Support Cython 3

Author: vzhurba01

cuda/cuda.pyx.in

@@ -2164,7 +2164,7 @@ cdef class cudaGraphNodeParams:
         Get memory address of class instance
 
     """
-    cdef ccudart.cudaGraphNodeParams __val
+    cdef ccudart.cudaGraphNodeParams* _val_ptr
     cdef ccudart.cudaGraphNodeParams* _ptr
     cdef cudaKernelNodeParamsV2 _kernel
     cdef cudaMemcpyNodeParams _memcpy

cuda/cudart.pxd.in

@@ -5,7 +5,7 @@
 # this software. Any use, reproduction, disclosure, or distribution of
 # this software and related documentation outside the terms of the EULA
 # is strictly prohibited.
-from typing import List, Tuple, Any
+from typing import List, Tuple, Any, Optional
 from enum import IntEnum
 import cython
 import ctypes

cuda/cudart.pyx.in

@@ -610,10 +610,12 @@ def test_device_get_name():
     expect = p.stdout.split(delimiter)
     size = 64
     _, got = cuda.cuDeviceGetName(size, device)
-    # Returned value is bytes, and we expect it to be of requested size
-    # assert len(got) == size
     got = got.split(b'\x00')[0]
-    assert got in expect
+    if any(b'Unable to determine the device handle for' in result for result in expect):
+        # Undeterministic devices get waived
+        pass
+    else:
+        assert got in expect
 
     err, = cuda.cuCtxDestroy(ctx)
     assert(err == cuda.CUresult.CUDA_SUCCESS)

cuda/nvrtc.pyx.in

@@ -96,7 +96,7 @@ def main():
     NUM_THREADS = 128
     NUM_BLOCKS = 32
 
-    a = np.array([2.0], dtype=np.float32)
+    a = np.float32(2.0)
     n = np.array(NUM_THREADS * NUM_BLOCKS, dtype=np.uint32)
     bufferSize = n * a.itemsize
 

cuda/tests/test_cuda.py

@@ -1,4 +1,4 @@
-Cython==0.29.34
+Cython>=3.0.0
 pytest>=6.2.4
 pytest-benchmark>=3.4.1
 numpy>=1.21.1