Merge branch 'master' into aspect_selector

oleksandr-pavlyk · web-flow · commit 5c57730a1580 · 2021-09-01T10:24:41.000-05:00
diff --git a/.github/workflows/conda-package.yml b/.github/workflows/conda-package.yml
@@ -8,52 +8,54 @@ env:
 
 jobs:
   build:
-    runs-on: ubuntu-latest
+    runs-on: ${{ matrix.os }}
     strategy:
       matrix:
+        os: [ubuntu-latest, windows-latest]
+        include:
+        - os: ubuntu-latest
+          pkgs_dirs: ~/.conda/pkgs
+          condarc: ~/.condarc
+          conda-bld: /usr/share/miniconda/conda-bld/linux-64/
+        - os: windows-latest
+          pkgs_dirs: C:\Users\runneradmin\.conda\pkgs
+          condarc: C:\Users\runneradmin\.condarc
+          conda-bld: C:\Miniconda\conda-bld\win-64\
         python: [3.8]
     steps:
       - uses: actions/checkout@v2
         with:
           fetch-depth: 0
+      - uses: conda-incubator/setup-miniconda@v2
+        with:
+          auto-activate-base: true
+          activate-environment: ""
 
       - name: Set pkgs_dirs
         run: |
-          echo "pkgs_dirs: [~/.conda/pkgs]" >> ~/.condarc
+          echo "pkgs_dirs: [${{ matrix.pkgs_dirs }}]" >> ${{ matrix.condarc }}
       - name: Cache conda packages
         uses: actions/cache@v2
         env:
           CACHE_NUMBER: 0  # Increase to reset cache
         with:
-          path: ~/.conda/pkgs
+          path: ${{ matrix.pkgs_dirs }}
           key:
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-${{hashFiles('**/meta.yaml') }}
           restore-keys: |
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-
-
-      - name: Add conda to system path
-        run: echo $CONDA/bin >> $GITHUB_PATH
       - name: Install conda-build
         run: conda install conda-build
       - name: Build conda package
-        run: |
-          CHANNELS="-c intel -c defaults --override-channels"
-          VERSIONS="--python ${{ matrix.python }}"
-          TEST="--no-test"
-
-          conda build \
-            $TEST \
-            $VERSIONS \
-            $CHANNELS \
-            conda-recipe
+        run: conda build --no-test --python ${{ matrix.python }} -c intel -c defaults --override-channels conda-recipe
       - name: Upload artifact
         uses: actions/upload-artifact@v2
         with:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
-          path: /usr/share/miniconda/conda-bld/linux-64/${{ env.PACKAGE_NAME }}-*.tar.bz2
+          path: ${{ matrix.conda-bld }}${{ env.PACKAGE_NAME }}-*.tar.bz2
 
-  test:
+  test_linux:
     needs: build
     runs-on:  ${{ matrix.runner }}
 
@@ -62,10 +64,10 @@ jobs:
         python: [3.8]
         experimental: [false]
         runner: [ubuntu-latest]
-        include:
-          - python: 3.8
-            experimental: true
-            runner: [self-hosted, gen9]
+        # include:
+        #   - python: 3.8
+        #     experimental: true
+        #     runner: [self-hosted, gen9]
     continue-on-error: ${{ matrix.experimental }}
     env:
       CHANNELS: -c intel -c defaults --override-channels
@@ -86,7 +88,6 @@ jobs:
           conda index $GITHUB_WORKSPACE/channel
           # Test channel
           conda search $PACKAGE_NAME -c $GITHUB_WORKSPACE/channel --override-channels
-
       - name: Collect dependencies
         run: |
           CHANNELS="-c $GITHUB_WORKSPACE/channel ${{ env.CHANNELS }}"
@@ -105,7 +106,6 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-
-
       - name: Install dpctl
         run: |
           CHANNELS="-c $GITHUB_WORKSPACE/channel ${{ env.CHANNELS }}"
@@ -119,8 +119,67 @@ jobs:
           # clinfo -l
           python -m pytest --pyargs $MODULE_NAME
 
