pystencils.jit.sycl.SYCLClangInfo

class pystencils.jit.sycl.SYCLClangInfo(openmp=True, optlevel='fast', cxx_standard='c++17', target=<Target._SYCL: 16384>, extra_cxxflags=<factory>, sycl_targets=<factory>, amd_offload_architecutres=<factory>)

Compiler info for the SYCL/oneAPI Clang compiler.

Parameters:

• openmp (bool)

• optlevel (str | None)

• cxx_standard (str)

• target (Target)

• extra_cxxflags (list[str])

• sycl_targets (list[str])

• amd_offload_architecutres (list[str])

target: Target = 16384

Hardware target to compile for.

The value of target is used to set the -march compiler option (or equivalent). Target.CurrentCPU translates to -march=native.

sycl_targets: list[str]

Arguments for -fsycl-targets can be nvptx64-nvidia-cuda, amdgcn-amd-amdhsa, and spir64

amd_offload_architecutres: list[str]

For targeting the AMD/HIP backend, the SYCL compiler needs the exact architecture specifier https://developer.codeplay.com/products/oneapi/amd/2025.2.0/guides/get-started-guide-amd.html#use-dpc-to-target-amd-gpus

cxxflags()

Compiler flags affecting C++ compilation