Patch 11.8.1

- Resolve issue 27
- Update install instructions to use latest CTK

Author: vzhurba01

README.md

@@ -11,63 +11,15 @@ work are as follows:
 * CUDA Toolkit 11.0 to 11.8
 * Cython - e.g. 0.29.21
 
-### Compilation
+### Installing
 
-To compile the extension in-place, run:
-
-```
-python setup.py build_ext --inplace
-```
-
-To compile for debugging the extension modules with gdb, pass the `--debug`
-argument to setup.py.
-
-
-### Develop installation
-
-You can use
-
-```
-pip install -e .
-```
-
-to install the module  as editible in your current Python environment (e.g. for
-testing of porting other libraries to use the binding).
-
-
-### Build the Docs
-
-```
-conda env create -f docs_src/environment-docs.yml
-conda activate cuda-python-docs
-```
-Then compile and install `cuda-python` following the steps above.
-
-```
-cd docs_src
-make html
-open build/html/index.html
-```
-
-### Publish the Docs
-
-```
-git checkout gh-pages
-cd docs_src
-make html
-cp -a build/html/. ../docs/
-```
+Refer to documentation for installation options and requirements: [nvidia.github.io/cuda-python/](https://nvidia.github.io/cuda-python/install.html)
 
 ## Testing
 
 ### Requirements
 
-Dependencies of the test execution and some versions that are known to
-work are as follows:
-
-* numpy-1.19.5
-* scipy-1.6.3
-* pytest-benchmark-3.4.1
+Latest dependencies can be found in [requirements.txt](https://github.com/NVIDIA/cuda-python/blob/main/requirements.txt).
 
 ### Unit-tests
 
@@ -76,6 +28,7 @@ You can run the included tests with:
 ```
 python -m pytest
 ```
+
 ### Benchmark
 
 You can run benchmark only tests with:

cuda/_lib/utils.pxd.in

@@ -88,6 +88,17 @@ cdef class HelperCUjit_option:
     cdef vector[char*] _charstarstar # list of names
     cdef InputVoidPtrPtrHelper _voidstarstar # list of addresses
 {{endif}}
+{{if 'CUmemAllocationHandleType_enum' in found_types}}
+
+cdef class HelperCUmemAllocationHandleType:
+    cdef void* _cptr
+    cdef ccuda.CUmemAllocationHandleType_enum _type
+
+    # Return values
+    cdef int _int
+    cdef void* _handle
+    cdef unsigned int _d3dkmt_handle
+{{endif}}
 
 cdef class InputVoidPtrPtrHelper:
     cdef void** _cptr

cuda/_lib/utils.pyx.in

@@ -426,6 +426,61 @@ cdef class HelperCUjit_option:
     def cptr(self):
         return <void_ptr>self._cptr
 {{endif}}
+{{if 'CUmemAllocationHandleType_enum' in found_types}}
+
+cdef class HelperCUmemAllocationHandleType:
+    def __cinit__(self, attr):
+        self._type = attr.value
+        if False:
+            pass
+        {{if 'CU_MEM_HANDLE_TYPE_NONE' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_NONE,):
+            self._cptr = <void*>&self._int
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR,):
+            self._cptr = <void*>&self._int
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_WIN32' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_WIN32,):
+            self._cptr = <void*>&self._handle
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_WIN32_KMT' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_WIN32_KMT,):
+            self._cptr = <void*>&self._d3dkmt_handle
+        {{endif}}
+        else:
+            raise TypeError('Unsupported attribute: {}'.format(attr.name))
+
+    def __dealloc__(self):
+        pass
+
+    @property
+    def cptr(self):
+        return <void_ptr>self._cptr
+
+    def pyObj(self):
+        if False:
+            pass
+        {{if 'CU_MEM_HANDLE_TYPE_NONE' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_NONE,):
+            return self._int
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR,):
+            return self._int
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_WIN32' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_WIN32,):
+            return <void_ptr>self._handle
+        {{endif}}
+        {{if 'CU_MEM_HANDLE_TYPE_WIN32_KMT' in found_values}}
+        elif self._type in (ccuda.CUmemAllocationHandleType_enum.CU_MEM_HANDLE_TYPE_WIN32_KMT,):
+            return self._d3dkmt_handle
+        {{endif}}
+        else:
+            raise TypeError('Unsupported attribute: {}'.format(self._type))
+{{endif}}
 
 cdef class InputVoidPtrPtrHelper:
     def __cinit__(self, lst):

cuda/cuda.pyx.in

@@ -13365,7 +13365,7 @@ def cuDeviceGetAttribute(attrib not None : CUdevice_attribute, dev):
 {{if 'cuDeviceGetNvSciSyncAttributes' in found_functions}}
 
