Difference between revisions of "ACML"
From CBLFS
Jump to navigationJump to search| (4 intermediate revisions by 2 users not shown) | |||
| Line 2: | Line 2: | ||
|- | |- | ||
!Download Source: | !Download Source: | ||
| − | | http://developer.amd.com/acml.jsp (to download one or more tarball acml-{{ACML-Version}}-gfortran-32bit, acml-{{ACML-Version}}-gfortran-64bit, acml-{{ACML-Version}}-gnu-32bit, or acml-{{ACML-Version}}-gnu-64bit) | + | |http://developer.amd.com/acml.jsp (to download one or more tarball acml-{{ACML-Version}}-gfortran-32bit, acml-{{ACML-Version}}-gfortran-64bit, acml-{{ACML-Version}}-gnu-32bit, or acml-{{ACML-Version}}-gnu-64bit) |
|} | |} | ||
{{Package-Introduction|AMD Core Mathematics Library (ACML) contains | {{Package-Introduction|AMD Core Mathematics Library (ACML) contains | ||
| − | * Full implementation of [[ | + | * Full implementation of [[ATLAS]], with key routines optimized for high performance on AMD processors. |
* Full suite of [[LAPACK]] routines optimized to achieve considerably higher performance than standard LAPACK implementations. | * Full suite of [[LAPACK]] routines optimized to achieve considerably higher performance than standard LAPACK implementations. | ||
* Comprehensive suite of Fast Fourier Transforms (FFTs) in both single-, double-, single-complex and double-complex data types. | * Comprehensive suite of Fast Fourier Transforms (FFTs) in both single-, double-, single-complex and double-complex data types. | ||
| Line 13: | Line 13: | ||
* Random Number Generators in both single- and double-precision. | * Random Number Generators in both single- and double-precision. | ||
| − | To obtain the ACML, you will be required to register with AMD and be at least 14-years old | + | To obtain the ACML, you will be required to register with AMD and be at least 14-years old. The packages compiled with gfortran provide libraries optimized for single- and multi-processor machines.|http://developer.amd.com/acml.jsp}} |
| − | Other vendors also provide | + | {{Note|The ACML license does not allow redistribution without written consent from AMD.}} |
| − | {| style="text-align: left; background-color: | + | |
| + | Other vendors also provide ATLAS and LAPACK libraries optimized for their processor. | ||
| + | {| style="text-align: left; background-color: #FCDFFF;" | ||
! Apple's | ! Apple's | ||
|[http://developer.apple.com/hardwaredrivers/ve/index.html Velocity Engine] | |[http://developer.apple.com/hardwaredrivers/ve/index.html Velocity Engine] | ||
| Line 36: | Line 38: | ||
|} | |} | ||
| − | = Non-Multilib or Multilib = | + | == Non-Multilib or Multilib == |
Create a directory to download and unpack the tarballs. | Create a directory to download and unpack the tarballs. | ||
| Line 70: | Line 72: | ||
|-valign="top" | |-valign="top" | ||
! Installed Libraries: | ! Installed Libraries: | ||
| − | | libacml.{so,a} | + | | libacml.{so,a}, libacml_mv.{so,a} |
| − | |||
|} | |} | ||
| Line 81: | Line 82: | ||
|binary utility to determine if your processor supports sse and sse2. Execute cat /proc/cpuinfo for the same results. | |binary utility to determine if your processor supports sse and sse2. Execute cat /proc/cpuinfo for the same results. | ||
|} | |} | ||
| + | |||
| + | [[Category:Science_Engineering]] | ||
Latest revision as of 16:38, 19 March 2009
| Download Source: | http://developer.amd.com/acml.jsp (to download one or more tarball acml-3.6.1-gfortran-32bit, acml-3.6.1-gfortran-64bit, acml-3.6.1-gnu-32bit, or acml-3.6.1-gnu-64bit) |
|---|
Contents
Introduction to ACML
AMD Core Mathematics Library (ACML) contains
- Full implementation of ATLAS, with key routines optimized for high performance on AMD processors.
- Full suite of LAPACK routines optimized to achieve considerably higher performance than standard LAPACK implementations.
- Comprehensive suite of Fast Fourier Transforms (FFTs) in both single-, double-, single-complex and double-complex data types.
- Fast scalar, vector, and array math transcendental library routines optimized for high performance on AMD processors.
- Random Number Generators in both single- and double-precision.
To obtain the ACML, you will be required to register with AMD and be at least 14-years old. The packages compiled with gfortran provide libraries optimized for single- and multi-processor machines.
Project Homepage: http://developer.amd.com/acml.jsp
Note
The ACML license does not allow redistribution without written consent from AMD.
Other vendors also provide ATLAS and LAPACK libraries optimized for their processor.
| Apple's | Velocity Engine |
|---|---|
| Hewlett-Packard's | Mathematical Software Library (MLIB) Free registration required. |
| IBM's | Engineering Scientific Subroutine Library (ESSL) |
| Intel's | Math Kernel Library (MKL) $399 US. |
| SGI's | Scientific Computing Software Library (SCSL) Free registration required. |
| Sun's | Sun Performance Library; Free to registered SDN members. |
Non-Multilib or Multilib
Create a directory to download and unpack the tarballs.
mkdir acml && cd acml
After retrieving and unpacking the tarballs, install each in turn:
./install-<package>.sh
where <package> is acml-3.6.1-gfortran-32bit, acml-3.6.1-gfortran-64bit, acml-3.6.1-gnu-32bit, or acml-3.6.1-gnu-64bit.
You will be required to accept the licensing agreement after which you will be prompted for a top level installation directory. The default is /opt/acml3.6.1. Each package will install its headers and libraries in separate subdirectories of the top level directory.
Move the documentation to a standard location:
install -v -m755 -d /usr/share/doc/acml-3.6.1 &&
install -v -m644 ACML-EULA.txt README.64-bit /usr/share/doc/acml-3.6.1 &&
mv -v /opt/acml3.6.1/{*.txt,ReleaseNote*} /usr/share/doc/acml-3.6.1 &&
mv -v /opt/acml3.6.1/Doc/*.{pdf,txt,inf*} /usr/share/doc/acml-3.6.1 &&
mv -v /opt/acml3.6.1/Doc/html/* /usr/share/doc/acml-3.6.1
Contents
| Installed Directories: | /opt/acml3.6.1/{gfortran32,gfortran32_mp,gfortran64,gfortran64_mp,gnu32,gnu64}/{examples,include,lib} |
|---|---|
| Installed Programs: | cpuid |
| Installed Libraries: | libacml.{so,a}, libacml_mv.{so,a} |
Short Descriptions
| cpuid: | binary utility to determine if your processor supports sse and sse2. Execute cat /proc/cpuinfo for the same results. |
|---|
