I had been struggling with the dpcpp compiler trying to run the examples from the book Data Parallel C++ and have finally resolved it.
To run the examples on an intel 630 HD graphics gpu, at least, I needed to do three things.
1. use the -Qstd=c++17 switch for the compiler.
2. Add the user variable SYCL_BE_=PI_OPENCL
3. Add the user variable SYCL_PROGRAM_COMPILE_OPTIONS= I just made this a space and it
worked.
There seems to be a lot of flux between the SYCL standard, header file rearrangements and the like so that the github samples do not work on the c++20 version of the compiler.
I stumbled upon the environment variables reading the Explicit SIMD extension (though it turns out that you don't have to do special coding). In fact the sample they provided utterly failed.
So if anyone has the Intel UHD 600 graphics or above GPU, then doing what I suggest above will alleviate a lot of problems until Intel and the contributors to the standard on GitHub settle down and mesh with each other. It literally took me days to come across what I needed to do. Thank goodness I am retired.
Enjoy and spread the word.
Thanks
