1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
|
.TH "MEMPERF" "1" "20 May 2017 (v0.0.1)" "bofc.pl"
.SH "NAME"
\fBmemperf\fR \- memory performance tool
.SH "SYNOPSIS"
\fBmemperf\fR [ \fB\-h\fR | \fB\-v\fR | \fIoptions\fR ]
.SH "OPTIONS"
All options that takes parameters must be passed without spaces. For example,
to set memory block, you should call it like \fB\-b10M\fR as \fB\-b 10M\fR is
invalid.
.TP
\fB\-h\fR
Prints short help on stdout and exits
.TP
\fB\-v\fR
Prints version number on stdout and exits
.PP
Options \fB\-b\fR, \fB\-r\fR and \fB\-l\fR accept number of bytes as a
parameters. jedec format can be used, that is one can pass \fB\-b1024\fR, which
will be equivalent to \fB\-b1K\fR. Another possible values can be \fB\-r1G\fR,
\fB-l15M\fR and \fB-b12345K\fR. 1K equals 1024 bytes.
.TP
\fB\-b\fR \fIblock_size\fR
Size of a single block of memory that will be copied in a burst without any
disturbace. Program will allocate 2 * \fIblock_size\fR of memory that will be
used to perform memory operations (default 16K)
.TP
\fB\-r\fR \fIreport_size\fR
Each time When at least \fIreport_size\fR bytes is copied, program will display
report with bytes copied, time used to copy data and bandwidth information.
Note that program can actually copy more bytes that set in \fIreport_size\fR
depending on the value in \fIblock_size\fR (default 100M)
.TP
\fB\-l\fR \fIcache_size\fR
Size of the l3 cpu cache. Each time \fIblock_size\fR is copied, application
will try to invalidate CPU cache so it does not interfere with benchmark. For
this, application needs to know the CPU cache size. If value is too low, memory
copy may be performed in cache making memory bandwith to be bigger than in it
is, setting this value too big will significally increase time neede to perform
benchmark. This copy is of course not counted (default 1M)
.TP
\fB\-i\fR \fIintervals\fR
Number of test inervals. Basically this defines how many reports will be printed.
.TP
\fB\-m\fR \fImethod\fR
Method to use to copy memory (default memcpy)
.RS
.TP
\fBmemcpy\fR
builtin memcpy function is used
.TP
\fBbbb\fR
byte by byte, a simple for loop is used
.RE
.TP
\fB\-c\fR \fIclock\fR
Clock to use to perform time measurement (default \fBCLOCK_REALTIME\fR if
\fBclock_gettime\fR is available in the system or or \fBclock_t\fR when it is
not)
.RS
.TP
\fBrealtime\fR
POSIX \fBCLOCK_REALTIME\fR will be used. As \fBCLOCK_REALTIME\fR can alter
itself (forward or backward), tests with this clock should be performed only
when it is sure, that time won\'t be changed. The reason to use this instead of
\fBCLOCK_MONOTONIC\fR is that \fBCLOCK_REALTIME\fR is mandatory and
\fBCLOCK_MONOTONIC\fR is an optional extension.
.TP
\fBclock\fR
C89 standard \fBclock_t\fR is used. If you want to build
\fBmemperf\fR for very minimalistic systems that implements only c89 libc, then
you are stuck with this clock. Its resolution may vary from system to system,
which may lead to bad measurements.
.RE
.SH AUTHOR
Michał Łyszczek <michal.lyszczek@bofc.pl>
|
