PGAS Programming with UPC and Fortran Coarrays

More documents

Recommendations

Info

•19/12/2012Vector Programming• Exploit hardware support for pipelining– and for fast data access• Early supercomputers were often vector systems– Such as the Cray-1• Allowed operations on vectors– A vector is a a series of values– e.g., a section of a Fortran array• Typical vector loopDO i = 1, ny(i) = a*x(i) + y(i)END DOParallel Programming Languages 5Vector Multiply• Multiply process is made up of a number of stages• Vector hardware allows stages to work independently and topass results to each other in an “assembly line” manner• Start-up cost as pipeline fills, but then a result every cycleR(5)R(4)R(3)R(2)R(1)SerialTimeR(5)R(4)R(3)R(2)R(1)VectorTimeParallel Programming Languages 6•3
•19/12/2012Vectorisation• Sometimes required restructuring of loops to allow efficientvectorisation• Directives used to provide information to the compiler aboutwhether a particular operation was vectorisable• Compilers became increasingly good at spottingopportunities for vectorisation• Vector supercomputers became less popular as parallelcomputing grew• However, many modern CPUs contain vector-like features– e.g., Interlagos Opteron processors in Cray XEParallel Programming Languages 7Data Parallel• Processors perform similar operations across data elements in an array• Higher level programming paradigm, characterised by:– single-threaded control– global name space– loosely synchronous processes– parallelism implied by operations applied to data– compiler directives• Data parallel languages: generally serial language (e.g., Fortran 90) plus– compiler directives (e.g., for data distribution)– first class language constructs to support parallelism– new intrinsics and library functions• Paradigm well suited to a number of early (SIMD) parallel computers– Connection Machine, DAP, MasPar,…Parallel Programming Languages 8•4
Page 1 and 2: PGAS Programming withUPC and Fortra
Page 3: •19/12/2012Contents• A Little B
Page 7 and 8: •19/12/2012Contents• A Little B
Page 9 and 10: •19/12/2012Shared Memory• Multi
Page 11 and 12: •19/12/2012Single-Sided Communica
Page 13 and 14: •19/12/2012PGAS Languages• The
Page 15 and 16: •19/12/2012Hybrid Approaches• U
Page 17 and 18: •19/12/2012References• PRACE-PP
Page 19 and 20: UPCUnified Parallel CParallel exten
Page 21 and 22: Private vs. shared memory spaceConc
Page 23 and 24: Shared array distributionIf a share
Page 25 and 26: Example: vector addition (1)#define
Page 27 and 28: “Static” keywordensures shared
Page 29 and 30: References• UPC Language Specific
Page 31 and 32: Sample PBS script for XE6#!/bin/bas
Page 33 and 34: Brief recap… private and shared d
Page 35 and 36: Multi-dimensional dataUPC can distr
Page 37 and 38: Multi-dimensional data - Case 1shar
Page 39 and 40: Multi-dimensional data - Case 3slab
Page 41 and 42: Static vs. dynamic compilation (3)s
Page 43 and 44: Debugging barriersthe optional valu
Page 45 and 46: Example: Vector addition (3/3)Advan
Page 47 and 48: Improved data distributionCAB00 0 0
Page 49 and 50: C pointersA pointer in C is a data
Page 51 and 52: UPC pointersUPC pointers have three
Page 53 and 54: Pointer castingcasting a shared poi
Page 55 and 56:
Non-collective dynamic allocation (
Page 57 and 58:
Creating locksinitial state of a ne
Page 59 and 60:
Example: Dining Philosophers (contd
Page 61 and 62:
Broadcastshared[] int A[2];shared[2
Page 63 and 64:
Final summarydata distribution & wo
Page 65 and 66:
12/19/2012The FortranProgramming Mo
Page 67 and 68:
12/19/2012Fortran 2008 coarray mode
Page 69 and 70:
12/19/2012Fortran coarraysBasic Fea
Page 71 and 72:
12/19/2012What are coarrays?• Arr
Page 73 and 74:
12/19/2012Making remote references
Page 75 and 76:
12/19/2012Synchronisation• code e
Page 77 and 78:
12/19/2012Example2: Calculate densi
Page 79 and 80:
12/19/2012Exercise Session 1• Loo
Page 81 and 82:
12/19/2012OpenMP implementationmemo
Page 83 and 84:
12/19/2012MPIprocess 0memoryprocess
Page 85 and 86:
12/19/2012High Performance Fortran
Page 87 and 88:
12/19/2012UPCthreadthreadthreadmemo
Page 89 and 90:
12/19/2012Fortran coarray modelproc
Page 91 and 92:
12/19/2012Parallel Programmingwith
Page 93 and 94:
12/19/2012Edge detection and pictur
Page 95 and 96:
12/19/2012More Coarray FeaturesPara
Page 97 and 98:
12/19/20122D Data• However, image
Page 99 and 100:
12/19/2012Codimensions: What They M
Page 101 and 102:
12/19/2012this_image() & image_inde
Page 103 and 104:
12/19/2012Allocatable Coarrays• C
Page 105 and 106:
12/19/2012Coarrays and Procedures
Page 107 and 108:
12/19/2012Summary• Coarrays with
Page 109 and 110:
segmentsegment12/19/2012More on Syn
Page 111 and 112:
12/19/2012Synchronisation mistakes
Page 113 and 114:
12/19/2012Other Synchronisation•
Page 115 and 116:
12/19/2012More on Cosubscripts• i
Page 117 and 118:
12/19/2012Coarray Technical Specifi
Page 119 and 120:
12/19/2012TS: Collective intrinsic
Page 121 and 122:
12/19/2012Experiences withCoarraysP
Page 123 and 124:
12/19/2012Implementations we have u
Page 125 and 126:
12/19/2012Adding coarrays to existi
Page 127 and 128:
12/19/2012AllReduce (everyone gets,
Page 129 and 130:
12/19/2012Synchronisation (one to m
Page 131 and 132:
12/19/2012Coarray Benchmark Suite
Page 133 and 134:
microseconds12/19/2012Pingpong (sma
Page 135 and 136:
12/19/20123D Halo Swap on XE6 (weak
Page 137 and 138:
12/19/2012Data Structuresuse,intrin
Page 139 and 140:
12/19/2012Solve the puzzles and wri
Page 141 and 142:
12/19/2012Distributed remote gather
Page 143 and 144:
Performance (TFlop/s)12/19/2012HIME
Page 145 and 146:
12/19/2012References• http://lacs
show all

PGAS Programming with UPC and Fortran Coarrays

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?