8a1f9ddb2c
Build the default package for SSE2 and add a SSE3 subpackage on x86_64 Apply patch (and arch defs.) to build on s390 and s390x (Dan Horák) Fix-up build on s390 and s390x (Christian Bornträger)
69 lines
3.6 KiB
Plaintext
69 lines
3.6 KiB
Plaintext
Notes on the Fedora version of ATLAS
|
|
|
|
by Quentin Spencer
|
|
updated: October 4, 2005
|
|
|
|
updated by Deji Akingunola
|
|
October 15, 2008
|
|
|
|
updated by Deji Akingunola
|
|
June 15, 2011
|
|
|
|
Because ATLAS relies on compile-time optimizations to obtain improved
|
|
performance over BLAS and LAPACK, the resulting binaries are closely
|
|
tied to the hardware on which they are compiled, and can likely result
|
|
in very poor performance on other hardware. For this reason,
|
|
including a package like ATLAS in Fedora requires some compromises.
|
|
Firstly, a binary ATLAS package must perform reasonably well on the
|
|
entire range of hardware on which it could potentially be installed.
|
|
Optimizing ATLAS for the most modern hardware can result in
|
|
significant performance penalties for users using the same package on
|
|
older hardware. Second, building from the same source package must
|
|
result in identical binaries for any computer of a particular
|
|
architecture. This is because the binaries installed on a user's
|
|
computer are built on a computer in the Fedora Extras build system,
|
|
which will have hardware different from the end user's hardware, and
|
|
quite possibly different from other available hardware in the build
|
|
system.
|
|
|
|
As of version 3.8.2 (in Fedora), ATLAS builds uses achitectural defaults,
|
|
are partial results of past searches when the compiler and architecture
|
|
are known, to discover the appropriate kernels used to build all the required
|
|
libraries. They make install time quicker and also ensure that good results are
|
|
obtained, since they typically represent several searches and/or user
|
|
intervention into the usual search so that maximum performance is found.
|
|
The result is a set of libraries that will not
|
|
necessarily achieve optimal performance on any given hardware but
|
|
should still offer significant performance gains over the reference
|
|
BLAS and LAPACK libraries on most hardware. The binary package
|
|
includes the atlas libraries as well as binary-compatible blas and
|
|
lapack libraries that should work as a drop-in replacement for the
|
|
standard ones (they are installed in /usr/lib{64}/atlas* rather than
|
|
/usr/lib{64}).
|
|
|
|
For 32bit x86 systems, the default atlas package on was built using Pentium Pro
|
|
architectural defaults using just x87 math optimization. In addition to the
|
|
base 32bit build, 4 ATLAS subpackages are built for 3Dnow, SSE, SSE2, and SSE3
|
|
ix86 extensions, using architectural defaults obtained from Athlon K7, PIII,
|
|
Pentium 4 with SSE2 extension and PENTIUM 4 with SSE3 extensions respectively.
|
|
|
|
On 64bit x86 systems the default atlas package on was built with SSE2 optimization using architetural default made for AMD's HAMMER processor, and an additional
|
|
SSE3-enabled subpackage was built also using architetural default made for AMD's HAMMER processor.
|
|
|
|
This packaging allows multiple installation of different atlas sub-packages
|
|
at the same time. The alternatives system (read 'man alternatives' for usage)
|
|
is used in the -devel subpackages to select the appropriate location for the
|
|
architectural dependent header files.
|
|
|
|
This package is designed to build RPMs that are identical regardless
|
|
of where they are compiled and that provide reasonable performance on
|
|
a wide range of hardware. For users who want optimal performance on
|
|
particular hardware, custom RPMs can be built from the source package
|
|
by setting the RPM macro "enable_native_atlas" to a value of 1. This
|
|
can be done from the command line as in the following example:
|
|
|
|
rpmbuild -D "enable_native_atlas 1" --rebuild atlas-3.8.3-1.src.rpm
|
|
|
|
This will cause the ATLAS build to use the achitectural default most
|
|
appropriate for the system on which the package is to be built.
|