 @cython.embedsignature(True)
-def cuDeviceGetNvSciSyncAttributes(dev, int flags):
+def cuDeviceGetNvSciSyncAttributes(nvSciSyncAttrList, dev, int flags):
     """ Return NvSciSync attributes that this device can support.
 
     Returns in `nvSciSyncAttrList`, the properties of NvSciSync that this
@@ -13405,6 +13405,8 @@ def cuDeviceGetNvSciSyncAttributes(dev, int flags):
 
     Parameters
     ----------
+    nvSciSyncAttrList : Any
+        Return NvSciSync attributes supported.
     dev : :py:obj:`~.CUdevice`
         Valid Cuda Device to get NvSciSync attributes for.
     flags : int
@@ -13414,8 +13416,6 @@ def cuDeviceGetNvSciSyncAttributes(dev, int flags):
     -------
     CUresult
 
-    nvSciSyncAttrList : Any
-        Return NvSciSync attributes supported.
 
     See Also
     --------
@@ -13431,10 +13431,10 @@ def cuDeviceGetNvSciSyncAttributes(dev, int flags):
         pdev = int(CUdevice(dev))
         cdev = <ccuda.CUdevice>pdev
 
-    cdef void_ptr nvSciSyncAttrList = 0
-    cdef void* cnvSciSyncAttrList_ptr = <void*>nvSciSyncAttrList
+    cnvSciSyncAttrList = utils.HelperInputVoidPtr(nvSciSyncAttrList)
+    cdef void* cnvSciSyncAttrList_ptr = <void*><void_ptr>cnvSciSyncAttrList.cptr
     err = ccuda.cuDeviceGetNvSciSyncAttributes(cnvSciSyncAttrList_ptr, cdev, flags)
-    return (CUresult(err), nvSciSyncAttrList)
+    return (CUresult(err),)
 {{endif}}
 
 {{if 'cuDeviceSetMemPool' in found_functions}}
@@ -20150,8 +20150,8 @@ def cuMemGetHandleForAddressRange(dptr, size_t size, handleType not None : CUmem
         pdptr = int(CUdeviceptr(dptr))
         cdptr = <ccuda.CUdeviceptr><void_ptr>pdptr
 
-    cdef void_ptr handle = 0
-    cdef void* chandle_ptr = <void*>handle
+    cdef int handle = 0
+    cdef void* chandle_ptr = <void*>&handle
     cdef ccuda.CUmemRangeHandleType chandleType = handleType.value
     err = ccuda.cuMemGetHandleForAddressRange(chandle_ptr, cdptr, size, chandleType, flags)
     return (CUresult(err), handle)
@@ -20794,11 +20794,11 @@ def cuMemExportToShareableHandle(handle, handleType not None : CUmemAllocationHa
         phandle = int(CUmemGenericAllocationHandle(handle))
         chandle = <ccuda.CUmemGenericAllocationHandle><void_ptr>phandle
 
-    cdef void_ptr shareableHandle = 0
-    cdef void* cshareableHandle_ptr = <void*>shareableHandle
+    cdef utils.HelperCUmemAllocationHandleType cshareableHandle = utils.HelperCUmemAllocationHandleType(handleType)
+    cdef void* cshareableHandle_ptr = <void*><void_ptr>cshareableHandle.cptr
     cdef ccuda.CUmemAllocationHandleType chandleType = handleType.value
     err = ccuda.cuMemExportToShareableHandle(cshareableHandle_ptr, chandle, chandleType, flags)
-    return (CUresult(err), shareableHandle)
+    return (CUresult(err), cshareableHandle.pyObj())
 {{endif}}
 
 {{if 'cuMemImportFromShareableHandle' in found_functions}}
@@ -21560,11 +21560,11 @@ def cuMemPoolExportToShareableHandle(pool, handleType not None : CUmemAllocation
         ppool = int(CUmemoryPool(pool))
         cpool = <ccuda.CUmemoryPool><void_ptr>ppool
 
