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
94
95
96
97
98
99
100
101
|
Notes on setting up cross compiling in CMake.
The Book Mastering CMake provides a chapter on this which discusses many of the
issues you will encounter. There is also some information online at:
http://www.cmake.org/Wiki/CMake_Cross_Compiling
The key to the cross compiling is the toolchain file. This file specifies the
compilers to use and the paths to look for libraries and headers.
To set the compilers you simply do:
set(CMAKE_C_COMPILER mpicc)
for each compiler that you want to replace, just like you would on the command
line if the configure check finds the wrong version etc. It is then important
that you do not override this on the command line too.
The most important part of the toolchain is setting up the paths that cmake
should look for the libraries, headers and potentially even programs(though that
is typically not the case). To do this you set the variable CMAKE_FIND_ROOT_PATH
to the list of paths that you want cmake to use in any of the FIND_* commands.
These paths are used as the prefix to the standard paths cmake uses for FIND_*.
You will need to make sure that they follow the typical form or <given path>/lib
and/or <given path>/include to ensure cmake will find things there properly.
There is a way to allow people to specify paths that do not follow this standard
that cmake expects, however, it does open up the possibility of allowing cmake
to find incompatible/wrong libraries and headers. There are some variables that
you can set to control where cmake will look for libraries and headers. The
variables:
CMAKE_FIND_ROOT_PATH_MODE_PROGRAM
CMAKE_FIND_ROOT_PATH_MODE_LIBRARY
CMAKE_FIND_ROOT_PATH_MODE_INCLUDE
can be set to "NEVER, ONLY or BOTH. "NEVER" tells cmake to only use the host
environment to find things. This is typically only used for finding programs
since any executable that you need will most likely be run on the host machine
and not the target machine. "Only" will allow cmake to only look in the
specified root paths given in the toolchain file. This is the recommended and
typical setup for libraries and includes. It is the safest way since it excludes
any headers or libraries that may exist in the host machine's environment.
"BOTH" will allow cmake to use the target(cross compile) environment _and_ the
host environment. This can be useful, but it also has risk. Using BOTH allows
for the possibility that cmake will find a library/header in the host
environment when it should only come from the target environment. However, it
does have a valid use.
Apparently the value given to the "PATHS" variable for find_* commands are
considered to be part of the host environment. This causes a problem in Trilinos
especially for TPLs. When trying to find headers and libraries for a TPL we use
the PATHS variable for the first attempt to find a TPL, this path is often
specified by the user in the configure script. Since that PATHS variable is
considered part of the host environment it is ignored in the case of cross
compiling if we set the find setting to ONLY. To allow users to set the paths
for TPLs in the way that we have encouraged them and in the way that they are
familiar we have to use the BOTH setting for libraries and headers.
It is possible to use TPL libraries and headers without using the "BOTH"
setting, but it requires that the path for the TPL be specified in the toolchain
and that the structure of that path take the form that cmake expects which is:
<path given in toolchain>
|----lib (where libraries are kept)
|----include (where headers are kept)
This may work in some cases, and if it is possible to do this it is the safest
method.
The example redstorm toolchain file had to use the "BOTH" setting for libraries
and headers to allow some pre-existing TPL libraries for redstorm to be linked
in.
The last bit of what needs to be set up is the "tryrunresults" file. During a
cross compile cmake make be asked to do a "try run" configure time check. Since
we are compiling for a machine other than what we are currently building on
cmake knows that it likely cannot run the executable so it doesn't even try.
Instead what it does is generate a "tryrunresults" file that needs to be
manually editted to set the variables to what the target machine would use. This
can be a little confusing since it may not be clear what those variables should
be set to on the target platform. Cmake does still build the executable that it
was supposed to run so that you can move it over to the target machine and run
it. You will then have to manually add the results of the executable to the
tryrunresults file though. Once this file has been completely set up you simply
have to tell cmake about it and it will use it to resolve any tryruns that it
comes across during configure time.
Now that we have both the toolchain and tryrunresults files all we need to do to
start cross compiling is to make a configure script like normal and add the
following to it:
-D CMAKE_TOOLCHAIN_FILE=<toolchain file> \
-C <tryrunresults file> \
and then run the new configure script.
you can look at in the redstorm directory for an example toolchain and try run
file.
|
