Hybrid MPI and OpenMP programming tutorial - Prace Training Portal

Likwid Tool Suite• Command line tools for Linux:– works with standard linux 2.6 kernel– supports Intel and AMD CPUs– Supports all compilers whose OpenMP implementation is based onpthreads• Current tools:– likwid-topology: Print thread and cache topology(similar to lstopo from the hwloc package)– likwid-pin: Pin threaded application without touching code– likwid-perfCtr: Measure performance counters (similar to SGI‘s perfex orlipfpm tools)– likwid-features: View and enable/disable hardware prefetchers (Core2only at the moment)– likwid-bench: Low-level benchmark construction toollikwid-topology – Topology information• Based on cpuid information• Functionality:– Measured clock frequency– Thread topology– Cache topology– Cache parameters (-c command line switch)– ASCII art output (-g command line switch)• Currently supported:– Intel Core 2 (45nm + 65 nm)– Intel Nehalem– AMD K10 (Quadcore and Hexacore)–AMDK8Hybrid Parallel ProgrammingSlide73/154Rabenseifner, Hager, JostHybrid Parallel ProgrammingSlide74/154Rabenseifner, Hager, JostOutput of likwid-topologyCPU name: Intel Core i7 processorCPU clock: 2666683826 Hz*************************************************************Hardware Thread Topology*************************************************************Sockets: 2Cores per socket: 4Threads per core: 2-------------------------------------------------------------HWThread Thread Core Socket0 0 0 01 1 0 02 0 1 03 1 1 04 0 2 05 1 2 06 0 3 07 1 3 08 0 0 19 1 0 110 0 1 111 1 1 112 0 2 113 1 2 114 0 3 115 1 3 1-------------------------------------------------------------— skipped —likwid-topology continuedSocket 0: ( 0 1 2 3 4 5 6 7 )Socket 1: ( 8 9 10 11 12 13 14 15 )-------------------------------------------------------------*************************************************************Cache Topology*************************************************************Level: 1Size: 32 kBCache groups: ( 0 1 ) ( 2 3 ) ( 4 5 ) ( 6 7 ) ( 8 9 ) ( 10 11 ) ( 12 13 ) ( 14 15 )-------------------------------------------------------------Level: 2Size: 256 kBCache groups: ( 0 1 ) ( 2 3 ) ( 4 5 ) ( 6 7 ) ( 8 9 ) ( 10 11 ) ( 12 13 ) ( 14 15 )-------------------------------------------------------------Level: 3Size: 8 MBCache groups: ( 0 1 2 3 4 5 6 7 ) ( 8 9 10 11 12 13 14 15 )-------------------------------------------------------------• … and also try the ultra-cool -g option!Hybrid Parallel ProgrammingSlide75/154Rabenseifner, Hager, JostHybrid Parallel ProgrammingSlide76/154Rabenseifner, Hager, Jost