Build a list of bin counts for histogram tests based on available memory for a device

conda-recipe/run_test.sh

@@ -37,4 +37,8 @@ set -e
 
 $PYTHON -c "import dpnp; print(dpnp.__version__)"
 $PYTHON -m dpctl -f
+
+timeout 10m gdb --batch -ex run -ex 'info sharedlibrary' -ex 'set print elements 1000' -ex thread apply all bt --args "$PYTHON" -m pytest -ra --disable-warnings --pyargs dpnp.tests.test_histogram || true
+$PYTHON -m pytest -sv --pyargs dpnp.tests.test_histogram
+
 $PYTHON -m pytest -ra --pyargs dpnp

dpnp/backend/extensions/statistics/bincount.cpp

@@ -201,7 +201,7 @@ std::tuple<sycl::event, sycl::event> Bincount::call(
             dpctl::utils::keep_args_alive(exec_q, {sample, histogram}, {ev});
     }
 
-    return {args_ev, ev};
+    return std::make_pair(args_ev, ev);
 }
 
 std::unique_ptr<Bincount> bincount;

dpnp/backend/extensions/statistics/histogram.cpp

@@ -259,7 +259,7 @@ std::tuple<sycl::event, sycl::event>
             exec_q, {sample, bins, histogram}, {ev});
     }
 
-    return {args_ev, ev};
+    return std::make_pair(args_ev, ev);
 }
 
 std::unique_ptr<Histogram> hist;

dpnp/backend/extensions/statistics/histogramdd.cpp

@@ -326,7 +326,7 @@ std::tuple<sycl::event, sycl::event> Histogramdd::call(
             exec_q, {sample, bins_edges, bins_edges_count, histogram}, {ev});
     }
 
-    return {args_ev, ev};
+    return std::make_pair(args_ev, ev);
 }
 
 std::unique_ptr<Histogramdd> histdd;

dpnp/backend/extensions/statistics/sliding_dot_product1d.cpp

@@ -135,10 +135,10 @@ std::tuple<sycl::event, sycl::event>
     auto ev = corr_func(exec_q, a.get_data(), v.get_data(), out.get_data(),
                         a.get_shape(0), v.get_shape(0), l_pad, r_pad, depends);
 
-    sycl::event args_ev;
-    args_ev = dpctl::utils::keep_args_alive(exec_q, {a, v, out}, {ev});
+    sycl::event args_ev =
+        dpctl::utils::keep_args_alive(exec_q, {a, v, out}, {ev});
 
-    return {args_ev, ev};
+    return std::make_pair(args_ev, ev);
 }
 
 std::unique_ptr<SlidingDotProduct1d> sdp;

dpnp/dpnp_algo/dpnp_fill.py

@@ -58,10 +58,10 @@ def dpnp_fill(arr, val):
         a_val = dpt.broadcast_to(a_val, arr.shape)
         _manager = dpu.SequentialOrderManager[exec_q]
         dep_evs = _manager.submitted_events
-        h_ev, c_ev = _copy_usm_ndarray_into_usm_ndarray(
+        ht_ev, c_ev = _copy_usm_ndarray_into_usm_ndarray(
             src=a_val, dst=arr, sycl_queue=exec_q, depends=dep_evs
         )
-        _manager.add_event_pair(h_ev, c_ev)
+        _manager.add_event_pair(ht_ev, c_ev)
         return
     elif not isinstance(val, (Number, dpnp.bool)):
         raise TypeError(
@@ -74,8 +74,8 @@ def dpnp_fill(arr, val):
 
