Transitioned dpctl to require DPC++ 2023, removed host related functions, fixed linker crash, enabled SyclKernel.max_sub_group_size property

.github/workflows/generate-coverage.yaml

@@ -96,11 +96,17 @@ jobs:
       - name: Upload coverage data to coveralls.io
         shell: bash -l {0}
         run: |
+          echo "Processing c-api-coverage"
+          export DPCTL_LCOV_FN=$(find _skbuild -name dpctl.lcov)
+          grep "/tmp" $DPCTL_LCOV_FN
           coveralls-lcov -v -n \
-              $(find _skbuild -name dpctl.lcov) > dpctl-c-api-coverage.json
+              $DPCTL_LCOV_FN > dpctl-c-api-coverage.json
+          echo "Processing pytest-coverage"
+          export DPCTL_PYTEST_LCOV=$(find . -name dpctl_pytest.lcov)
+          grep "/tmp" $DPCTL_PYTEST_LCOV
           coveralls-lcov -v -n \
-              $(find . -name dpctl_pytest.lcov) > pytest-dpctl-c-api-coverage.json
-          # merge json files
+              $DPCTL_PYTEST_LCOV > pytest-dpctl-c-api-coverage.json
+          echo "Merging JSON files"
           python -c "import json; \
               fh1 = open('dpctl-c-api-coverage.json', 'r'); \
               f1 = json.load(fh1); fh1.close(); \
@@ -113,6 +119,7 @@ jobs:
           ls -lh dpctl-c-api-coverage.json pytest-dpctl-c-api-coverage.json \
                combined-dpctl-c-api-coverage.json \
                $(find _skbuild -name dpctl.lcov) $(find . -name dpctl_pytest.lcov)
+          echo "Merging combined files with coverage data"
           coveralls --service=github --merge=combined-dpctl-c-api-coverage.json
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/generate-docs.yml

@@ -63,7 +63,7 @@ jobs:
           source /opt/intel/oneapi/setvars.sh
           wget https://github.com/vovkos/doxyrest/releases/download/doxyrest-2.1.2/doxyrest-2.1.2-linux-amd64.tar.xz
           tar xf doxyrest-2.1.2-linux-amd64.tar.xz
-          python setup.py develop -G Ninja --build-type=Debug \
+          python setup.py develop -G Ninja --build-type=Release \
               --                                              \
               -DCMAKE_C_COMPILER:PATH=$(which icx)            \
               -DCMAKE_CXX_COMPILER:PATH=$(which icpx)         \

CHANGELOG.md

@@ -8,8 +8,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+* Added `dpctl.program.SyclKernel.max_sub_group_size` property [#1208](https://github.com/IntelPython/dpctl/pull/1028).
+
 ### Changed
 
+* Removed `dpctl.select_host_device`, `dpctl.has_host_device`, `dpctl.SyclDevice.is_host`, and `dpctl.SyclDevice.has_aspect_host` since support for host device has been removed in DPC++ 2023 and from SYCL 2020 spec [#1208](https://github.com/IntelPython/dpctl/pull/1028).
+
 ### Fixed
 
 

