Intel C++

From ArchWiki

Tango-view-refresh-red.pngThis article or section is out of date.Tango-view-refresh-red.png

Reason: Intel has discontinued Parallel Studio XE and replaced it with the oneAPI Base & HPC Toolkit. The instructions here no longer apply. (Discuss in Talk:Intel C++)

Installation and basic usage of Intel Performance Libraries and other tools for Linux on Arch. They are included in the Intel Parallel Studio XE (formerly Intel C++ Composer XE or Intel C++ Compiler Professional Edition).

Some libraries and tools are free and open source, other require a license file which can be obtained upon registration.

Free libraries

Intel Parallel Studio XE

Intel Parallel Studio XE packages are available in the AUR. The PKGBUILD uses a free download link, but a license file needs to be obtained in order to actually use (most of) the packages. Intel offers a free license for open-source contributors, students, researchers, and educators. The requisite license file is emailed to users upon registration and should be copied into the $startdir prior to building the packages.

The intel-parallel-studio-xe PKGBUILD assembles 10 split packages:

Using icc with makepkg

Tango-view-refresh-red.pngThis article or section is out of date.Tango-view-refresh-red.png

Reason: Information is from the time before intel-common-libsAUR was created. intel-openmpAUR does not depend on intel-compiler-baseAUR, so the compiled packages probably do not need to depend on the compiler package. (Discuss in Talk:Intel C++)
Note:
  • Packages that have been compiled by icc will depend on the associated libs contained in the intel-openmp package in order to run. Since intel-openmp depends on intel-compiler-base, users must have both packages installed at all times!
  • Not every package will successfully compile with icc without heavy modifications to the underlying source.

There is currently no official guide to using icc with makepkg. This section is meant to capture various methods suggested by users. Please make a new sub-section with your suggested method titled as such.

Method 1 (12/08/2012)

Modify /etc/makepkg.conf inserting the following code under the existing line defining CXXFLAGS to enable makepkg to use icc. No special switches are needed when calling makepkg to build.

_CC=icc
if [ $_CC = "icc" ]; then
export CC="icc"
export CXX="icpc"
export CFLAGS="-march=native -O3 -no-prec-div -fno-alias -pipe"
export CXXFLAGS="${CFLAGS}"
export LDFLAGS="-Wl,-O1,--sort-common,--as-needed"
export AR="xiar"
export LD="xild"
fi
Note:
  • To toggle between the native gcc and icc, simple comment or uncomment the newly created _CC variable.
  • In some case the compilation method described above fails and the compilation will be performed with gcc, so you should test if yours application has been effectively compiled with icc.

Method 2

Insert the following code anywhere near the top of the PKGBUILD:

groups=('modified')
export CC="icc"
export CXX="icpc"
export CFLAGS="-march=native -O3 -no-prec-div -fno-alias -pipe"
export CXXFLAGS="-march=native -O3 -no-prec-div -fno-alias -pipe"
export LDFLAGS="-Wl,-O1,--sort-common,--as-needed"
export AR="xiar"
export LD="xild"
Note: The use of groups=('modified') is explained further in Arch Build System#Preserve modified packages. You can delete this line if necessary. However, remember that the package will be replaced with the GCC version every time it is upgraded.

Add intel-openmp and intel-compiler-base to the depends array.

Note: To test if your package has been really compiled with icc, observe the build output and watch if it is using the icc or icpc command.

icc CFLAGS

In general, icc supports many of the same CFLAGS gcc supports and is also pretty tolerant to gcc flags it cannot use. In most cases it will happily ignore the flag warning the user and moving on. For an exhaustive list and explanation of available compiler flags, consult the icc manpage or better yet by invoking the compiler with the help flag:

icc --help

-xX

Use to generate specialized code to run exclusively on processors supporting it. If unsure which option to use, simply inspect the flags section of /proc/cpuinfo. In the example below, SSE4.1 would be the correct selection:

$ grep -m 1 flags /proc/cpuinfo
flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush 
dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc arch_perfmon pebs 
bts rep_good nopl aperfmperf pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr
pdcm sse4_1 lahf_lm dts tpr_shadow vnmi flexpriority
  • -xHost
  • -xSSE2
  • -xSSE3
  • -xSSSE3
  • -xSSE4.1
  • -xSSE4.2
  • -xAVX
  • -xCORE-AVX-I
  • -xSSSE3_ATOM
Tip: Use the -xHost flag if unsure what your specific processor supports.

-Ox

Same behavior as gcc. x is one of the following options:

  • 0 - disables optimizations
  • 1 - optimize for maximum speed, but disable some optimizations which increase code size for a small speed benefit
  • 2 - optimize for maximum speed (DEFAULT)
  • 3 - optimize for maximum speed and enable more aggressive optimizations that may not improve performance on some programs (recommended for math intensive looping programs)

-w

Similar to the gcc:

  • -w - disable all warnings (recommended for the package compilation)
  • -Wbrief - print brief one-line diagnostics
  • -Wall - enable all warnings
  • -Werror - force warnings to be reported as errors

Software compiled with Intel C / C++

In the following table we report a list of packages from the official repository that we have tried to compile with the intel C/C++ compiler. The compilation should be done by using the PKGBUILD from ABS.

Application Method 1 Comments
xvidcore OK
kdebase OK
conky 1.9.0 OK
nginx 1.4.2 OK
gzip 1.6 OK
xz OK
lz4 OK Must edit the PKGBUILD.
minetest OK
opus OK
zlib 1.2.8 Not recommended Causes bugs in some applications, like tightvnc
Gimp 2.8 / 2.9 OK
Pacman 4.0.3 OK
x264 OK
MySql OK
SqlLite OK
lame OK
xaos OK
gegl OK
VLC Unsuccessful There is some problem with the compiler flags
bzip2 Unsuccessful There is some problem with the compiler flags
mplayer Out of date Does not recognize the Intel compiler
optipng OK Comment out LD=xild in makepkg.conf
python-numpy OK Must edit the PKGBUILD. python-numpy-mklAUR
python-scipy OK Must edit the PKGBUILD. python-scipy-mklAUR
Qt OK We must add the option -platform linux-icc-64 (or 32) in the configure command
systemd Fail undefined reference to `server_dispatch_message'

Legend:

OK The compilation with ICC works!
OK The compilation works but is needed an editing of the PKGBUILD
Unsuccessful The compilation may work, but there are some compilations errors.
Not recommended The compilation works, but is not recommended
Fail It is impossible to compile the PKG with ICC.
Out of date It is unsuccessful or fails with older CFLAGS.