1============================================================== 2 High Performance Computing Linpack Benchmark (HPL) 3 HPL - 2.3 - December 2, 2018 4============================================================== 5 6 History 7 8 - 09/09/00 Public release of Version 1.0 9 10 - 09/27/00 A couple of mistakes in the VSIPL port have been 11 corrected. The tar file as well as the web site were updated 12 on September 27th, 2000. Note that these problems were not 13 affecting the BLAS version of the software in any way. 14 15 - 01/01/04 Version 1.0a 16 The MPI process grid numbering scheme is now an run-time 17 option. 18 The inlined assembly timer routine that caused the compila- 19 tion to fail when using gcc version 3.3 and above has been 20 removed from the package. 21 Various building problems on the T3E have been fixed; Thanks 22 to Edward Anderson. 23 24 - 15/12/04 Version 1.0b 25 Weakness of the pseudo-random matrix generator found for pro- 26 blem sizes being power of twos and larger than 2^15; Thanks 27 to Gregory Bauer. This problem has not been fixed. It is thus 28 currently recommended to HPL users willing to test matrices 29 of size larger than 2^15 to not use power twos. 30 31 When the matrix size is such that one needs > 16 GB per MPI 32 rank, the intermediate calculation (mat.ld+1) * mat.nq in 33 HPL_pdtest.c ends up overflowing because it is done using 34 32-bit arithmetic. This issue has been fixed by typecasting 35 to size_t; Thanks to John Baron. 36 37 - 09/10/08 Version 2.0 38 39 Piotr Luszczek changed to 64-bit RNG, modified files: 40 -- [M] include/hpl_matgen.h 41 -- [M] testing/matgen/HPL_ladd.c 42 -- [M] testing/matgen/HPL_lmul.c 43 -- [M] testing/matgen/HPL_rand.c 44 -- [M] testing/ptest/HPL_pdinfo.c 45 46 For a motivation for the change, see: 47 Dongarra and Langou, ``The Problem with the Linpack 48 Benchmark Matrix Generator'', LAWN 206, June 2008. 49 50 -- [M] testing/ptest/HPL_pdtest.c -- 51 52 Julien Langou changed the test for correctness from 53 ||Ax-b||_oo / ( eps * ||A||_1 * N ) 54 ||Ax-b||_oo / ( eps * ||A||_1 * ||x||_1 ) 55 ||Ax-b||_oo / ( eps * ||A||_oo * ||x||_oo * N ) 56 to the normwise backward error 57 || r ||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N ) 58 See: 59 Nicholas J. Higham, ``Accuracy and Stability of Numerical Algorithms'', 60 Society for Industrial and Applied Mathematics, Philadelphia, PA, USA, 61 Second Edition, pages = xxx+680, ISBN = 0-89871-521-0, 2002. 62 63 Note that in our case || b ||_oo is almost for sure 64 1/2, we compute it anyway. 65 66 - 10/26/2012 Version 2.1 67 68 Piotr Luszczek introduced exact time stamping for HPL_pdgesv(): 69 -- [M] dist/include/hpl_misc.h 70 -- [M] dist/testing/ptest/HPL_pdtest.c 71 72 Piotr Luszczek fixed out-of-bounds access in data spreading functions 73 and exact time stamping for HPL_pdgesv(): 74 -- [M] dist/src/pgesv/HPL_spreadN.c 75 -- [M] dist/src/pgesv/HPL_spreadT.c 76 Thanks to Stephen Whalen from Cray. 77 78 - 02/24/2016 Version 2.2 79 80 Piotr Luszczek added continuous reporting of factorization progress 81 submitted by Intel and make scripts that uses Intel software tools and 82 libraries and their Apple's Mac OS X equivalents. 83 84 - 12/02/2018 Version 2.3 85 86 Piotr Luszczek removed deprecated MPI functions that are no longer 87 supported in some MPI implementations (for example Open MPI 4.0) and 88 replaced them with 89 modern equivalents in HPL_packL(): 90 -- [M] src/comm/HPL_packL.c 91 92 Piotr Luszczek added one digit to the display of performance result 93 and changed display of scaled residual to scientific notation with 94 extra digits in HPL_pdtest(): 95 -- [M] testing/ptest/HPL_pdtest.c 96 97 Piotr Luszczek added support for Autotools configuration packages 98 autoconf and automake: 99 -- [A] Makefile.am 100 -- [A] configure.ac 101 -- [A] acinclude.m4 102 -- [A] src/Makefile.am 103 -- [A] testing/Makefile.am 104