19 releases
0.8.1 | Sep 13, 2022 |
---|---|
0.8.0-rc.1 | Aug 25, 2022 |
0.7.0+mkl2020.1 | Jul 29, 2022 |
0.6.0+mkl2020.1 | Jun 23, 2020 |
0.2.5 | Sep 14, 2017 |
#137 in Math
9,948 downloads per month
Used in 104 crates
(38 directly)
15KB
100 lines
intel-mkl-src
crate | crate.io | docs.rs | master | description |
---|---|---|---|---|
intel-mkl-src | Source crate for Intel-MKL | |||
intel-mkl-sys | FFI for Intel-MKL vector math, and statistical functions | |||
intel-mkl-tool | Seek Intel-MKL libraries from filesystem |
Usage
intel-mkl-src
crate is a *-src
crate. This links MKL libraries to executable build by cargo, but does not provide Rust bindings.
Please use blas-sys
, lapack-sys
, or fftw-sys
to use BLAS, LAPACK, FFTW interface of MKL, e.g.
[dependencies]
fftw-sys = { version = "0.4", features = ["intel-mkl"] }
Binding to MKL specific features are provided by intel-mkl-sys
crate. This contains
How to find system MKL libraries
intel-mkl-tool
crate seeks system MKL library installed by various installer as following manner:
- Seek using
pkg-config
command - Seek
${MKLROOT}
directory - Seek default installation path
/opt/intel/mkl
for LinuxC:/Program Files (x86)/IntelSWTools/
andC:/Program Files (x86)/Intel/oneAPI
for Windows
If intel-mkl-tool
does not find MKL library, intel-mkl-src
try to download MKL binaries from GitHub Container Registry (ghcr.io).
Supported features
There are 8 (=2x2x2) mkl-*-*-*
features to specify how to link MKL libraries.
If any feature is set, default to mkl-static-ilp64-iomp
.
Link type (static
or dynamic
)
dynamic
means MKL is linked dynamically, i.e. the executable does not contains MKL libraries
and will seek them from filesystem while execution.
This is better choice when the MKL libraries are managed by the system package manager e.g. apt
.
static
means MKL is linked statically, i.e. the MKL binaries are embedded in the executable file.
This is better choice when creating portable executable.
Data model (lp64
or ilp64
)
This specify the data model:
ilp64
meansint
(i),long
(l), and pointers (p) are 64-bit.lp64
meanslong
(l) and pointers (p) are 64-bit,int
is 32-bit.
Thread management (iomp
or seq
)
iomp
means MKL uses Intel OpenMP runtimeseq
means sequential (single thread) execution
Using GNU OpenMP runtime (libgomp
) is not supported yet. Please see https://github.com/rust-math/intel-mkl-src/issues/97
License
MKL is distributed under the Intel Simplified Software License for Intel(R) Math Kernel Library, See License.txt. Some wrapper codes are licensed by MIT License (see the header of each file).
Dependencies
~0–2.1MB
~32K SLoC