set of data-centric hardware ▷ No common programming language or APIs ▷ Inconsistent tool support across platforms ▷ Proprietary solutions on individual platforms S V M S ▷ Each platform requires unique software investment Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. Getting started withoneAPI 5 *Other names and brands may be claimed as the property of others. INTEL'S ONEAPI CORECONCEPT ▷ Project oneAPI delivers a unified programming model to simplify development across diverse architectures ▷ Common developer experience across SVMS ▷ Uncompromised native high-level language performance ▷ Support for CPU, GPU, AI, and FPGA ▷ Unified language and libraries for ▷ Based on industry standards and expressing parallelism open specifications https://www.oneapi.com/spec/ Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. Getting started withoneAPI 6 *Other names and brands may be claimed as the property of others. ONEAPI FOR CROSS-ARCHITECTUREPERFORMANCE Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. Getting started withoneAPI 7 *Other names and brands may be claimed as the property of others. WHAT IS DATA PARALLELC++? WHAT ISDPC++? The language is: C++ + SYCL https://www.khronos.org/sycl/ + Additional Features such as..
    [Show full text]
  • Michael Steyer Technical Consulting Engineer Intel Architecture, Graphics & Software Analysis Tools
    Michael Steyer Technical Consulting Engineer Intel Architecture, Graphics & Software Analysis Tools Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. Aspects of HPC/Throughput Application Performance What are the Aspects of Performance Intel Hardware Features Multi-core Intel® Omni Intel® Optane™ Intel® Advanced Intel® Path HBM DC persistent Vector Xeon® Extensions 512 Architecture memory (Intel® AVX-512) processor Distributed memory Memory I/O Threading CPU Core Message size False Sharing File I/O Threaded/serial ratio uArch issues (IPC) Rank placement Access with strides I/O latency Thread Imbalance Vectorization Rank Imbalance Latency I/O waits RTL overhead FPU usage efficiency RTL Overhead Bandwidth System-wide I/O (scheduling, forking) Network Bandwidth NUMA Synchronization Cluster Node Core Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. IntelWhat Parallel are the Studio Aspects Tools covering of Performance the Aspects Intel Hardware Features Multi-core Intel® Intel® Omni Intel® Optane™ Advanced Intel®Path DC persistent Intel® Vector HBM Extensions Architectur Intel® VTune™memory AmplifierXeon® processor 512 (Intel® Tracee Intel®AVX-512) DistributedAnalyzer memory Memory I/O Threading AdvisorCPU Core Messageand size False Sharing File I/O Threaded/serial ratio uArch issues (IPC) Rank placement Access with strides I/O latency Thread Imbalance Vectorization RankCollector Imbalance Latency I/O waits RTL overhead FPU usage efficiency RTL Overhead Bandwidth System-wide I/O (scheduling, forking) Network Bandwidth NUMA Synchronization Cluster Node Core Optimization Notice Copyright © 2020, Intel Corporation. All rights reserved.
    [Show full text]
  • Hands-On Intel® Software Development & Oneapi WORKSHOP
    Hands-on Intel® Software Development & oneAPI WORKSHOP May 26-27, 2020 Scandic Solli, Parkveien 68 Box 2458 Solli, 0202 Oslo AGENDA DAY 1 - Technical Computing & Developer Tools - May 26 Timing Sessions 08:30 – 09:00 Registration & Light breakfast Part 1: Coding for maximum performance using the new Intel® Parallel Studio XE 2020 A refresher on the Intel® Hardware Architecture for Software Developers and Architects This session will offer in-depth insights into the current and future Intel® hardware platforms tailored to the 09:00 -09:45 needs of software developers, software architects, HPC and AI experts. We will cover the latest Intel® processors and the future Intel® GPU architecture. Developing code for Intel® architecture: how to achieve maximum performance using the new Intel® Parallel Studio XE 2020 09:45 – 10:30 Learn how Intel® Software Development Tools will help you to achieve optimal performance in your High Performance Computing, Artificial Intelligence ,and IoT projects. Includes a look at the new Intel® Parallel Studio XE 2020 tools which are designed to take advantage of the latest generation of Intel processors. 10:30 – 11:00 Coffee Break How to optimize and maximize code performance Learn how to use some of the advanced features of Intel® VTune™ Amplifier profile your applications. See how you can use event-based and architectural analysis to fine-tune your code so that it is taking full 11:00 – 12:00 advantage of the latest processor features of the target CPU. Learn how to use Intel Advisor, a powerful tool for tracking down and solving vectorization problems. In this session we will demonstrate how the Intel Advisor vector analysis and associated Roofline Model can be used to identify and help fixing vectorization problems.