     # can leverage efficient memset when val is 0
     if arr.flags["FORC"] and val == 0:
-        h_ev, zeros_ev = _zeros_usm_ndarray(arr, exec_q, depends=dep_evs)
-        _manager.add_event_pair(h_ev, zeros_ev)
+        ht_ev, zeros_ev = _zeros_usm_ndarray(arr, exec_q, depends=dep_evs)
+        _manager.add_event_pair(ht_ev, zeros_ev)
     else:
-        h_ev, fill_ev = _full_usm_ndarray(val, arr, exec_q, depends=dep_evs)
-        _manager.add_event_pair(h_ev, fill_ev)
+        ht_ev, fill_ev = _full_usm_ndarray(val, arr, exec_q, depends=dep_evs)
+        _manager.add_event_pair(ht_ev, fill_ev)

dpnp/dpnp_iface_histograms.py

@@ -284,8 +284,6 @@ def _bincount_run_native(
         size, dtype=n_dtype, usm_type=n_usm_type, sycl_queue=queue
     )
 
-    _manager = dpu.SequentialOrderManager[queue]
-
     x_usm = dpnp.get_usm_ndarray(x_casted)
     weights_usm = (
         dpnp.get_usm_ndarray(weights_casted)
@@ -294,7 +292,9 @@ def _bincount_run_native(
     )
     n_usm = dpnp.get_usm_ndarray(n_casted)
 
-    mem_ev, bc_ev = statistics_ext.bincount(
+    _manager = dpu.SequentialOrderManager[queue]
+
+    ht_ev, bc_ev = statistics_ext.bincount(
         x_usm,
         min_v.item(),
         max_v.item(),
@@ -303,7 +303,7 @@ def _bincount_run_native(
         depends=_manager.submitted_events,
     )
 
-    _manager.add_event_pair(mem_ev, bc_ev)
+    _manager.add_event_pair(ht_ev, bc_ev)
 
     return n_casted
 
@@ -647,8 +647,6 @@ def histogram(a, bins=10, range=None, density=None, weights=None):
         usm_type=n_usm_type,
     )
 
-    _manager = dpu.SequentialOrderManager[queue]
-
     a_usm = dpnp.get_usm_ndarray(a_casted)
     bins_usm = dpnp.get_usm_ndarray(bin_edges_casted)
     weights_usm = (
@@ -658,14 +656,16 @@ def histogram(a, bins=10, range=None, density=None, weights=None):
     )
     n_usm = dpnp.get_usm_ndarray(n_casted)
 
-    mem_ev, ht_ev = statistics_ext.histogram(
+    _manager = dpu.SequentialOrderManager[queue]
+
+    ht_ev, hist_ev = statistics_ext.histogram(
         a_usm,
         bins_usm,
         weights_usm,
         n_usm,
         depends=_manager.submitted_events,
     )
-    _manager.add_event_pair(mem_ev, ht_ev)
+    _manager.add_event_pair(ht_ev, hist_ev)
 
     n = dpnp.asarray(n_casted, dtype=ntype, usm_type=usm_type)
 
@@ -1039,7 +1039,7 @@ def _histdd_run_native(
 
     _manager = dpu.SequentialOrderManager[queue]
 
-    mem_ev, hdd_ev = statistics_ext.histogramdd(
+    ht_ev, histdd_ev = statistics_ext.histogramdd(
         sample_usm,
         edges_usm,
         edges_count_usm,
@@ -1048,7 +1048,7 @@ def _histdd_run_native(
         depends=_manager.submitted_events,
     )
 
-    _manager.add_event_pair(mem_ev, hdd_ev)
+    _manager.add_event_pair(ht_ev, histdd_ev)
 
     return n
 

dpnp/dpnp_iface_indexing.py

@@ -177,8 +177,8 @@ def _choose_run(inds, chcs, q, usm_type, out=None, mode=0):
     _manager = dpu.SequentialOrderManager[q]
     dep_evs = _manager.submitted_events
 
-    h_ev, choose_ev = indexing_ext._choose(inds, chcs, out, mode, q, dep_evs)
-    _manager.add_event_pair(h_ev, choose_ev)
+    ht_ev, choose_ev = indexing_ext._choose(inds, chcs, out, mode, q, dep_evs)
+    _manager.add_event_pair(ht_ev, choose_ev)
 
     return out
 
@@ -335,7 +335,7 @@ def _take_index(x, inds, axis, q, usm_type, out=None, mode=0):
     _manager = dpu.SequentialOrderManager[q]
     dep_evs = _manager.submitted_events
 
