Replace nvrtc Tempita templates with pre-generated Cython files (#1900)

* [DRAFT] Replace nvrtc Tempita templates with pre-generated Cython files

Replace .pyx.in / .pxd.in Tempita templates with plain .pyx / .pxd files
generated by the refactored cython-gen (cybind leof/cybind\!369).

Changes:
- Remove 6 Tempita template files (.pyx.in / .pxd.in) for nvrtc
- Add 7 pre-generated Cython files with zero Tempita
- Move internal layer from _bindings/ to _internal/ following cybind convention
- Split platform-specific code into nvrtc_linux.pyx / nvrtc_windows.pyx
- Internal layer now uses cybind-style loading (RTLD_DEFAULT + fallback,
  FunctionNotFoundError, _init_nvrtc/_check_or_init_nvrtc naming)

This is a companion PR to cybind leof/cybind\!369. Build system changes
(build_hooks.py, pyproject.toml) to remove pyclibrary are not yet
included — this PR shows the generated output for review.

Part of https://gitlab-master.nvidia.com/leof/cybind/-/issues/173

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Regenerate nvrtc files with template-based generation

Update generated artifacts to use cybind-style templates:
- Fix excessive blank lines from Tempita stripping
- Internal layer files now generated from template files under
  cybind/assets/templates/ via string.Template.substitute()
- Proper license headers from templates

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Update cynvrtc.pyx to use _internal cimport (cybind convention)

- cynvrtc.pyx now uses "from ._internal cimport nvrtc as _nvrtc"
  instead of the legacy "cimport _bindings.cynvrtc as cynvrtc"
- Function wrappers reference _nvrtc._funcName() matching cybind style
- cynvrtc.pxd gets proper template header

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Update build system: skip pyclibrary for nvrtc, fix cynvrtc source path

- Remove nvrtc from _REQUIRED_HEADERS (no pyclibrary parsing needed;
  nvrtc .pyx/.pxd files are now pre-generated by cybind)
- Update sources_list: cynvrtc.pyx moved from _bindings/ to bindings/
- _rename_architecture_specific_files already handles
  _internal/nvrtc_linux.pyx -> _internal/nvrtc.pyx

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Restore generator version in headers, remove duplicate cynvrtc source entry

- Regenerate files with generator version in header comments
- Remove explicit cynvrtc.pyx from sources_list (already covered by
  cuda_bindings_files glob)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Fix pre-commit: strip trailing whitespace and ensure final newline

Regenerated from latest cybind which now cleans up trailing whitespace
and ensures files end with exactly one newline.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: leofang

cuda_bindings/build_hooks.py

@@ -94,9 +94,7 @@ def _get_cuda_path() -> str:
         "driver_functions.h",
         "cuda_profiler_api.h",
     ],
-    "nvrtc": [
-        "nvrtc.h",
-    ],
+    # nvrtc: headers no longer parsed at build time (pre-generated by cybind).
     # During compilation, Cython will reference C headers that are not
     # explicitly parsed above. These are the known dependencies:
     #
@@ -423,7 +421,8 @@ def _cleanup_dst_files():
     sources_list = [
         # private
         (["cuda/bindings/_bindings/cydriver.pyx", "cuda/bindings/_bindings/loader.cpp"], None),
-        (["cuda/bindings/_bindings/cynvrtc.pyx"], None),
+        # cynvrtc.pyx is now in cuda/bindings/ (pre-generated by cybind),
+        # picked up by the cuda_bindings_files glob below.
         (["cuda/bindings/_bindings/cyruntime.pyx"], static_runtime_libraries),
         (["cuda/bindings/_bindings/cyruntime_ptds.pyx"], static_runtime_libraries),
         # utils

cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in

@@ -1,136 +1,63 @@
 # SPDX-FileCopyrightText: Copyright (c) 2021-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
+#
+# This code was automatically generated with version 13.2.0, generator version 0.3.1.dev1364+ged01d643e. Do not modify it directly.
 
-# This code was automatically generated with version 13.2.0, generator version 0.3.1.dev1422+gf4812259e.d20260318. Do not modify it directly.
-from cuda.bindings.cynvrtc cimport *
+from ..cynvrtc cimport *
 
-{{if 'nvrtcGetErrorString' in found_functions}}
+###############################################################################
+# Wrapper functions
+###############################################################################
 
 cdef const char* _nvrtcGetErrorString(nvrtcResult result) except ?NULL nogil
-{{endif}}
-
-{{if 'nvrtcVersion' in found_functions}}
 
 cdef nvrtcResult _nvrtcVersion(int* major, int* minor) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetNumSupportedArchs' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetNumSupportedArchs(int* numArchs) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetSupportedArchs' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetSupportedArchs(int* supportedArchs) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcCreateProgram' in found_functions}}
 
 cdef nvrtcResult _nvrtcCreateProgram(nvrtcProgram* prog, const char* src, const char* name, int numHeaders, const char** headers, const char** includeNames) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcDestroyProgram' in found_functions}}
 
 cdef nvrtcResult _nvrtcDestroyProgram(nvrtcProgram* prog) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcCompileProgram' in found_functions}}
 
 cdef nvrtcResult _nvrtcCompileProgram(nvrtcProgram prog, int numOptions, const char** options) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetPTXSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetPTXSize(nvrtcProgram prog, size_t* ptxSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetPTX' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetPTX(nvrtcProgram prog, char* ptx) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetCUBINSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetCUBINSize(nvrtcProgram prog, size_t* cubinSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetCUBIN' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetCUBIN(nvrtcProgram prog, char* cubin) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetLTOIRSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetLTOIRSize(nvrtcProgram prog, size_t* LTOIRSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetLTOIR' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetLTOIR(nvrtcProgram prog, char* LTOIR) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetOptiXIRSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetOptiXIRSize(nvrtcProgram prog, size_t* optixirSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetOptiXIR' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetOptiXIR(nvrtcProgram prog, char* optixir) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetProgramLogSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetProgramLogSize(nvrtcProgram prog, size_t* logSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetProgramLog' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetProgramLog(nvrtcProgram prog, char* log) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcAddNameExpression' in found_functions}}
 
 cdef nvrtcResult _nvrtcAddNameExpression(nvrtcProgram prog, const char* name_expression) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetLoweredName' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetLoweredName(nvrtcProgram prog, const char* name_expression, const char** lowered_name) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetPCHHeapSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetPCHHeapSize(size_t* ret) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcSetPCHHeapSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcSetPCHHeapSize(size_t size) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetPCHCreateStatus' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetPCHCreateStatus(nvrtcProgram prog) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetPCHHeapSizeRequired' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetPCHHeapSizeRequired(nvrtcProgram prog, size_t* size) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcSetFlowCallback' in found_functions}}
 
 cdef nvrtcResult _nvrtcSetFlowCallback(nvrtcProgram prog, void* callback, void* payload) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetTileIRSize' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetTileIRSize(nvrtcProgram prog, size_t* TileIRSizeRet) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-
-{{if 'nvrtcGetTileIR' in found_functions}}
 
 cdef nvrtcResult _nvrtcGetTileIR(nvrtcProgram prog, char* TileIR) except ?NVRTC_ERROR_INVALID_INPUT nogil
-{{endif}}
-