package interval_intel
Install
Dune Dependency
Authors
Maintainers
Sources
sha256=a2807d7860a9f965a3f91b40de17b98321e5c3f7da9e69dccf3dc20f1ac77e96
sha512=76196b171a41ce1374cae946b7c7a1dea6d614db61743ead14d628a12dce30a0dbd82db1960ad17b58f1f65a6f3fc61aaa5eac42face474baf840efb670feee8
Description
This library uses assembly code to compute all operations with proper roundings, and currently ONLY works on intel processors. It supports Linux, Windows and MacOS, with gcc and clang. Unfortunately, the Intel processor does not properly round trigonometric functions so this library cannot be used whenever proved correct enclosures are needed.
More information is given in the paper presented in the OCaml meeting 2012: http://www.alliot.fr/papers/oud2012.pdf
README
README.md
Interval
This is an interval arithmetic library for OCaml. Here is a small example in the REPL:
# #require "interval.top";;
...
# open Interval_crlibm;;
# let v = I.singleton 1.;;
val v : Interval_crlibm.I.t = [1, 1]
# I.sin v;;
- : Interval_crlibm.I.t = [0.841471, 0.841471]
# I.Precision.set (Some 16);;
- : unit = ()
# I.sin v;;
- : Interval_crlibm.I.t = [0.8414709848078965, 0.8414709848078966]
Several OPAM packages are provided by this repository:
interval_base
: basic interval library that defines the datatypeInterval.t
and uses Intel assembly if possible or C99 instructions to perform arithmetic operations (module Interval).interval_crlibm
: relies on crlibm to implement the interval operations. CRlibm provides proved correctly-rounded operations, so this is the library of choice for computer assisted proofs.interval_intel
: use Intel FPU instructions with directed rounding to implement interval operations. However, the Intel FPU operations may not always be correctly rounded for, say, trigonometric functions.interval_intel
uses assembly code to compute all operations with proper rounding, and currently ONLY works on Intel processors. The package has been developed for Linux systems but works on MacOSX and Windows when compiled with GCC.interval
is a meta-package installing all the packages above. It also providesinterval.top
used above.
Note that ocamlopt
does float constant propagation in round to the nearest mode which may invalidate interval computations. Use the compiler flag -no-float-const-prop
to deactivate it.
Happy interval programming...
Installation
The easier way to install this library is to use OPAM:
opam install interval
interval
is a meta-package that will install all packages mentioned above.
If you cloned this repository, first install dune and type make
in the main directory. This will compile the libraries, the examples and run basic tests. You can compile the examples with The programs of the examples will be in _build/default/examples/
.
Documentation
To documentation is build using dune build @doc
and will be in _build/default/_doc/
in HTML format. You can also consult the interfaces of
or online. It is extremely wise to read the whole documentation, even if you intend to only use the Interval_intel
module.
Some examples are available in the examples/
directory. There is a B_AND_B
sub-directory with an example of a branch-and-bound algorithm that uses interval arithmetics for function optimization (the example is for the Griewank function, but you can substitute any function you like).
Tests
Tests are available in the tests/
directory. They are mainly for debugging purpose and quite complicated. You may run them (make tests
) to check that everything is working properly for your machine. The test
program runs also a speed test for your particular architecture.
Bug reports should be open at https://github.com/Chris00/ocaml-interval/issues
Remark: This library was originally published on Jean-Marc Alliot website but was moved to Github with the permission of the authors.
Dependencies (2)
- dune
-
interval_base
= version
Dev Dependencies
None
Used by (1)
-
interval
>= "1.6"
Conflicts
None