-    h_ev, take_ev = ti._take(
+    ht_ev, take_ev = ti._take(
         src=x,
         ind=(inds,),
         dst=out,
@@ -344,7 +344,7 @@ def _take_index(x, inds, axis, q, usm_type, out=None, mode=0):
         sycl_queue=q,
         depends=dep_evs,
     )
-    _manager.add_event_pair(h_ev, take_ev)
+    _manager.add_event_pair(ht_ev, take_ev)
 
     return out
 

dpnp/dpnp_iface_logic.py

@@ -137,7 +137,7 @@ def _isclose_scalar_tol(a, b, rtol, atol, equal_nan):
     )
 
     _manager = dpu.SequentialOrderManager[exec_q]
-    mem_ev, ht_ev = ufi._isclose_scalar(
+    ht_ev, isclose_ev = ufi._isclose_scalar(
         a.get_array(),
         b.get_array(),
         rtol,
@@ -147,7 +147,7 @@ def _isclose_scalar_tol(a, b, rtol, atol, equal_nan):
         exec_q,
         depends=_manager.submitted_events,
     )
-    _manager.add_event_pair(mem_ev, ht_ev)
+    _manager.add_event_pair(ht_ev, isclose_ev)
 
     return output
 

dpnp/dpnp_iface_mathematical.py

@@ -2945,7 +2945,7 @@ def interp(x, xp, fp, left=None, right=None, period=None):
     right_usm = right.get_array() if right is not None else None
 
     _manager = dpu.SequentialOrderManager[exec_q]
-    mem_ev, ht_ev = ufi._interpolate(
+    ht_ev, inter_ev = ufi._interpolate(
         x.get_array(),
         idx.get_array(),
         xp.get_array(),
@@ -2956,7 +2956,7 @@ def interp(x, xp, fp, left=None, right=None, period=None):
         exec_q,
         depends=_manager.submitted_events,
     )
-    _manager.add_event_pair(mem_ev, ht_ev)
+    _manager.add_event_pair(ht_ev, inter_ev)
 
     return output
 
@@ -3517,11 +3517,11 @@ def nan_to_num(x, copy=True, nan=0.0, posinf=None, neginf=None):
     q = x.sycl_queue
     _manager = dpu.SequentialOrderManager[q]
 
-    h_ev, comp_ev = ufi._nan_to_num(
+    ht_ev, comp_ev = ufi._nan_to_num(
         x_ary, nan, max_f, min_f, out_ary, q, depends=_manager.submitted_events
     )
 
-    _manager.add_event_pair(h_ev, comp_ev)
+    _manager.add_event_pair(ht_ev, comp_ev)
 
     return dpnp.get_result_array(out)
 

dpnp/dpnp_iface_statistics.py

@@ -699,15 +699,15 @@ def _run_native_sliding_dot_product1d(a, v, l_pad, r_pad, rdtype):
 
     _manager = dpu.SequentialOrderManager[queue]
 
-    mem_ev, corr_ev = statistics_ext.sliding_dot_product1d(
+    ht_ev, corr_ev = statistics_ext.sliding_dot_product1d(
         a_usm,
         v_usm,
         out_usm,
         l_pad,
         r_pad,
         depends=_manager.submitted_events,
     )
-    _manager.add_event_pair(mem_ev, corr_ev)
+    _manager.add_event_pair(ht_ev, corr_ev)
 
     return out
 

dpnp/tests/conftest.py

@@ -44,7 +44,7 @@
 
 import dpnp
 
-from .helper import get_dev_id
+from .helper import get_dev_id, get_dev_info
 
 skip_mark = pytest.mark.skip(reason="Skipping test.")
 