-  upload:
-    needs: test
+  test_windows:
+    needs: build
+    runs-on:  ${{ matrix.runner }}
+
+    strategy:
+      matrix:
+        python: [3.8]
+        experimental: [false]
+        runner: [windows-latest]
+    continue-on-error: ${{ matrix.experimental }}
+    env:
+      CHANNELS: -c intel -c defaults --override-channels
+
+    steps:
+      - name: Download artifact
+        uses: actions/download-artifact@v2
+        with:
+          name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
+      - uses: conda-incubator/setup-miniconda@v2
+        with:
+          auto-activate-base: true
+          activate-environment: ""
+      - name: Install conda-build
+        run: conda install conda-build
+      - name: Create conda channel
+        run: |
+          mkdir -p $GITHUB_WORKSPACE/channel/win-64
+          mv ${{ env.PACKAGE_NAME }}-*.tar.bz2 ${{ env.GITHUB_WORKSPACE }}/channel/win-64
+          conda index ${{ env.GITHUB_WORKSPACE }}/channel
+          # Test channel
+          conda search ${{ env.PACKAGE_NAME }} -c ${{ env.GITHUB_WORKSPACE }}/channel --override-channels
+      - name: Collect dependencies
+        run: conda install ${{ env.PACKAGE_NAME }} python=${{ matrix.python }} -c ${{ env.GITHUB_WORKSPACE }}/channel ${{ env.CHANNELS }} --only-deps --dry-run > lockfile
+      - name: Set pkgs_dirs
+        run: |
+          echo "pkgs_dirs: [C:\Users\runneradmin\.conda\pkgs]" >> C:\Users\runneradmin\.condarc
+      - name: Cache conda packages
+        uses: actions/cache@v2
+        env:
+          CACHE_NUMBER: 0  # Increase to reset cache
+        with:
+          path: C:\Users\runneradmin\.conda\pkgs
+          key:
+            ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-${{hashFiles('lockfile') }}
+          restore-keys: |
+            ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-
+            ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-
+      - name: Install opencl_rt
+        run: conda install opencl_rt -c intel --override-channels
+      - name: Install dpctl
+        run: |
+          conda install ${{ env.PACKAGE_NAME }} pytest python=${{ matrix.python }} -c ${{ env.GITHUB_WORKSPACE }}/channel ${{ env.CHANNELS }}
+          # Test installed packages
+          conda list
+      - name: Add library
+        run: echo "OCL_ICD_FILENAMES=C:\Miniconda\Library\lib\intelocl64.dll" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append
+      - name: Run tests
+        run: python -m pytest --pyargs ${{ env.MODULE_NAME }}
+
+  upload_linux:
+    needs: test_linux
     if: ${{ github.ref == 'refs/heads/master' }}
     runs-on: ubuntu-latest
     strategy:
@@ -143,3 +202,29 @@ jobs:
         run: |
           conda install anaconda-client
           anaconda --token $ANACONDA_TOKEN upload --user dppy --label dev ${PACKAGE_NAME}-*.tar.bz2
+
+  upload_windows:
+    needs: test_windows
+    if: ${{ github.ref == 'refs/heads/master' }}
+    runs-on: windows-latest
+    strategy:
+      matrix:
+        python: [3.8]
+    steps:
+      - name: Download artifact
+        uses: actions/download-artifact@v2
+        with:
+          name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
+      - uses: conda-incubator/setup-miniconda@v2
+        with:
+          auto-activate-base: true
+          activate-environment: ""
+      - name: Install anaconda-client
+        run: conda install anaconda-client
+
+      - name: Upload
+        env:
+          ANACONDA_TOKEN: ${{ secrets.ANACONDA_TOKEN }}
+        run: |
+          conda install anaconda-client
+          anaconda --token ${{ env.ANACONDA_TOKEN }} upload --user dppy --label dev ${{ env.PACKAGE_NAME }}-*.tar.bz2
diff --git a/dpctl/__init__.py b/dpctl/__init__.py
@@ -65,6 +65,7 @@
 )
 
 from ._device_selection import select_device_with_aspects
+from ._sycl_timer import SyclTimer
 from ._version import get_versions
 from .enum_types import backend_type, device_type, event_status_type
 
