topic Re: Support for bessel functions inside a SYCL kernel in Intel® oneAPI DPC++/C++ Compiler

Support for bessel functions inside a SYCL kernel

RCB1 — Tue, 24 Sep 2024 15:09:58 GMT

Hello

I am trying to test some SYCL kernels in oneAPI DPC++ that have Bessel function calls inside the parallel_for loop of the kernel.

If I use either Boost or the Standard library, or the Intel mathimf.h library as shown below, I get compile time errors.

float f1 = boost::math::cyl_bessel_j(0.0f, 1.0f);
float f2 = std::cyl_bessel_j(0.0f, 1.0f);

float f3 = j0f(1.0f);

Is there support for Bessel functions in SYCL kernels.

The source file is a modification of vector-add-buffers.cpp in the samples, and the compiler output is attached in errors.txt.

Thanks

RCB

Re: Support for bessel functions inside a SYCL kernel

Alex_Y_Intel — Wed, 25 Sep 2024 00:09:23 GMT

For your vector-add-buffers.cpp code, only "float f3 = j0f(1.0f)" can work, but the proper usage is "float f3 = ext::intel::math::j0(1.0f);" and you need to include "#include <sycl/ext/intel/math.hpp>"

Re: Support for bessel functions inside a SYCL kernel

RCB1 — Wed, 25 Sep 2024 06:34:31 GMT

Thank you Alex much appreciated - the sycl::ext::intel::math::j0(1.0f) works for me. Incidentally is it out of the question that the Boost or the STL Bessel functions may work at some point as there is some additional functionality that I am interested in those libraries?

Re: Support for bessel functions inside a SYCL kernel

Alex_Y_Intel — Wed, 25 Sep 2024 23:50:25 GMT

STL Bessel functions only work for host code but not SYCL kernels. We might file Boost and STL functions for feature requests if we get enough inquiries.