Quadruple-precision

[SkyScriptor]

Hi, I have installed heyoka.py library on my intel-mac and i have trouble enabling quadruple -precision. I do not understand how i am supposed to compile the heyoka C++ library with the HEYOKA_WITH_MPPP as described in the installation instructions. Any assistance will be helpful. Thanks in advance.

[bluescarni]

@SkyScriptor OSX compilers do not provide the __float128 floating-point type on top of which mp++'s and heyoka's quadruple precision support is implemented. If you want true quadruple-precision support, you will have to build heyoka/heyoka.py on a Linux machine.

As a (probably-better) alternative, you can enable multiprecision support and use that to work in quadruple precision (i.e., by selecting a mantissa of 113 bits). According to my testing, this is only slightly slower than "native" quadruple precision, and multiprecision is supported on all platforms.

In order to enable multiprecision computations, you first need to:

• compile and install the mp++ library with the option MPPP_WITH_MPFR set to TRUE (see here: https://github.com/bluescarni/mppp/blob/master/CMakeLists.txt#L36),
• compile and install the heyoka C++ library with the option HEYOKA_WITH_MPPP set to TRUE.

Then, you can install heyoka.py and you will have multiprecision support. A tutorial showing how to perform computations in multiple precision is available here:

https://bluescarni.github.io/heyoka.py/notebooks/arbitrary_precision.html

Finally, I would recommend to use the conda packages instead of doing all the compilation/installation by hand - the conda packages are already built with multiprecision support.