Gentoo Archives: gentoo-soc

From: Andrea Arteaga <andyspiros@×××××.com>
To: gentoo-soc@l.g.o
Subject: [gentoo-soc] Benchmarking suite - Report 10
Date: Wed, 03 Aug 2011 00:00:02
Message-Id: CAPtnn5HaJq4=Nd=sbptX9zgUNR1-Zjj_SN1z5_vupzBe8nvr5g@mail.gmail.com
1 This is the report of the project "Automated benchmark suite for
2 numerical libraries in Gentoo" for the week 25 July-1 August.
3 This is also available on my blog [1].
4
5 == Project description ==
6
7 The project aims to develop a simple yet powerful automated system of
8 benchmarking for numerical libraries. The Gentoo software system
9 provides many implementations of widely used standards such as BLAS,
10 CBLAS, LAPACK, ScaLAPACK and some other numerical library such as
11 FFTW, MKL. The developed tools will aid the system maintener to choose
12 the best suited implementation with respect to the machine hardware
13 and to test the samer implementation or different ones with different
14 compilers, compiler versions and compile flags.
15
16 == Progrees during the week ==
17
18 This week has been devoted to the following features:
19
20 * Management of the dependencies
21 * Improvement of the PBLAS/ScaLAPACK tests and investigation of some
22 BTL-related problem -- see below
23 * 3-Dimensional FFTW tests
24
25 Before this week's work, only packages without unsatisfied
26 dependencies on the main system could be tested. For example, if one
27 did not have cmake installed in the system, he could not test eigen,
28 as it requires cmake to be installed in order to compile. This is not
29 a problem anymore. The dependencies are emerged into the "sandbox"
30 root along with the package to be tested, and the environment is
31 adjusted in order to make use of the software installed. The delay of
32 this report is due to this work, which has required more work as
33 expected -- a small rewriting of the preliminary part of the script
34 has been done --; now the report can be complete.
35
36 Some investigation has been done regarding the input matrices and
37 vectors for the tests. In order to be fair, the tests should produce
38 the same inputfor the libraries, as different input could require
39 different work by the libraries. The linear congruential random number
40 generatorhas been written, along with the new-style matrix generator.
41 This will not completely solve the problem, but it represents a step
42 in the right direction. Now the PBLAS/ScaLAPACK tests use this method,
43 and the results are satisfactory.
44
45 The work with FFTW is finished for now. 1-dimensional, 2-dimensional
46 and 3-dimensional complex discrete Fourier transformations are tested
47 in both forward and backward versions and with the "estimate" and
48 "measure" strategies. If the time will allow it, I will also add sinus
49 and cosinus tranformations.
50
51 == Plan for the next week ==
52
53 During the following week I will implement the tests for METIS. This
54 is a set of executables and libraries that compute graph and mesh
55 partitioning -- along with other numerical functions --, which are
56 widely used in the field of numerical simulations and other scientific
57 computations.
58
59 The project has intensively gained features in the last weeks. After
60 the METIS tests implementation, the quality of the whole project will
61 be tested and improved. A considerable part of the time has been
62 devoted to the BTL library, and more work will be performed in order
63 to submit the new code upstream. Some work has to be also done in
64 order to make the script more user-friendly; this means managing the
65 exceptions that occur sometimes, avoid conflicts between tests already
66 done and the currect ones, reduce the probability of errors. Much
67 documentation has to be written. The ebuild will be splitted in two
68 part, a Gentoo-specific one and a general one containing the modified
69 BTL.
70
71 All this work will eventually provide a finished and usable project,
72 which is my primary objective for the programme. I do not want to
73 provide an incomplete or undocumented software; I rather prefer
74 pospone the inclusion of new features after the programme end, if the
75 time is not enough.
76
77 Best regards
78 Andrea Arteaga
79
80 [1] http://wp.me/pzWEm-8A