fix(core): reject use_libdevice on the NVRTC and PTX backends - #2573

Problem

ProgramOptions.use_libdevice is documented as "Only supported for the NVVM backend" (_program.pyx:465-468). The guard that was supposed to enforce that sits in Program_init's final else — the branch reached only when code_type is not a backend at all:

    else:
        supported_code_types = tuple(x.value for x in SourceCodeType)
        if options.use_libdevice:
            raise ValueError("use_libdevice is only supported by the NVVM backend")
        raise RuntimeError(f"Unsupported {code_type=} ({supported_code_types=})")

Two wrong behaviours follow:

1. The two real non-NVVM backends accept the option silently. Program(src, "c++", ProgramOptions(use_libdevice=True)) and the "ptx" equivalent construct fine. self._use_libdevice is initialised to False (_program.pyx:765) and only flipped inside the nvvm branch, so libdevice is never loaded — the caller finds out via undefined-symbol errors at link time instead of via the documented up-front ValueError.

2. A typo'd code_type reports the wrong error. Program(src, "bogus", ProgramOptions(use_libdevice=True)) raises "use_libdevice is only supported by the NVVM backend" and never mentions that code_type is the actual problem.

The sibling option shows the intended shape: extra_sources has the same "NVVM only" contract, and its guard is duplicated into each real branch — _program.pyx:772-773 (c++) and :790-791 (ptx).

Fix

Move the use_libdevice check next to those extra_sources guards in the c++ and ptx branches, and drop it from the else so an unrecognised code_type reports itself. No behaviour changes for code_type="nvvm".

This is an error-path-only change: any program that compiled before still compiles, and any program that is newly rejected was already producing a build the user did not ask for.

Tests

Three cases in cuda_core/tests/test_program.py, placed next to the existing extra_sources tests they mirror:

• test_cpp_program_with_use_libdevice — NVRTC rejects it.
• test_ptx_program_use_libdevice_unsupported — the PTX/linker path rejects it.
• test_program_init_invalid_code_type_reports_the_code_type — use_libdevice=True no longer shadows the unrecognised-code_type error.

What I ran

Environment: macOS, no CUDA driver and no CUDA toolkit, so cuda.core cannot be built or imported here.

• Did not run: the three new tests, or anything else in cuda_core/tests/ — they need a built cuda.core and a GPU. They are written against the existing extra_sources tests immediately adjacent to them, which use the same fixtures.
• Ran: a reduction of Program_init's dispatch with the CUDA calls removed and the guard placement kept verbatim, before and after the change:

--- before ---
  code_type='c++'      use_libdevice=True -> accepted, _use_libdevice=False
  code_type='ptx'      use_libdevice=True -> accepted, _use_libdevice=False
  code_type='nvvm'     use_libdevice=True -> accepted, _use_libdevice=True
  code_type='fortran'  use_libdevice=True -> ValueError: use_libdevice is only supported by the NVVM backend
--- after ---
  code_type='c++'      use_libdevice=True -> ValueError: use_libdevice is not supported by the NVRTC backend (C++ code_type)
  code_type='ptx'      use_libdevice=True -> ValueError: use_libdevice is not supported by the PTX backend.
  code_type='nvvm'     use_libdevice=True -> accepted, _use_libdevice=True
  code_type='fortran'  use_libdevice=True -> RuntimeError: Unsupported code_type='fortran' (SUPPORTED=('c++', 'ptx', 'nvvm'))

• Ran: ruff check / ruff format --check on cuda_core/tests/test_program.py — clean, no new findings against a main baseline for that file.
• Checked: use_libdevice appears in _program.pyx only at the two sites above (the nvvm set and this guard), so nothing else depended on the old placement. test_program_init_invalid_code_type still passes unchanged because its ProgramOptions leaves use_libdevice at its False default.