dpctl/CMakeLists.txt

@@ -18,10 +18,10 @@ if(WIN32)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /Ox ${WARNING_FLAGS}")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Ox ${WARNING_FLAGS}")
     set(CMAKE_C_FLAGS_DEBUG
-        "${CMAKE_C_FLAGS_DEBUG} ${WARNING_FLAGS} -O0 -ggdb3 -DDEBUG"
+        "${CMAKE_C_FLAGS_DEBUG} ${WARNING_FLAGS} -O0 -g1 -DDEBUG"
     )
     set(CMAKE_CXX_FLAGS_DEBUG
-        "${CMAKE_CXX_FLAGS_DEBUG} ${WARNING_FLAGS} -O0 -ggdb3 -DDEBUG"
+        "${CMAKE_CXX_FLAGS_DEBUG} ${WARNING_FLAGS} -O0 -g1 -DDEBUG"
     )
     set(DPCTL_LDFLAGS "/link /NXCompat /DynamicBase")
 elseif(UNIX)
@@ -57,10 +57,10 @@ elseif(UNIX)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3 ${CFLAGS}")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 ${CXXFLAGS}")
     set(CMAKE_C_FLAGS_DEBUG
-        "${CMAKE_C_FLAGS_DEBUG} ${CFLAGS} -O0 -ggdb3 -DDEBUG"
+        "${CMAKE_C_FLAGS_DEBUG} ${CFLAGS} -O0 -g1 -DDEBUG"
     )
     set(CMAKE_CXX_FLAGS_DEBUG
-        "${CMAKE_CXX_FLAGS_DEBUG} ${CXXFLAGS} -O0 -ggdb3 -DDEBUG"
+        "${CMAKE_CXX_FLAGS_DEBUG} ${CXXFLAGS} -O0 -g1 -DDEBUG"
     )
     set(DPCTL_LDFLAGS "-z,noexecstack,-z,relro,-z,now")
 else()

dpctl/__init__.py

@@ -48,12 +48,10 @@
     has_accelerator_devices,
     has_cpu_devices,
     has_gpu_devices,
-    has_host_device,
     select_accelerator_device,
     select_cpu_device,
     select_default_device,
     select_gpu_device,
-    select_host_device,
 )
 from ._sycl_event import SyclEvent
 from ._sycl_platform import SyclPlatform, get_platforms, lsplatform

dpctl/_backend.pxd

@@ -37,7 +37,6 @@ cdef extern from "syclinterface/dpctl_sycl_enum_types.h":
     ctypedef enum _backend_type 'DPCTLSyclBackendType':
         _ALL_BACKENDS    'DPCTL_ALL_BACKENDS'
         _CUDA            'DPCTL_CUDA'
-        _HOST            'DPCTL_HOST'
         _LEVEL_ZERO      'DPCTL_LEVEL_ZERO'
         _OPENCL          'DPCTL_OPENCL'
         _UNKNOWN_BACKEND 'DPCTL_UNKNOWN_BACKEND'
@@ -49,7 +48,6 @@ cdef extern from "syclinterface/dpctl_sycl_enum_types.h":
         _CPU            'DPCTL_CPU'
         _CUSTOM         'DPCTL_CUSTOM'
         _GPU            'DPCTL_GPU'
-        _HOST_DEVICE    'DPCTL_HOST_DEVICE'
         _UNKNOWN_DEVICE 'DPCTL_UNKNOWN_DEVICE'
 
     ctypedef enum _arg_data_type 'DPCTLKernelArgType':
@@ -174,7 +172,6 @@ cdef extern from "syclinterface/dpctl_sycl_device_interface.h":
     cdef bool DPCTLDevice_IsAccelerator(const DPCTLSyclDeviceRef DRef)
     cdef bool DPCTLDevice_IsCPU(const DPCTLSyclDeviceRef DRef)
     cdef bool DPCTLDevice_IsGPU(const DPCTLSyclDeviceRef DRef)
-    cdef bool DPCTLDevice_IsHost(const DPCTLSyclDeviceRef DRef)
     cdef bool DPCTLDevice_GetSubGroupIndependentForwardProgress(const DPCTLSyclDeviceRef DRef)
     cdef uint32_t DPCTLDevice_GetPreferredVectorWidthChar(const DPCTLSyclDeviceRef DRef)
     cdef uint32_t DPCTLDevice_GetPreferredVectorWidthShort(const DPCTLSyclDeviceRef DRef)
@@ -236,7 +233,6 @@ cdef extern from "syclinterface/dpctl_sycl_device_selector_interface.h":
     DPCTLSyclDeviceSelectorRef DPCTLCPUSelector_Create()
     DPCTLSyclDeviceSelectorRef DPCTLFilterSelector_Create(const char *)
     DPCTLSyclDeviceSelectorRef DPCTLGPUSelector_Create()
-    DPCTLSyclDeviceSelectorRef DPCTLHostSelector_Create()
     void DPCTLDeviceSelector_Delete(DPCTLSyclDeviceSelectorRef)
     int DPCTLDeviceSelector_Score(DPCTLSyclDeviceSelectorRef, DPCTLSyclDeviceRef)
 