    [Show full text]
  • Evaluating Techniques for Parallelization Tuning in MPI, Ompss and MPI/Ompss
    Evaluating techniques for parallelization tuning in MPI, OmpSs and MPI/OmpSs Advisors: Author: Prof. Jesús Labarta Vladimir Subotic´ Prof. Mateo Valero Prof. Eduard Ayguade´ A THESIS SUBMITTED IN FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF Doctor per la Universitat Politècnica de Catalunya Departament d’Arquitectura de Computadors Barcelona, 2013 i Abstract Parallel programming is used to partition a computational problem among multiple processing units and to define how they interact (communicate and synchronize) in order to guarantee the correct result. The performance that is achieved when executing the parallel program on a parallel architec- ture is usually far from the optimal: computation unbalance and excessive interaction among processing units often cause lost cycles, reducing the efficiency of parallel computation. In this thesis we propose techniques oriented to better exploit parallelism in parallel applications, with especial emphasis in techniques that increase asynchronism. Theoretically, this type of parallelization tuning promises multiple benefits. First, it should mitigate communication and synchro- nization delays, thus increasing the overall performance. Furthermore, parallelization tuning should expose additional parallelism and therefore increase the scalability of execution. Finally, increased asynchronism would allow more flexible communication mechanisms, providing higher toler- ance to slower networks and external noise. In the first part of this thesis, we study the potential for tuning MPI par- allelism. More specifically, we explore automatic techniques to overlap communication and computation. We propose a speculative messaging technique that increases the overlap and requires no changes of the orig- inal MPI application. Our technique automatically identifies the applica- tion’s MPI activity and reinterprets that activity using optimally placed non-blocking MPI requests.
    [Show full text]
  • Introduction to Intel Performance Tools Part
    Introduction to Intel Performance Tools Part 1/2 Doug Roberts SHARCNET / COMPUTE CANADA Intel® Performance Tools o Intel Advisor - Optimize Vectorization and Thread Prototyping for C, C++, Fortran o Intel Inspector - Easy-to-use Memory and Threading Error Debugger for C, C++, Fortran o Intel Vtune Amplifier - Serial/Threaded Performance Profiler for C, C++, Fortran, Mixed Python o Intel Trace Analyzer and Collector - Understand MPI application behavior for C, C++, Fortran, OpenSHMEM o Intel Distribution for Python - High-performance Python powered by native Intel Performance Libraries Intel® Parallel Studio XE – Cluster Edition https://software.intel.com/en-us/parallel-studio-xe o Intel Advisor* https://software.intel.com/en-us/intel-advisor-xe o Intel Inspector* https://software.intel.com/en-us/intel-inspector-xe o Intel Vtune Amplifier* https://software.intel.com/en-us/intel-vtune-amplifier-xe o Intel Trace Analyzer and Collector* https://software.intel.com/en-us/intel-trace-analyzer o Intel Distribution for Python https://software.intel.com/en-us/distribution-for-python * Product Support → Training, Docs, Faq, Code Samples Initializating the Components – The Intel Way ssh graham.sharcnet.ca cd /opt/software/intel/18.0.1/parallel_studio_xe_2018.1.038 source psxevars.sh → linux/bin/compilervars.sh → clck_2018/bin/clckvars.sh → itac_2018/bin/itacvars.sh → inspector_2018/inspxe-vars.sh → vtune_amplifier_2018/amplxe-vars.sh → advisor_2018/advixe-vars.sh Examples ls /opt/software/intel/18.0.1/parallel_studio_xe_2018.1.038/samples_2018/en
    [Show full text]
  • Concurrent Cilk: Lazy Promotion from Tasks to Threads in C/C++
    Concurrent Cilk: Lazy Promotion from Tasks to Threads in C/C++ Christopher S. Zakian, Timothy A. K. Zakian Abhishek Kulkarni, Buddhika Chamith, and Ryan R. Newton Indiana University - Bloomington, fczakian, tzakian, adkulkar, budkahaw, [email protected] Abstract. Library and language support for scheduling non-blocking tasks has greatly improved, as have lightweight (user) threading packages. How- ever, there is a significant gap between the two developments. In previous work|and in today's software packages|lightweight thread creation incurs much larger overheads than tasking libraries, even on tasks that end up never blocking. This limitation can be removed. To that end, we describe an extension to the Intel Cilk Plus runtime system, Concurrent Cilk, where tasks are lazily promoted to threads. Concurrent Cilk removes the overhead of thread creation on threads which end up calling no blocking operations, and is the first system to do so for C/C++ with legacy support (standard calling conventions and stack representations). We demonstrate that Concurrent Cilk adds negligible overhead to existing Cilk programs, while its promoted threads remain more efficient than OS threads in terms of context-switch overhead and blocking communication. Further, it enables development of blocking data structures that create non-fork-join dependence graphs|which can expose more parallelism, and better supports data-driven computations waiting on results from remote devices. 1 Introduction Both task-parallelism [1, 11, 13, 15] and lightweight threading [20] libraries have become popular for different kinds of applications. The key difference between a task and a thread is that threads may block|for example when performing IO|and then resume again.