-    cdef void_ptr handle_out = 0
-    cdef void* chandle_out_ptr = <void*>handle_out
+    cdef utils.HelperCUmemAllocationHandleType chandle_out = utils.HelperCUmemAllocationHandleType(handleType)
+    cdef void* chandle_out_ptr = <void*><void_ptr>chandle_out.cptr
     cdef ccuda.CUmemAllocationHandleType chandleType = handleType.value
     err = ccuda.cuMemPoolExportToShareableHandle(chandle_out_ptr, cpool, chandleType, flags)
-    return (CUresult(err), handle_out)
+    return (CUresult(err), chandle_out.pyObj())
 {{endif}}
 
 {{if 'cuMemPoolImportFromShareableHandle' in found_functions}}

cuda/cudart.pyx.in

@@ -12312,7 +12312,7 @@ def cudaDeviceGetMemPool(int device):
 {{if 'cudaDeviceGetNvSciSyncAttributes' in found_functions}}
 
 @cython.embedsignature(True)
-def cudaDeviceGetNvSciSyncAttributes(int device, int flags):
+def cudaDeviceGetNvSciSyncAttributes(nvSciSyncAttrList, int device, int flags):
     """ Return NvSciSync attributes that this device can support.
 
     Returns in `nvSciSyncAttrList`, the properties of NvSciSync that this
@@ -12347,6 +12347,8 @@ def cudaDeviceGetNvSciSyncAttributes(int device, int flags):
 
     Parameters
     ----------
+    nvSciSyncAttrList : Any
+        Return NvSciSync attributes supported.
     device : int
         Valid Cuda Device to get NvSciSync attributes for.
     flags : int
@@ -12356,17 +12358,15 @@ def cudaDeviceGetNvSciSyncAttributes(int device, int flags):
     -------
     cudaError_t
 
-    nvSciSyncAttrList : Any
-        Return NvSciSync attributes supported.
 
     See Also
     --------
     :py:obj:`~.cudaImportExternalSemaphore`, :py:obj:`~.cudaDestroyExternalSemaphore`, :py:obj:`~.cudaSignalExternalSemaphoresAsync`, :py:obj:`~.cudaWaitExternalSemaphoresAsync`
     """
-    cdef void_ptr nvSciSyncAttrList = 0
-    cdef void* cnvSciSyncAttrList_ptr = <void*>nvSciSyncAttrList
+    cnvSciSyncAttrList = utils.HelperInputVoidPtr(nvSciSyncAttrList)
+    cdef void* cnvSciSyncAttrList_ptr = <void*><void_ptr>cnvSciSyncAttrList.cptr
     err = ccudart.cudaDeviceGetNvSciSyncAttributes(cnvSciSyncAttrList_ptr, device, flags)
-    return (cudaError_t(err), nvSciSyncAttrList)
+    return (cudaError_t(err),)
 {{endif}}
 
 {{if 'cudaDeviceGetP2PAttribute' in found_functions}}
@@ -19805,11 +19805,11 @@ def cudaMemPoolExportToShareableHandle(memPool, handleType not None : cudaMemAll
         pmemPool = int(cudaMemPool_t(memPool))
         cmemPool = <ccudart.cudaMemPool_t><void_ptr>pmemPool
 
-    cdef void_ptr shareableHandle = 0
-    cdef void* cshareableHandle_ptr = <void*>shareableHandle
+    cdef utils.HelperCUmemAllocationHandleType cshareableHandle = utils.HelperCUmemAllocationHandleType(handleType)
+    cdef void* cshareableHandle_ptr = <void*><void_ptr>cshareableHandle.cptr
     cdef ccudart.cudaMemAllocationHandleType chandleType = handleType.value
     err = ccudart.cudaMemPoolExportToShareableHandle(cshareableHandle_ptr, cmemPool, chandleType, flags)
-    return (cudaError_t(err), shareableHandle)
+    return (cudaError_t(err), cshareableHandle.pyObj())
 {{endif}}
 