@@ -90,7 +91,7 @@
 ]
 __all__ += [
     "SyclEvent",
-    "SyclEventRaw",
+    "SyclTimer",
 ]
 __all__ += [
     "get_platforms",
diff --git a/dpctl/_sycl_timer.py b/dpctl/_sycl_timer.py
@@ -17,27 +17,65 @@
 
 import timeit
 
-import dpctl
+from . import SyclQueue
 
 
 class SyclTimer:
-    def __init__(self, host_time=timeit.default_timer, time_scale=1):
-        self.timer = host_time
+    """
+    SyclTimer(host_timer=timeit.default_timer, time_scale=1)
+    Python class to measure device time of execution of commands submitted to
+    :class:`dpctl.SyclQueue` as well as the wall-time.
+
+    :Example:
+        .. code-block:: python
+
+            import dpctl
+
+            # Create a default SyclQueue
+            q = dpctl.SyclQueue(property='enable_profiling')
+
+            # create the timer
+            miliseconds_sc = 1e-3
+            timer = dpctl.SyclTimer(time_scale = miliseconds_sc)
+
+            # use the timer
+            with timer(queue=q):
+                code_block
+
+            # retrieve elapsed times in miliseconds
+            sycl_dt, wall_dt = timer.dt
+
+    Remark:
+       The timer synchronizes the queue at the entrance and the
+       exit of the context.
+
+    Args:
+       host_timer (callable): A callable such that host_timer() returns current
+       host time in seconds.
+       time_scale (int, float): Ratio of the unit of time of interest and
+       one second.
+    """
+
+    def __init__(self, host_timer=timeit.default_timer, time_scale=1):
+        self.timer = host_timer
         self.time_scale = time_scale
+        self.queue = None
 
     def __call__(self, queue=None):
-        if isinstance(queue, dpctl.SyclQueue):
+        if isinstance(queue, SyclQueue):
             if queue.has_enable_profiling:
                 self.queue = queue
             else:
                 raise ValueError(
-                    "The queue does not contain the enable_profiling property"
+                    "The given queue was not created with the "
+                    "enable_profiling property"
                 )
         else:
-            raise ValueError(
-                "The passed queue must be <class 'dpctl._sycl_queue.SyclQueue'>"
+            raise TypeError(
+                "The passed queue must have type dpctl.SyclQueue, "
+                "got {}".format(type(queue))
             )
-        return self.__enter__()
+        return self
 
     def __enter__(self):
         self.event_start = self.queue.submit_barrier()
@@ -48,6 +86,7 @@ def __exit__(self, *args):
         self.event_finish = self.queue.submit_barrier()
         self.host_finish = self.timer()
 
+    @property
     def dt(self):
         self.event_start.wait()
         self.event_finish.wait()
diff --git a/dpctl/tensor/_usmarray.pyx b/dpctl/tensor/_usmarray.pyx
@@ -130,12 +130,17 @@ cdef class usm_ndarray:
         cdef Py_ssize_t _offset = offset
         cdef Py_ssize_t ary_min_displacement = 0
         cdef Py_ssize_t ary_max_displacement = 0
+        cdef Py_ssize_t tmp = 0
         cdef char * data_ptr = NULL
 
