Merge remote-tracking branch 'origin/master' into gold/2021

Author: oleksandr-pavlyk

.github/workflows/conda-package.yml

@@ -14,7 +14,7 @@ env:
 
 jobs:
   build_linux:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-20.04
 
     strategy:
       matrix:
@@ -104,7 +104,7 @@ jobs:
       matrix:
         python: ['3.8', '3.9', '3.10']
         experimental: [false]
-        runner: [ubuntu-latest]
+        runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
       CHANNELS: -c intel -c main --override-channels
@@ -150,9 +150,10 @@ jobs:
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-
       - name: Install dpctl
         run: |
-          CHANNELS="-c $GITHUB_WORKSPACE/channel ${{ env.CHANNELS }}"
+          export CHANNELS="-c $GITHUB_WORKSPACE/channel ${{ env.CHANNELS }}"
+          export TEST_DEPENDENCIES="pytest pytest-cov cython"
           export PACKAGE_VERSION=$(python -c "${VER_SCRIPT1} ${VER_SCRIPT2}")
-          conda create -n test_dpctl $PACKAGE_NAME=${PACKAGE_VERSION} pytest python=${{ matrix.python }} $CHANNELS
+          conda create -n test_dpctl $PACKAGE_NAME=${PACKAGE_VERSION} ${TEST_DEPENDENCIES} python=${{ matrix.python }} ${CHANNELS}
           # Test installed packages
           conda list -n test_dpctl
       - name: Smoke test
@@ -249,10 +250,12 @@ jobs:
           FOR /F "tokens=* USEBACKQ" %%F IN (`python -c "%SCRIPT%"`) DO (
              SET PACKAGE_VERSION=%%F
           )
-          conda install ${{ env.PACKAGE_NAME }}=%PACKAGE_VERSION% pytest python=${{ matrix.python }} -c ${{ env.workdir }}/channel ${{ env.CHANNELS }}
-      - name: Report content of test environemtn
+          SET "TEST_DEPENDENCIES=pytest pytest-cov cython"
+          conda install ${{ env.PACKAGE_NAME }}=%PACKAGE_VERSION% %TEST_DEPENDENCIES% python=${{ matrix.python }} -c ${{ env.workdir }}/channel ${{ env.CHANNELS }}
+      - name: Report content of test environment
         shell: cmd /C CALL {0}
         run: |
+          pip install --no-cache-dir brotli
           echo "Value of CONDA enviroment variable was: " %CONDA%
           conda list
       - name: Add library
@@ -307,7 +310,7 @@ jobs:
   upload_linux:
     needs: test_linux
     if: ${{github.ref == 'refs/heads/master' || (startsWith(github.ref, 'refs/heads/release') == true) || github.event_name == 'push' && contains(github.ref, 'refs/tags/')}}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-20.04
     strategy:
       matrix:
         python: ['3.8', '3.9', '3.10']
@@ -362,7 +365,7 @@ jobs:
       matrix:
         python: ['3.9']
         experimental: [false]
-        runner: [ubuntu-latest]
+        runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
       CHANNELS: -c intel -c main --override-channels
@@ -507,7 +510,7 @@ jobs:
       matrix:
         python: ['3.10']
         experimental: [false]
-        runner: [ubuntu-latest]
+        runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
       CHANNELS: -c intel -c main --override-channels

.github/workflows/pre-commit.yml

@@ -7,10 +7,13 @@ on:
 
 jobs:
   pre-commit:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-20.04
     steps:
     - uses: actions/checkout@v3
     - uses: actions/setup-python@v4
       with:
         python-version: '3.10'
+    - name: Version of clang-format
+      run: |
+        clang-format --version
     - uses: pre-commit/[email protected]

CHANGELOG.md

@@ -4,6 +4,15 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [dev]
+
+### Added
+
+### Changed
+
+### Fixed
+
+
 ## [0.14.0] - 11/18/2022
 
 ### Added

conda-recipe/build.sh

@@ -5,7 +5,9 @@
 # activation scripts.
 export LDFLAGS="$LDFLAGS -Wl,-rpath,$PREFIX/lib"
 
-${PYTHON} setup.py clean --all
+if [ -e "_skbuild" ]; then
+    ${PYTHON} setup.py clean --all
+fi
 export CMAKE_GENERATOR="Ninja"
 SKBUILD_ARGS="-- -DCMAKE_C_COMPILER:PATH=icx -DCMAKE_CXX_COMPILER:PATH=icpx"
 echo "${PYTHON} setup.py install ${SKBUILD_ARGS}"

conda-recipe/meta.yaml

@@ -16,10 +16,11 @@ requirements:
         - {{ compiler('dpcpp') }}  >=2022.1  # [not osx]
     host:
         - setuptools
-        - cython
         - cmake  >=3.21
-        - python
         - ninja
+        - git
+        - cython
+        - python
         - scikit-build
         - numpy 1.19
         - wheel

dpctl/__main__.py

@@ -19,6 +19,7 @@
 import os.path
 import platform
 import sys
+import warnings
 
 import dpctl
 
@@ -50,6 +51,24 @@ def print_library() -> None:
     print(ld_flags + " -lSyclInterface")
 
