This projects purpose is to compile GNU Octave with some of it's library dependencies consistently using 64-bit indices on Linux systems. The GNU Octave manual describes this problem in more detail.
If all GNU Octave build dependencies are installed, just type the following commands:
git clone https://github.com/octave-de/GNU-Octave-enable-64.git
cd GNU-Octave-enable-64
make -j2 2>&1 | tee log/build.log
./install/bin/octave
In case of any problems, a detailed log of all console output is saved to
log/build.log
this way.
To illustrate the limitation this project addresses, consider the following Octave code, to determine the integer size of the BLAS library used by Octave:
clear all;
N = 2^31;
## The following line requires about 8 GB of RAM!
a = b = ones (N, 1, "single");
c = a' * b
If the BLAS library uses 32-bit integers, an error will be thrown:
error: integer dimension or index out of range for Fortran INTEGER type
Otherwise, if the BLAS library uses 64-bit integers, the result is:
c = 2^31 = 2147483648
Note that the test case above usually requires twice the memory, if a
and b
are not assigned by a = b = ...
. Note further, that the data type "single"
has a precision of about 23 binary bits. In this particular example no
rounding errors occur.
In particular, a Makefile is provided containing all necessary information to compile
- OpenBLAS (0.3.9),
- SuiteSparse (5.7.1),
- QRUPDATE (1.1.2),
- ARPACK-NG (3.7.0), and
- GNU Octave (5.2.0)
using 64-bit indices. To get a quick overview about the library dependencies, the following figure visualizes them top-down: "The above requires all libraries below".
+-------------------------------------------------------+
| GNU Octave |
+-------------+-------------+-------------+-------------+
| | SuiteSparse | QRUPDATE | ARPACK-NG |
| +-------------+-------------+-------------+
| OpenBLAS |
+-------------------------------------------------------+
Notice: It is assumed, that the user of this Makefile is already capable of building the "usual" Octave development version!
Means, that all other GNU Octave build dependencies (e.g. libtool,
gfortran, ...) are properly installed on the system and, even better,
building the "usual" Octave development version runs flawless. Building
this project requires approximately 5 GB disc space and 1 hour,
depending on your system and the number of parallel jobs make -j
.
Debian like distributions (Ubuntu, ...) make SuiteSparse a dependency for GLPK. For this reason, this Makefile can optionally build and install the original GLPK version by calling:
make -j2 glpk 2>&1 | tee log/build_glpk.log
Using this Makefile is especially of interest, if one pursues the following goals:
- No root privileges required.
- No collision with system libraries.
Both aforementioned goals are archived by building and deploying the required libraries in arbitrary directories. The following directories can be set when calling the Makefile:
ROOT_DIR
(default: current directorypwd
)BUILD_DIR
(default:ROOT_DIR)/build
) The build directory for each library.INSTALL_DIR
(default:ROOT_DIR)/install
) The library installation directory/usr
or/usr/local
. If the installation directory is a system directory, the flagSUDO_INSTALL=sudo
ensures that the installation operations are performed with elevated privileges.SRC_CACHE
(default:ROOT_DIR/source-cache
) The location for downloaded library sources.
Two typical examples:
make ROOT_DIR=$HOME/some/path
make BUILD_DIR=/tmp \
INSTALL_DIR=/usr/local SUDO_INSTALL=1 \
SRC_CACHE=$HOME/Downloads
The internal directory structure relative to ROOT_DIR
is:
ROOT_DIR
|-- build # BUILD_DIR
| |-- arpack
| |-- octave
| +-- ...
|-- install # INSTALL_DIR
| |-- bin
| |-- include
| +-- lib
+-- source-cache # SRC_CACHE
|-- arpack-$(VER).tar.gz
+-- ...
All required libraries are built according to this pattern:
- Download the source code archive to
SRC_CACHE
- Extract the source code archive to
BUILD_DIR
- Configure and build the library ensuring 64-bit indices
- Deploy the library in
INSTALL_DIR
Notice: For reducing the required disc space to 1 GB, it is possible to remove the directories
BUILD_DIR
andSRC_CACHE
entirely after a successfully build of this project.
For more information on the topic of building GNU Octave using large indices, see the GNU Octave manual or the GNU Octave wiki.
This project is greatly inspired by a blog post by Richard Calaba, who created one year ahead a GitHub repository for GNU Octave 3.8.
A very similar project to this one is octave-blas64-builder by Mike Miller. A typical case of "I should have known earlier about it!" :wink:
There is a cross-compiling solution, called MXE-Octave. It compiles all dependencies from scratch, thus requires much more disc space and time than this approach.
As all of this project deals with the correct treatment of shared libraries, it follows a short review on how to work with them and debug their usage.
A method to ensure that shared libraries located in a certain folder are found,
is to export the system environment variable LD_LIBRARY_PATH
. In the
example below, the environment variable gets extended by the path /opt/lib
.
export LD_LIBRARY_PATH=/opt/lib:$LD_LIBRARY_PATH
./run-your-program
Another very verbose method to figure out what shared libraries your
system is actually loading and at which paths your system is looking for them,
is exporting the system environment variable LD_DEBUG
:
export LD_DEBUG={files|bindings|libs|versions|help}
./run-your-program
The meaning of the individual values of LD_DEBUG
can be found at
tldp.org.
Some useful tools for checking the shared library dependencies of a program or library or whether the "right" shared library was loaded at runtime by a program are:
- ld - The GNU linker
- ldd - print shared library dependencies
- nm - list symbols from object files
- objdump - display information from object files
- pmap - report memory map of a process
- readelf - Displays information about ELF files.
Assume your program binary is called <PROG>
and at runtime it has the process
ID <PID>
(which can for example be found out with pgrep -l octave
) and a
shared library of interest is called <LIB>
, then the commands can be used as
follows:
ldd {<PROG>|<LIB>}
nm -D {<PROG>|<LIB>}
objdump {<PROG>|<LIB>}
readelf -d {<PROG>|<LIB>}
pmap <PID>
To get a further insight into shared libraries, here are some more advanced references on this topic: