Update third party tests to pass without fp64 aspect (#2284)

Some of third party tests are intended to run towards the latest NumPy,
but previously it was run only in public CI on OpenCL CPU device.
With the new `stable` workflow available in the internal CI, such
scenario is also covered for L0 GPU device.
Thus the tests has to consider dtype check of result based on fp64
aspect of the device (some of GPU does not provide it).

The PR proposes to update the affected tests and to modify dtype check
based on device capabilities where the result is allocating.

dpnp/tests/third_party/cupy/core_tests/test_nep50_examples.py

@@ -2,6 +2,7 @@
 import pytest
 
 import dpnp as cp
+from dpnp.tests.helper import has_support_aspect64
 from dpnp.tests.third_party.cupy import testing
 
 # TODO: remove once all dtype aliases added
@@ -15,8 +16,6 @@
     "uint8(1) + 2",
     "array([1], uint8) + int64(1)",
     "array([1], uint8) + array(1, int64)",
-    "array([1.], float32) + float64(1.)",
-    "array([1.], float32) + array(1., float64)",
     "array([1], uint8) + 1",
     "array([1], uint8) + 200",
     "array([100], uint8) + 200",
@@ -25,7 +24,6 @@
     "uint8(100) + 200",
     "float32(1) + 3e100",
     "array([1.0], float32) + 1e-14 == 1.0",
-    "array([0.1], float32) == float64(0.1)",
     "array(1.0, float32) + 1e-14 == 1.0",
     "array([1.], float32) + 3",
     "array([1.], float32) + int64(3)",
@@ -42,25 +40,34 @@
     "1.0 + array([1, 2, 3], int8)",
     "array([1], float32) + 1j",
 ]
+if has_support_aspect64():
+    examples += [
+        "array([1.], float32) + float64(1.)",
+        "array([1.], float32) + array(1., float64)",
+        "array([0.1], float32) == float64(0.1)",
+    ]
 
 
 @pytest.mark.filterwarnings("ignore::RuntimeWarning")
 @testing.with_requires("numpy>=2.0")
 @pytest.mark.parametrize("example", examples)
-@testing.numpy_cupy_allclose(atol=1e-15, accept_error=OverflowError)
+@testing.numpy_cupy_allclose(
+    atol=1e-15, accept_error=OverflowError, type_check=has_support_aspect64()
+)
 def test_nep50_examples(xp, example):
     dct = {
         "array": xp.array,
         "uint8": xp.uint8,
         "int64": xp.int64,
         "float32": xp.float32,
-        "float64": xp.float64,
         "int16": xp.int16,
         "bool_": xp.bool_,
         "int32": xp.int32,
         "complex64": xp.complex64,
         "int8": xp.int8,
     }
+    if has_support_aspect64():
+        dct["float64"] = xp.float64
 
     if isinstance(example, tuple):
         example, mesg = example

dpnp/tests/third_party/cupy/linalg_tests/test_product.py

@@ -172,7 +172,7 @@ class TestLinalgCrossProduct(unittest.TestCase):
 
     @testing.with_requires("numpy>=2.0")
     @testing.for_all_dtypes_combination(["dtype_a", "dtype_b"])
-    @testing.numpy_cupy_allclose()
+    @testing.numpy_cupy_allclose(type_check=has_support_aspect64())
     def test_cross(self, xp, dtype_a, dtype_b):
         if dtype_a == dtype_b == numpy.bool_:
             # cross does not support bool-bool inputs.

dpnp/tests/third_party/cupy/math_tests/test_misc.py

@@ -198,7 +198,7 @@ def test_fabs(self, xp, dtype):
 
     @testing.with_requires("numpy>=2.0")
     @testing.for_all_dtypes(no_complex=True)
-    @testing.numpy_cupy_allclose(atol=1e-5)
+    @testing.numpy_cupy_allclose(atol=1e-5, type_check=has_support_aspect64())
     def test_fabs_negative(self, xp, dtype):
         if numpy.issubdtype(dtype, numpy.unsignedinteger):
             pytest.skip("trying to set negative value to unsigned integer")

dpnp/tests/third_party/cupy/math_tests/test_sumprod.py

@@ -1162,51 +1162,75 @@ def test_ediff1d_ed2(self, xp, dtype):
 class TestTrapezoid:
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_1dim(self, xp, dtype):
         a = testing.shaped_arange((5,), xp, dtype)
         return xp.trapezoid(a)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_1dim_with_x(self, xp, dtype):
         a = testing.shaped_arange((5,), xp, dtype)
         x = testing.shaped_arange((5,), xp, dtype)
         return xp.trapezoid(a, x=x)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_1dim_with_dx(self, xp, dtype):
         a = testing.shaped_arange((5,), xp, dtype)
         return xp.trapezoid(a, dx=0.1)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_2dim_without_axis(self, xp, dtype):
         a = testing.shaped_arange((4, 5), xp, dtype)
         return xp.trapezoid(a)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_2dim_with_axis(self, xp, dtype):
         a = testing.shaped_arange((4, 5), xp, dtype)
         return xp.trapezoid(a, axis=-2)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_2dim_with_x_and_axis(self, xp, dtype):
         a = testing.shaped_arange((4, 5), xp, dtype)
         x = testing.shaped_arange((5,), xp, dtype)
         return xp.trapezoid(a, x=x, axis=1)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_2dim_with_dx_and_axis(self, xp, dtype):
         a = testing.shaped_arange((4, 5), xp, dtype)
         return xp.trapezoid(a, dx=0.1, axis=1)
 
     @testing.for_all_dtypes()
-    @testing.numpy_cupy_allclose(rtol={numpy.float16: 1e-1, "default": 1e-7})
+    @testing.numpy_cupy_allclose(
+        rtol={numpy.float16: 1e-1, "default": 1e-7},
+        type_check=has_support_aspect64(),
+    )
     def test_trapz_1dim_with_x_and_dx(self, xp, dtype):
         a = testing.shaped_arange((5,), xp, dtype)
         x = testing.shaped_arange((5,), xp, dtype)

dpnp/tests/third_party/cupy/testing/_loops.py

@@ -583,12 +583,12 @@ def numpy_cupy_allclose(
     # When `type_check` is `False`, cupy result and numpy result may have
     # different dtypes so we can not determine the dtype to use from the
     # tolerance associations.
-    if not type_check:
-        if isinstance(rtol, dict) or isinstance(atol, dict):
-            raise TypeError(
-                "When `type_check` is `False`, `rtol` and `atol` "
-                "must be supplied as float."
-            )
+    # if not type_check:
+    #     if isinstance(rtol, dict) or isinstance(atol, dict):
+    #         raise TypeError(
+    #             "When `type_check` is `False`, `rtol` and `atol` "
+    #             "must be supplied as float."
+    #         )
 
     def check_func(c, n):
         rtol1, atol1 = _resolve_tolerance(type_check, c, rtol, atol)