@@ -271,8 +267,7 @@ cdef extern from "syclinterface/dpctl_sycl_kernel_interface.h":
     cdef size_t DPCTLKernel_GetPreferredWorkGroupSizeMultiple(const DPCTLSyclKernelRef KRef)
     cdef size_t DPCTLKernel_GetPrivateMemSize(const DPCTLSyclKernelRef KRef)
     cdef uint32_t DPCTLKernel_GetMaxNumSubGroups(const DPCTLSyclKernelRef KRef)
-##  Next line is commented out due to issue in DPC++ runtime
-#   cdef uint32_t DPCTLKernel_GetMaxSubGroupSize(const DPCTLSyclKernelRef KRef)
+    cdef uint32_t DPCTLKernel_GetMaxSubGroupSize(const DPCTLSyclKernelRef KRef)
     cdef uint32_t DPCTLKernel_GetCompileNumSubGroups(const DPCTLSyclKernelRef KRef)
     cdef uint32_t DPCTLKernel_GetCompileSubGroupSize(const DPCTLSyclKernelRef KRef)
 

dpctl/_sycl_device.pyx

@@ -73,7 +73,6 @@ from ._backend cimport (  # noqa: E211
     DPCTLDevice_IsAccelerator,
     DPCTLDevice_IsCPU,
     DPCTLDevice_IsGPU,
-    DPCTLDevice_IsHost,
     DPCTLDeviceMgr_GetDeviceInfoStr,
     DPCTLDeviceMgr_GetDevices,
     DPCTLDeviceMgr_GetPositionInDevices,
@@ -162,8 +161,6 @@ cdef list _get_devices(DPCTLDeviceVectorRef DVRef):
 cdef str _backend_type_to_filter_string_part(_backend_type BTy):
     if BTy == _backend_type._CUDA:
         return "cuda"
-    elif BTy == _backend_type._HOST:
-        return "host"
     elif BTy == _backend_type._LEVEL_ZERO:
         return "level_zero"
     elif BTy == _backend_type._OPENCL:
@@ -181,8 +178,6 @@ cdef str _device_type_to_filter_string_part(_device_type DTy):
         return "cpu"
     elif DTy == _device_type._GPU:
         return "gpu"
-    elif DTy == _device_type._HOST_DEVICE:
-        return "host"
     else:
         return "unknown"
 
@@ -222,7 +217,6 @@ cdef class SyclDevice(_SyclDevice):
           :func:`dpctl.select_cpu_device()`,
           :func:`dpctl.select_default_device()`,
           :func:`dpctl.select_gpu_device()`,
-          :func:`dpctl.select_host_device()`.
 
 
         :Example:
@@ -359,8 +353,6 @@ cdef class SyclDevice(_SyclDevice):
         )
         if BTy == _backend_type._CUDA:
             return backend_type.cuda
-        elif BTy == _backend_type._HOST:
-            return backend_type.host
         elif BTy == _backend_type._LEVEL_ZERO:
             return backend_type.level_zero
         elif BTy == _backend_type._OPENCL:
@@ -388,21 +380,9 @@ cdef class SyclDevice(_SyclDevice):
             return device_type.cpu
         elif DTy == _device_type._GPU:
             return device_type.gpu
-        elif DTy == _device_type._HOST_DEVICE:
-            return device_type.host
         else:
             raise ValueError("Unknown device type.")
 
