[PERF]: Faster void * conversion

cuda_bindings/cuda/bindings/_lib/utils.pxd.in

@@ -1,10 +1,11 @@
-# SPDX-FileCopyrightText: Copyright (c) 2021-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2021-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
 cimport cuda.bindings.driver as driver
 cimport cuda.bindings.cydriver as cydriver
 cimport cuda.bindings.cyruntime as cyruntime
 from libcpp.vector cimport vector
+from cpython.buffer cimport PyBuffer_Release, Py_buffer
 
 cdef class _HelperKernelParams:
     cdef Py_buffer _pybuffer
@@ -14,10 +15,19 @@ cdef class _HelperKernelParams:
     cdef int _length
     cdef bint _malloc_list_created
 
+cdef struct _HelperInputVoidPtrStruct:
+    Py_buffer _pybuffer
+
 cdef class _HelperInputVoidPtr:
-    cdef Py_buffer _pybuffer
+    cdef _HelperInputVoidPtrStruct _helper 
     cdef void* _cptr
-    cdef bint _pyobj_acquired
+
+cdef void * _helper_input_void_ptr(ptr, _HelperInputVoidPtrStruct *buffer)
+
+cdef inline void * _helper_input_void_ptr_free(_HelperInputVoidPtrStruct *helper):
+    if helper[0]._pybuffer.buf != NULL:
+        PyBuffer_Release(&helper[0]._pybuffer)
+
 {{if 'CUmemPool_attribute_enum' in found_types}}
 
 cdef class _HelperCUmemPool_attribute:

cuda_bindings/cuda/bindings/_lib/utils.pxi.in

@@ -128,31 +128,32 @@ cdef class _HelperKernelParams:
 
 cdef class _HelperInputVoidPtr:
     def __cinit__(self, ptr):
-        self._pyobj_acquired = False
+        self._cptr = _helper_input_void_ptr(ptr, &self._helper)
+
+    def __dealloc__(self):
+        _helper_input_void_ptr_free(&self._helper)
+
+    @property
+    def cptr(self):
+        return <void_ptr>self._cptr 
+
+
+cdef void * _helper_input_void_ptr(ptr, _HelperInputVoidPtrStruct *helper):
+    helper[0]._pybuffer.buf = NULL
+    try:
+        return <void *><void_ptr>ptr        
+    except:
         if ptr is None:
-            self._cptr = NULL
-        elif isinstance(ptr, (int)):
-            # Easy run, user gave us an already configured void** address
-            self._cptr = <void*><void_ptr>ptr
-        elif isinstance(ptr, (_driver["CUdeviceptr"])):
-            self._cptr = <void*><void_ptr>int(ptr)
+            return NULL
         elif PyObject_CheckBuffer(ptr):
             # Easy run, get address from Python Buffer Protocol
-            err_buffer = PyObject_GetBuffer(ptr, &self._pybuffer, PyBUF_SIMPLE | PyBUF_ANY_CONTIGUOUS)
+            err_buffer = PyObject_GetBuffer(ptr, &helper[0]._pybuffer, PyBUF_SIMPLE | PyBUF_ANY_CONTIGUOUS)
             if err_buffer == -1:
                 raise RuntimeError("Failed to retrieve buffer through Buffer Protocol")
-            self._pyobj_acquired = True
-            self._cptr = <void*><void_ptr>self._pybuffer.buf
+            return <void*><void_ptr>(helper[0]._pybuffer.buf)
         else:
             raise TypeError("Provided argument is of type {} but expected Type {}, {} or object with Buffer Protocol".format(type(ptr), type(None), type(int)))
 
-    def __dealloc__(self):
-        if self._pyobj_acquired is True:
-            PyBuffer_Release(&self._pybuffer)
-
-    @property
-    def cptr(self):
-        return <void_ptr>self._cptr
 
 {{if 'CUmemPool_attribute_enum' in found_types}}