 {{if 'cudaMemPoolImportFromShareableHandle' in found_functions}}
@@ -23522,7 +23522,7 @@ def cudaGraphMemFreeNodeGetParams(node):
         cnode = <ccudart.cudaGraphNode_t><void_ptr>pnode
 
     cdef void_ptr dptr_out = 0
-    cdef void* cdptr_out_ptr = <void*>dptr_out
+    cdef void* cdptr_out_ptr = <void*>&dptr_out
     err = ccudart.cudaGraphMemFreeNodeGetParams(cnode, cdptr_out_ptr)
     return (cudaError_t(err), dptr_out)
 {{endif}}

docs/.buildinfo

@@ -1,4 +1,4 @@
 # Sphinx build info version 1
 # This file hashes the configuration used when building these files. When it is not found, a full rebuild will be done.
-config: 2173daa5e1d85d4fd21eb27f7801cf13
+config: ad24118d52f507b84ce833bbfdacae42
 tags: 645f666f9bcd5a90fca523b33c5a78b7

docs/_sources/install.md.txt

@@ -27,15 +27,14 @@ conda install -c nvidia cuda-python
 
 ### Requirements
 
-Building dependencies:
-* cython>=0.29.24
+Installing from source requires the latest CUDA Toolkit (CTK), matching the major.minor of CUDA Python. The installed package will still be compatible with all minor CTK versions.
 
-Unit-test dependencies:
-* pytest>=6.2.4
-* pytest-benchmark>=3.4.1
-* numpy>=1.21.1
+Environment variable CUDA_HOME must be set to CTK root directory:
+```
+export CUDA_HOME=/usr/local/cuda
+```
 
-Latest list of dependencies can be found at [requirements.txt](https://github.com/NVIDIA/cuda-python/blob/main/requirements.txt)
+Remaining build and test dependencies are outlined in [requirements.txt](https://github.com/NVIDIA/cuda-python/blob/main/requirements.txt)
 
 ### In-place
 
@@ -56,7 +55,7 @@ You can use
 pip install -e .
 ```
 
-to install the module  as editible in your current Python environment (e.g. for
+to install the module as editible in your current Python environment (e.g. for
 testing of porting other libraries to use the binding).
 
 ## Build the Docs

docs/_sources/release.md.txt

@@ -5,6 +5,7 @@
 maxdepth: 3
 ---
 
+    11.8.1 <release/11.8.1-notes>
     11.8.0 <release/11.8.0-notes>
     11.7.1 <release/11.7.1-notes>
     11.7.0 <release/11.7.0-notes>

docs/_sources/release/11.8.1-notes.md.txt

@@ -0,0 +1,32 @@
+# CUDA Python 11.8.1 Release notes
+
+Released on November 4, 2022
+
+## Hightlights
+- Resolves [issue #27](https://github.com/NVIDIA/cuda-python/issues/24)
+- Update install instructions to use latest CTK
+
+## Limitations
+
+### CUDA Functions Not Supported in this Release
+
+- Symbol APIs
+    - cudaGraphExecMemcpyNodeSetParamsFromSymbol
+    - cudaGraphExecMemcpyNodeSetParamsToSymbol
+    - cudaGraphAddMemcpyNodeToSymbol
+    - cudaGraphAddMemcpyNodeFromSymbol
+    - cudaGraphMemcpyNodeSetParamsToSymbol
+    - cudaGraphMemcpyNodeSetParamsFromSymbol
+    - cudaMemcpyToSymbol
+    - cudaMemcpyFromSymbol
+    - cudaMemcpyToSymbolAsync
+    - cudaMemcpyFromSymbolAsync
+    - cudaGetSymbolAddress
+    - cudaGetSymbolSize
+    - cudaGetFuncBySymbol
+- Launch Options
+    - cudaLaunchKernel
+    - cudaLaunchCooperativeKernel
+    - cudaLaunchCooperativeKernelMultiDevice
+- cudaSetValidDevices
+- cudaVDPAUSetVDPAUDevice

docs/_static/documentation_options.js

@@ -1,6 +1,6 @@
 var DOCUMENTATION_OPTIONS = {
     URL_ROOT: document.getElementById("documentation_options").getAttribute('data-url_root'),
-    VERSION: '11.8.0',
+    VERSION: '11.8.1',
     LANGUAGE: 'None',
     COLLAPSE_INDEX: false,
     BUILDER: 'html',