-    @property
-    def has_aspect_host(self):
-        """ Returns True if this device is a host device, False otherwise.
-
-        Returns:
-            bool: Indicates if the device is a host device.
-        """
-        cdef _aspect_type AT = _aspect_type._host
-        return DPCTLDevice_HasAspect(self._device_ref, AT)
-
     @property
     def has_aspect_cpu(self):
         """ Returns True if this device is a CPU device, False otherwise.
@@ -729,15 +709,6 @@ cdef class SyclDevice(_SyclDevice):
         """
         return DPCTLDevice_IsGPU(self._device_ref)
 
-    @property
-    def is_host(self):
-        """ Returns True if the SyclDevice instance is a SYCL host device.
-
-        Returns:
-            bool: True if the SyclDevice is a SYCL host device, else False.
-        """
-        return DPCTLDevice_IsHost(self._device_ref)
-
     @property
     def max_work_item_dims(self):
         """ Returns the maximum dimensions that specify the global and local
@@ -866,11 +837,9 @@ cdef class SyclDevice(_SyclDevice):
             int: The maximum number of sub-groups support per work-group by
             the device.
         """
-        cdef uint32_t max_num_sub_groups = 0
-        if (not self.is_host):
-            max_num_sub_groups = (
-                DPCTLDevice_GetMaxNumSubGroups(self._device_ref)
-            )
+        cdef uint32_t max_num_sub_groups = (
+            DPCTLDevice_GetMaxNumSubGroups(self._device_ref)
+        )
         return max_num_sub_groups
 
     @property

dpctl/_sycl_device_factory.pxd

@@ -31,10 +31,8 @@ cpdef SyclDevice select_accelerator_device()
 cpdef SyclDevice select_cpu_device()
 cpdef SyclDevice select_default_device()
 cpdef SyclDevice select_gpu_device()
-cpdef SyclDevice select_host_device()
 cpdef list get_devices(backend=*, device_type=*)
 cpdef int get_num_devices(backend=*, device_type=*)
 cpdef cpp_bool has_gpu_devices()
 cpdef cpp_bool has_cpu_devices()
 cpdef cpp_bool has_accelerator_devices()
-cpdef cpp_bool has_host_device()

dpctl/_sycl_device_factory.pyx

@@ -39,7 +39,6 @@ from ._backend cimport (  # noqa: E211
     DPCTLDeviceVector_Size,
     DPCTLDeviceVectorRef,
     DPCTLGPUSelector_Create,
-    DPCTLHostSelector_Create,
     DPCTLSyclDeviceRef,
     DPCTLSyclDeviceSelectorRef,
     _backend_type,
@@ -56,12 +55,10 @@ __all__ = [
     "select_cpu_device",
     "select_default_device",
     "select_gpu_device",
-    "select_host_device",
     "get_num_devices",
     "has_cpu_devices",
     "has_gpu_devices",
     "has_accelerator_devices",
-    "has_host_device",
 ]
 
 
@@ -71,8 +68,6 @@ cdef _backend_type _string_to_dpctl_sycl_backend_ty(str backend_str):
         return _backend_type._ALL_BACKENDS
     elif backend_str == "cuda":
         return _backend_type._CUDA
-    elif backend_str == "host":
-        return _backend_type._HOST
     elif backend_str == "level_zero":
         return _backend_type._LEVEL_ZERO
     elif backend_str == "opencl":
@@ -95,8 +90,6 @@ cdef _device_type _string_to_dpctl_sycl_device_ty(str dty_str):
         return _device_type._CUSTOM
     elif dty_str == "gpu":
         return _device_type._GPU
-    elif dty_str == "host":
-        return _device_type._HOST_DEVICE
     else:
         return _device_type._UNKNOWN_DEVICE
 