 
+def _warn_if_any_set(args, li) -> None:
+    opts_set = [it for it in li if getattr(args, it, True)]
+    if opts_set:
+        if len(opts_set) == 1:
+            warnings.warn(
+                "The option " + str(opts_set[0]) + " is being ignored.",
+                stacklevel=3,
+            )
+        else:
+            warnings.warn(
+                "Options " + str(opts_set) + " are being ignored.", stacklevel=3
+            )
+
+
+def print_lsplatform(verbosity: int) -> None:
+    dpctl.lsplatform(verbosity=verbosity)
+
+
 def main() -> None:
     """Main entry-point."""
     parser = argparse.ArgumentParser()
@@ -68,9 +87,45 @@ def main() -> None:
         action="store_true",
         help="Linker flags for SyclInterface library.",
     )
+    parser.add_argument(
+        "-f",
+        "--full-list",
+        action="store_true",
+        help="Enumerate system platforms, using dpctl.lsplatform(verbosity=2)",
+    )
+    parser.add_argument(
+        "-l",
+        "--long-list",
+        action="store_true",
+        help="Enumerate system platforms, using dpctl.lsplatform(verbosity=1)",
+    )
+    parser.add_argument(
+        "-s",
+        "--summary",
+        action="store_true",
+        help="Enumerate system platforms, using dpctl.lsplatform()",
+    )
     args = parser.parse_args()
     if not sys.argv[1:]:
         parser.print_help()
+    if args.full_list:
+        _warn_if_any_set(
+            args, ["long_list", "summary", "includes", "cmakedir", "library"]
+        )
+        print_lsplatform(2)
+        return
+    if args.long_list:
+        _warn_if_any_set(
+            args, ["full_list", "summary", "includes", "cmakedir", "library"]
+        )
+        print_lsplatform(1)
+        return
+    if args.summary:
+        _warn_if_any_set(
+            args, ["long_list", "full_list", "includes", "cmakedir", "library"]
+        )
+        print_lsplatform(0)
+        return
     if args.includes:
         print_includes()
     if args.cmakedir:

dpctl/_backend.pxd

@@ -206,6 +206,7 @@ cdef extern from "syclinterface/dpctl_sycl_device_interface.h":
         const DPCTLSyclDeviceRef DRef)
     cdef size_t *DPCTLDevice_GetSubGroupSizes(const DPCTLSyclDeviceRef DRef,
         size_t *res_len)
+    cdef uint32_t DPCTLDevice_GetPartitionMaxSubDevices(const DPCTLSyclDeviceRef DRef)
 
 
 cdef extern from "syclinterface/dpctl_sycl_device_manager.h":

dpctl/_sycl_device.pyx

@@ -55,6 +55,7 @@ from ._backend cimport (  # noqa: E211
     DPCTLDevice_GetMaxWriteImageArgs,
     DPCTLDevice_GetName,
     DPCTLDevice_GetParentDevice,
+    DPCTLDevice_GetPartitionMaxSubDevices,
     DPCTLDevice_GetPlatform,
     DPCTLDevice_GetPreferredVectorWidthChar,
     DPCTLDevice_GetPreferredVectorWidthDouble,
@@ -1292,9 +1293,25 @@ cdef class SyclDevice(_SyclDevice):
         """
         cdef uint64_t cache_line_sz = DPCTLDevice_GetGlobalMemCacheLineSize(
             self._device_ref
-	)
+	    )
         return cache_line_sz
 
+    @property
+    def partition_max_sub_devices(self):
+        """ The maximum number of sub-devices this :class:`dpctl.SyclDevice`
+        instance can be partitioned into. The value returned cannot exceed the
+        value returned by :attr:`dpctl.SyclDevice.max_compute_units`.
+
+        Returns:
+            int: The maximum number of sub-devices that can be created when this
+                device is partitioned. Zero value indicates that device can not
+                be partitioned.
+        """
+        cdef uint32_t max_part = DPCTLDevice_GetPartitionMaxSubDevices(
+            self._device_ref
+        )
+        return max_part
+
     cdef cpp_bool equals(self, SyclDevice other):
         """ Returns ``True`` if the :class:`dpctl.SyclDevice` argument has the
         same _device_ref as this SyclDevice.

dpctl/apis/include/dpctl4pybind11.hpp

@@ -150,7 +150,13 @@ class dpctl_capi
         return PyObject_TypeCheck(obj, PySyclProgramType_) != 0;
     }
 