@@ -143,7 +143,9 @@ def pytest_collection_modifyitems(config, items):
     print(
         f"DPNP Test scope includes all integer dtypes: {bool(dtype_config.all_int_types)}"
     )
-    print(f"DPNP current device ID: 0x{get_dev_id(dev):04X}")
+    print(
+        f"DPNP current device ID: 0x{get_dev_id(dev):04X}, info: {get_dev_info(dev)}"
+    )
     print(f"DPNP current device is CPU: {is_cpu}")
     print(f"DPNP current device is GPU: {is_gpu}")
     print(f"DPNP current device supports fp64: {support_fp64}")

dpnp/tests/helper.py

@@ -309,9 +309,18 @@ def get_dev_id(device=None):
     Obtain Intel Device ID for a device (the default device if not provided).
     """
 
+    return get_dev_info(device).get("device_id", 0)
+
+
+def get_dev_info(device=None):
+    """
+    Obtain a dictionary with the info for a device (the default device if not
+    provided).
+
+    """
+
     dev = dpctl.select_default_device() if device is None else device
-    dev_info = dpctl.utils.intel_device_info(dev)
-    return dev_info.get("device_id", 0)
+    return dpctl.utils.intel_device_info(dev)
 
 
 def get_float_dtypes(no_float16=True, device=None):
@@ -531,3 +540,14 @@ def requires_intel_mkl_version(version):  # pragma: no cover
 
     build_deps = numpy.show_config(mode="dicts")["Build Dependencies"]
     return build_deps["blas"]["version"] >= version
+
+
+def requires_memory(no_of_gbs, device=None):
+    """
+    Check if the required number of GBs in memory of a device (the default one
+    if not provided) is available.
+
+    """
+
+    free_mem = get_dev_info(device).get("free_memory", 0)
+    return free_mem >= no_of_gbs * (1024**3)

dpnp/tests/test_histogram.py

@@ -22,8 +22,14 @@
     get_integer_float_dtypes,
     has_support_aspect64,
     numpy_version,
+    requires_memory,
 )
 
+# Build a list of bin sizes to check the histogram
+_bin_counts = [10, 10**2, 10**3, 10**4, 10**5]
+if requires_memory(16):
+    _bin_counts += [10**6]
+
 
 class TestDigitize:
     @pytest.mark.parametrize("dtype", get_integer_float_dtypes())
@@ -486,10 +492,7 @@ def test_weights_another_sycl_queue(self):
         with assert_raises(ValueError):
             dpnp.histogram(v, weights=w)
 
-    @pytest.mark.parametrize(
-        "bins_count",
-        [10, 10**2, 10**3, 10**4, 10**5, 10**6],
-    )
+    @pytest.mark.parametrize("bins_count", _bin_counts)
     def test_different_bins_amount(self, bins_count):
         v = numpy.linspace(0, bins_count, bins_count, dtype=numpy.float32)
         iv = dpnp.array(v)
@@ -584,10 +587,7 @@ def test_weights_unsupported_dtype(self, xp, dt):
         w = xp.arange(5, dtype=dt)
         assert_raises((TypeError, ValueError), xp.bincount, v, weights=w)
 
-    @pytest.mark.parametrize(
-        "bins_count",
-        [10, 10**2, 10**3, 10**4, 10**5, 10**6],
-    )
+    @pytest.mark.parametrize("bins_count", _bin_counts)
     def test_different_bins_amount(self, bins_count):
         v = numpy.arange(0, bins_count, dtype=int)
         iv = dpnp.array(v)
@@ -863,10 +863,7 @@ def test_weights_another_sycl_queue(self):
         with assert_raises(ValueError):
             dpnp.histogramdd(v, weights=w)
 
-    @pytest.mark.parametrize(
-        "bins_count",
-        [10, 10**2, 10**3, 10**4, 10**5, 10**6],
-    )
+    @pytest.mark.parametrize("bins_count", _bin_counts)
     def test_different_bins_amount(self, bins_count):
         v = numpy.linspace(0, bins_count, bins_count, dtype=numpy.float32)
         iv = dpnp.array(v)