@@ -106,8 +99,6 @@ cdef _backend_type _enum_to_dpctl_sycl_backend_ty(BTy):
         return _backend_type._ALL_BACKENDS
     elif BTy == backend_type.cuda:
         return _backend_type._CUDA
-    elif BTy == backend_type.host:
-        return _backend_type._HOST
     elif BTy == backend_type.level_zero:
         return _backend_type._LEVEL_ZERO
     elif BTy == backend_type.opencl:
@@ -129,8 +120,6 @@ cdef _device_type _enum_to_dpctl_sycl_device_ty(DTy):
         return _device_type._CUSTOM
     elif DTy == device_type_t.gpu:
         return _device_type._GPU
-    elif DTy == device_type_t.host:
-        return _device_type._HOST_DEVICE
     else:
         return _device_type._UNKNOWN_DEVICE
 
@@ -288,19 +277,6 @@ cpdef cpp_bool has_accelerator_devices():
     return <cpp_bool>num_accelerator_dev
 
 
-cpdef cpp_bool has_host_device():
-    """ A helper function to check if there are any SYCL Host devices available.
-
-    Returns:
-        bool: ``True`` if ``sycl::device_type::host`` devices are present,
-        ``False`` otherwise.
-    """
-    cdef int num_host_dev = DPCTLDeviceMgr_GetNumDevices(
-        _device_type._HOST_DEVICE
-    )
-    return <cpp_bool>num_host_dev
-
-
 cpdef SyclDevice select_accelerator_device():
     """ A wrapper for SYCL's ``accelerator_selector`` class.
 
@@ -379,23 +355,3 @@ cpdef SyclDevice select_gpu_device():
         raise SyclDeviceCreationError("Device unavailable.")
     Device = SyclDevice._create(DRef)
     return Device
-
-
-cpdef SyclDevice select_host_device():
-    """ A wrapper for SYCL's ``host_selector`` class.
-
-    Returns:
-        dpctl.SyclDevice: A Python object wrapping the SYCL ``device``
-        returned by the SYCL ``host_selector``.
-    Raises:
-        dpctl.SyclDeviceCreationError: If the SYCL ``host_selector`` is
-        unable to select a ``device``.
-    """
-    cdef DPCTLSyclDeviceSelectorRef DSRef = DPCTLHostSelector_Create()
-    cdef DPCTLSyclDeviceRef DRef = DPCTLDevice_CreateFromSelector(DSRef)
-    # Free up the device selector
-    DPCTLDeviceSelector_Delete(DSRef)
-    if DRef is NULL:
-        raise SyclDeviceCreationError("Host device is unavailable.")
-    Device = SyclDevice._create(DRef)
-    return Device

dpctl/_sycl_event.pyx

@@ -302,8 +302,6 @@ cdef class SyclEvent(_SyclEvent):
             return backend_type.opencl
         elif BE == _backend_type._LEVEL_ZERO:
             return backend_type.level_zero
-        elif BE == _backend_type._HOST:
-            return backend_type.host
         elif BE == _backend_type._CUDA:
             return backend_type.cuda
         else:

dpctl/_sycl_platform.pyx

@@ -251,8 +251,6 @@ cdef class SyclPlatform(_SyclPlatform):
         )
         if BTy == _backend_type._CUDA:
             return backend_type.cuda
-        elif BTy == _backend_type._HOST:
-            return backend_type.host
         elif BTy == _backend_type._LEVEL_ZERO:
             return backend_type.level_zero
         elif BTy == _backend_type._OPENCL:

dpctl/_sycl_queue.pyx

@@ -676,8 +676,6 @@ cdef class SyclQueue(_SyclQueue):
             return backend_type.opencl
         elif BE == _backend_type._LEVEL_ZERO:
             return backend_type.level_zero
-        elif BE == _backend_type._HOST:
-            return backend_type.host
         elif BE == _backend_type._CUDA:
             return backend_type.cuda
         else:

dpctl/enum_types.py

@@ -54,7 +54,6 @@ class device_type(Enum):
     cpu = auto()
     custom = auto()
     gpu = auto()
-    host = auto()
 
 
 class backend_type(Enum):
@@ -74,7 +73,6 @@ class backend_type(Enum):
 
     all = auto()
     cuda = auto()
-    host = auto()
     level_zero = auto()
     opencl = auto()