-    ~dpctl_capi(){};
+    ~dpctl_capi()
+    {
+        as_usm_memory_.reset();
+        default_usm_ndarray_.reset();
+        default_usm_memory_.reset();
+        default_sycl_queue_.reset();
+    };
 
     static auto &get()
     {
@@ -194,8 +200,36 @@ class dpctl_capi
     std::shared_ptr<py::object> as_usm_memory_;
 
     dpctl_capi()
-        : default_sycl_queue_{}, default_usm_memory_{}, default_usm_ndarray_{},
-          as_usm_memory_{}
+        : Py_SyclDeviceType_(nullptr), PySyclDeviceType_(nullptr),
+          Py_SyclContextType_(nullptr), PySyclContextType_(nullptr),
+          Py_SyclEventType_(nullptr), PySyclEventType_(nullptr),
+          Py_SyclQueueType_(nullptr), PySyclQueueType_(nullptr),
+          Py_MemoryType_(nullptr), PyMemoryUSMDeviceType_(nullptr),
+          PyMemoryUSMSharedType_(nullptr), PyMemoryUSMHostType_(nullptr),
+          PyUSMArrayType_(nullptr), PySyclProgramType_(nullptr),
+          PySyclKernelType_(nullptr), SyclDevice_GetDeviceRef_(nullptr),
+          SyclDevice_Make_(nullptr), SyclContext_GetContextRef_(nullptr),
+          SyclContext_Make_(nullptr), SyclEvent_GetEventRef_(nullptr),
+          SyclEvent_Make_(nullptr), SyclQueue_GetQueueRef_(nullptr),
+          SyclQueue_Make_(nullptr), Memory_GetUsmPointer_(nullptr),
+          Memory_GetContextRef_(nullptr), Memory_GetQueueRef_(nullptr),
+          Memory_GetNumBytes_(nullptr), Memory_Make_(nullptr),
+          SyclKernel_GetKernelRef_(nullptr), SyclKernel_Make_(nullptr),
+          SyclProgram_GetKernelBundleRef_(nullptr), SyclProgram_Make_(nullptr),
+          UsmNDArray_GetData_(nullptr), UsmNDArray_GetNDim_(nullptr),
+          UsmNDArray_GetShape_(nullptr), UsmNDArray_GetStrides_(nullptr),
+          UsmNDArray_GetTypenum_(nullptr), UsmNDArray_GetElementSize_(nullptr),
+          UsmNDArray_GetFlags_(nullptr), UsmNDArray_GetQueueRef_(nullptr),
+          UsmNDArray_GetOffset_(nullptr), USM_ARRAY_C_CONTIGUOUS_(0),
+          USM_ARRAY_F_CONTIGUOUS_(0), USM_ARRAY_WRITABLE_(0), UAR_BOOL_(-1),
+          UAR_SHORT_(-1), UAR_USHORT_(-1), UAR_INT_(-1), UAR_UINT_(-1),
+          UAR_LONG_(-1), UAR_ULONG_(-1), UAR_LONGLONG_(-1), UAR_ULONGLONG_(-1),
+          UAR_FLOAT_(-1), UAR_DOUBLE_(-1), UAR_CFLOAT_(-1), UAR_CDOUBLE_(-1),
+          UAR_TYPE_SENTINEL_(-1), UAR_HALF_(-1), UAR_INT8_(-1), UAR_UINT8_(-1),
+          UAR_INT16_(-1), UAR_UINT16_(-1), UAR_INT32_(-1), UAR_UINT32_(-1),
+          UAR_INT64_(-1), UAR_UINT64_(-1), default_sycl_queue_{},
+          default_usm_memory_{}, default_usm_ndarray_{}, as_usm_memory_{}
+
     {
         // Import Cython-generated C-API for dpctl
         // This imports python modules and initializes

dpctl/tensor/__init__.py

@@ -58,6 +58,11 @@
     squeeze,
     stack,
 )
+from dpctl.tensor._print import (
+    get_print_options,
+    print_options,
+    set_print_options,
+)
 from dpctl.tensor._reshape import reshape
 from dpctl.tensor._usmarray import usm_ndarray
 
@@ -129,4 +134,7 @@
     "can_cast",
     "result_type",
     "meshgrid",
+    "get_print_options",
+    "set_print_options",
+    "print_options",
 ]

dpctl/tensor/_ctors.py

@@ -773,14 +773,20 @@ def full(
 
     sycl_queue = normalize_queue_device(sycl_queue=sycl_queue, device=device)
     usm_type = usm_type if usm_type is not None else "device"
-    dtype = _get_dtype(dtype, sycl_queue, ref_type=type(fill_value))
+    fill_value_type = type(fill_value)
+    dtype = _get_dtype(dtype, sycl_queue, ref_type=fill_value_type)
     res = dpt.usm_ndarray(
         sh,
         dtype=dtype,
         buffer=usm_type,
         order=order,
         buffer_ctor_kwargs={"queue": sycl_queue},
     )
+    if fill_value_type in [float, complex] and np.issubdtype(dtype, np.integer):
+        fill_value = int(fill_value.real)
+    elif fill_value_type is complex and np.issubdtype(dtype, np.floating):
+        fill_value = fill_value.real
+
     hev, _ = ti._full_usm_ndarray(fill_value, res, sycl_queue)
     hev.wait()
     return res