    [Show full text]
  • Intel® Software Products Highlights and Best Practices
    Intel® Software Products Highlights and Best Practices Edmund Preiss Business Development Manager Entdecken Sie weitere interessante Artikel und News zum Thema auf all-electronics.de! Hier klicken & informieren! Agenda • Key enhancements and highlights since ISTEP’11 • Industry segments using Intel® Software Development Products • Customer Demo and Best Practices Copyright© 2012, Intel Corporation. All rights reserved. 2 *Other brands and names are the property of their respective owners. Key enhancements & highlights since ISTEP’11 3 All in One -- Intel® Cluster Studio XE 2012 Analysis & Correctness Tools Shared & Distributed Memory Application Development Intel Cluster Studio XE supports: -Shared Memory Processing MPI Libraries & Tools -Distributed Memory Processing Compilers & Libraries Programming Models -Hybrid Processing Copyright© 2012, Intel Corporation. All rights reserved. *Other brands and names are the property of their respective owners. Intel® VTune™ Amplifier XE New VTune Amplifier XE features very well received by Software Developers Key reasons : • More intuitive – Improved GUI points to application inefficiencies • Preconfigured & customizable analysis profiles • Timeline View highlights concurrency issues • New Event/PC counter ratio analysis concept easy to grasp Copyright© 2012, Intel Corporation. All rights reserved. *Other brands and names are the property of their respective owners. Intel® VTune™ Amplifier XE The Old Way versus The New Way The Old Way: To see if there is an issue with branch misprediction, multiply event value (86,400,000) by 14 cycles, then divide by CPU_CLK_UNHALTED.THREAD (5,214,000,000). Then compare the resulting value to a threshold. If it is too high, investigate. The New Way: Look at the Branch Mispredict metric, and see if any cells are pink.
    [Show full text]
  • Intel® Offload Advisor
    NHR@ZIB - Intel oneAPI Workshop, 2-3 March 2021 Intel® Advisor Offload Modelling and Analysis Klaus-Dieter Oertel Intel® Advisor for High Performance Code Design Rich Set of Capabilities Offload Modelling Design offload strategy and model performance on GPU. One Intel Software & Architecture (OISA) 2 Agenda ▪ Offload Modelling ▪ Roofline Analysis – Recap ▪ Roofline Analysis for GPU code ▪ Flow Graph Analyzer One Intel Software & Architecture (OISA) 3 Offload Modelling 4 Intel® Advisor - Offload Advisor Find code that can be profitably offloaded Starting from an optimized binary (running on CPU): ▪ Helps define which sections of the code should run on a given accelerator ▪ Provides performance projection on accelerators One Intel Software & Architecture (OISA) 5 Intel® Advisor - Offload Advisor What can be expected? Speedup of accelerated code 8.9x One Intel Software & Architecture (OISA) 6 Modeling Performance Using Intel® Advisor – Offload Advisor Baseline HW (Programming model) Target HW 1. CPU (C,C++,Fortran, Py) CPU + GPU measured measured estimated 1.a CPU (DPC++, OCL, OMP, CPU + GPU measured “target=host”) measured estimated 2 CPU+iGPU (DPC++, OCL, OMP, CPU + GPU measured “target=offload”) measured Estimated Optimization Notice Copyright © 2019, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. Modeling Performance Using Intel® Advisor – Offload Advisor Region X Region Y Execution time on baseline platform (CPU) • Execution time on accelerator. Estimate assuming bounded exclusively
    [Show full text]
  • Tips and Tricks: Designing Low Power Native and Webapps
    Tips and Tricks: Designing low power Native and WebApps Harita Chilukuri and Abhishek Dhanotia Acknowledgements • William Baughman for his help with the browser analysis • Ross Burton & Thomas Wood for information on Tizen Architecture • Tom Baker, Luis “Fernando” Recalde, Raji Shunmuganathan and Yamini Nimmagadda for reviews and comments 2 tizen.org Power – Onus lies on Software too! Use system resources to provide best User WEBAPPS Experience with minimum power NATIVE APPS RUNTIME Interfaces with HW components, DRIVERS / MIDDLEWARE Independent device power management Frequency Governors, CPU Power OS Management ACPI/RTPM Provides features for low power HARDWARE Clock Gating, Power Gating, Sleep States 3 tizen.org Power – Onus lies on Software too! • A single bad application can lead to exceeding power budget • Hardware and OS provide many features for low power – Apps need to use them smartly to improve power efficiency • Good understanding of underlying system can help in designing better apps Images are properties of their respective owners 4 tizen.org Agenda Tips for Power Tools General Low Power Q&A & Metrics guidelines Applications 5 tizen.org Estimating Power - Metrics • CPU utilization • Memory bandwidth • CPU C and P state residencies • Device D states - For non-CPU components • S states – system sleep states • Wakeups, interrupts Soft metrics can help tune the application for optimal power 6 tizen.org Estimating Power - Tools • CPU utilization – Vmstat, Top – VTune, Perf for CPU cycles • Memory bandwidth – Vtune • CPU C and P states, Device D states – VTune, Powertop • Wakeups, Interrupts, Timers – Powertop, /proc stats • Tracing tool in Chrome browser ** VTune is an Intel product and can be purchased, others are publicly available Linux tools * Other names and brands may be claimed as the property of others.
    [Show full text]