Using fsycl-link

James_S_9 — Tue, 13 Apr 2021 08:46:53 GMT

Using Intel 2021.2 from oneapi-hpckit docker image.

I'm attempting to follow the documentation here: https://software.intel.com/content/www/us/en/develop/documentation/oneapi-dpcpp-cpp-compiler-dev-guide-and-reference/top/compilation/use-a-third-party-compiler-as-a-host-compiler-for-dpc-code.html

but am running into a few issues:

1. The generated header seems to be specialising a type that doesn't exist:

In file included from <command-line>:0:0: ./a_host.h:27:20: error: 'KernelInfoData' is not a class template template <> struct KernelInfoData<'_', 'Z', 'T', 'S', 'Z', 'Z', '4', 'm', 'a', 'i', 'n', 'E', 'N', 'K', 'U', 'l', 'R', 'N', '2', 'c', 'l', '4', 's', 'y', 'c', 'l', '7', 'h', 'a', 'n', 'd', 'l', 'e', 'r', 'E', 'E', '1', '1', '_', '2', '0', 'c', 'l', 'E', 'S', '2', '_', 'E', 'U', 'l', 'N', 'S', '0', '_', '4', 'i', 't', 'e', 'm', 'I', 'L', 'i', '1', 'E', 'L', 'b', '1', 'E', 'E', 'E', 'E', '1', '3', '_', '2', '7'> { ^~~~~~~~~~~~~~ In file included from <command-line>:0:0: ./a_host.h:27:415: error: explicit specialization of non-template 'cl::sycl::detail::KernelInfoData' ', 'l', 'E', 'S', '2', '_', 'E', 'U', 'l', 'N', 'S', '0', '_', '4', 'i', 't', 'e', 'm', 'I', 'L', 'i', '1', 'E', 'L', 'b', '1', 'E', 'E', 'E', 'E', '1', '3', '_', '2', '7'> { ^

2. CL_TARGET_OPENCL_VERSION is not defined:

In file included from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/cl.h:20:0, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/cl_ext_intel.h:44, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl/detail/common.hpp:11, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl/access/access.hpp:10, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl/ONEAPI/atomic_enums.hpp:12, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl/ONEAPI/atomic_accessor.hpp:11, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl/ONEAPI/atomic.hpp:11, from /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/sycl.hpp:11, from sample.cc:1: /opt/intel/oneapi/compiler/latest/linux/include/sycl/CL/cl_version.h:22:104: note: #pragma message: cl_version.h: CL_TARGET_OPENCL_VERSION is not defined. Defaulting to 300 (OpenCL 3.0) #pragma message("cl_version.h: CL_TARGET_OPENCL_VERSION is not defined. Defaulting to 300 (OpenCL 3.0)") ^

3. The linked device.o file only contains sycl symbols and none of the compiled kernels:

$ nm -A device.o device.o:0000000000000000 r .sycl_offloading.0.data device.o:0000000000000000 r .sycl_offloading.0.data.size device.o:0000000000000060 d .sycl_offloading.descriptor device.o:0000000000000000 d .sycl_offloading.device_images device.o:0000000000000007 r .sycl_offloading.opts.compile.0 device.o:0000000000000008 r .sycl_offloading.opts.link.0 device.o:0000000000000000 r .sycl_offloading.target.0 device.o: U __sycl_register_lib device.o: U __sycl_unregister_lib device.o:0000000000000000 t sycl.descriptor_reg device.o:0000000000000010 t sycl.descriptor_unreg

Is there any updated documentation on how the -fsycl-link flag should be used as it currently seems broken?

Re: Using fsycl-link

James_S_9 — Tue, 13 Apr 2021 10:22:29 GMT

The first error seems to be due to __SYCL_UNNAMED_LAMBDA__ not being defined when compiling with g++ as KernelInfoData only exists when that is turned on.

Re:Using fsycl-link

VidyalathaB_Intel — Wed, 14 Apr 2021 10:25:56 GMT

Hi James,

Thanks for reaching out to us.

Could you please share a small reproducer and steps to reproduce the issues (compilation and linking commands used).

Regards,

Vidya

Re: Re:Using fsycl-link

James_S_9 — Wed, 14 Apr 2021 10:32:39 GMT

The commands are as in the documentation as stated in my original post: https://software.intel.com/content/www/us/en/develop/documentation/oneapi-dpcpp-cpp-compiler-dev-gui...

I was just using a single file though:

#include <CL/sycl.hpp> #include <iostream> constexpr int num = 16; using namespace sycl; int main() { auto r = range{num}; buffer<int> a{r}; queue{}.submit([&](handler& h) { accessor out{a, h}; h.parallel_for(r, [=](item<1> idx) { out[idx] = idx; }); }); host_accessor result{a}; for(int i = 0; i < num; ++i) std::cout << result[i] << "\n"; }

The integration header generated by this file gives the errors described in my original post.

topic Re:Using fsycl-link in Intel® oneAPI DPC++/C++ Compiler

Using fsycl-link

Re: Using fsycl-link

Re:Using fsycl-link

Re: Re:Using fsycl-link