         self._reset()
         if (not isinstance(shape, (list, tuple))
                 and not hasattr(shape, 'tolist')):
-            raise TypeError("Argument shape must be a list of a tuple.")
+            try:
+                tmp = <Py_ssize_t> shape
+                shape = [shape, ]
+            except Exception:
+                raise TypeError("Argument shape must be a list or a tuple.")
         nd = len(shape)
         typenum = dtype_to_typenum(dtype)
         itemsize = type_bytesize(typenum)
diff --git a/dpctl/tests/test_sycl_event.py b/dpctl/tests/test_sycl_event.py
@@ -157,3 +157,26 @@ def test_profiling_info():
         assert event.profiling_info_end
     else:
         pytest.skip("No OpenCL CPU queues available")
+
+
+def test_sycl_timer():
+    try:
+        q = dpctl.SyclQueue(property="enable_profiling")
+    except dpctl.SyclQueueCreationError:
+        pytest.skip("Queue creation of default device failed")
+    timer = dpctl.SyclTimer()
+    m1 = dpctl_mem.MemoryUSMDevice(256 * 1024, queue=q)
+    m2 = dpctl_mem.MemoryUSMDevice(256 * 1024, queue=q)
+    with timer(q):
+        # device task
+        m1.copy_from_device(m2)
+        # host task
+        [x ** 2 for x in range(1024)]
+    host_dt, device_dt = timer.dt
+    assert host_dt > device_dt
+    q_no_profiling = dpctl.SyclQueue()
+    assert q_no_profiling.has_enable_profiling is False
+    with pytest.raises(ValueError):
+        timer(queue=q_no_profiling)
+    with pytest.raises(TypeError):
+        timer(queue=None)
diff --git a/dpctl/tests/test_sycl_kernel_submit.py b/dpctl/tests/test_sycl_kernel_submit.py
@@ -37,19 +37,20 @@ def test_create_program_from_source(self):
             size_t index = get_global_id(0);                                   \
             c[index] = d*a[index] + b[index];                                  \
         }"
-        q = dpctl.SyclQueue("opencl:gpu")
+        q = dpctl.SyclQueue("opencl:gpu", property="enable_profiling")
         prog = dpctl_prog.create_program_from_source(q, oclSrc)
         axpyKernel = prog.get_sycl_kernel("axpy")
 
-        bufBytes = 1024 * np.dtype("i").itemsize
+        n_elems = 1024 * 512
+        bufBytes = n_elems * np.dtype("i").itemsize
         abuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
         bbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
         cbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
-        a = np.ndarray((1024), buffer=abuf, dtype="i")
-        b = np.ndarray((1024), buffer=bbuf, dtype="i")
-        c = np.ndarray((1024), buffer=cbuf, dtype="i")
-        a[:] = np.arange(1024)
-        b[:] = np.arange(1024, 0, -1)
+        a = np.ndarray((n_elems,), buffer=abuf, dtype="i")
+        b = np.ndarray((n_elems,), buffer=bbuf, dtype="i")
+        c = np.ndarray((n_elems,), buffer=cbuf, dtype="i")
+        a[:] = np.arange(n_elems)
+        b[:] = np.arange(n_elems, 0, -1)
         c[:] = 0
         d = 2
         args = []
@@ -59,10 +60,17 @@ def test_create_program_from_source(self):
         args.append(c.base)
         args.append(ctypes.c_int(d))
 
-        r = [1024]
+        r = [
+            n_elems,
+        ]
 
-        q.submit(axpyKernel, args, r)
-        self.assertTrue(np.allclose(c, a * d + b))
+        timer = dpctl.SyclTimer()
+        with timer(q):
+            q.submit(axpyKernel, args, r)
+            ref_c = a * d + b
+        host_dt, device_dt = timer.dt
+        self.assertTrue(host_dt > device_dt)
+        self.assertTrue(np.allclose(c, ref_c))
 
 
 if __name__ == "__main__":
diff --git a/dpctl/tests/test_usm_ndarray_ctor.py b/dpctl/tests/test_usm_ndarray_ctor.py
diff --git a/examples/python/dppy_kernel.py b/examples/python/dppy_kernel.py

Original file line number	Diff line number	Diff line change
`@@ -65,6 +65,7 @@`
`65`	`65`	`)`
`66`	`66`
`67`	`67`	`from ._device_selection import select_device_with_aspects`
	`68`	`+from ._sycl_timer import SyclTimer`
`68`	`69`	`from ._version import get_versions`
`69`	`70`	`from .enum_types import backend_type, device_type, event_status_type`
`70`	`71`
`@@ -90,7 +91,7 @@`
`90`	`91`	`]`
`91`	`92`	`__all__ += [`
`92`	`93`	`"SyclEvent",`
`93`		`- "SyclEventRaw",`
	`94`	`+ "SyclTimer",`
`94`	`95`	`]`
`95`	`96`	`__all__ += [`
`96`	`97`	`"get_platforms",`