Fix test_matmul_simple to avoid use of out-of-bounds Python scalars

Change the test so that input matrices that get multiplied only have
blocks of ones no larger than the max integer for the type, rest is
populated with zeros. This change applies to integral types only.

dpctl/tests/test_usm_ndarray_linalg.py

@@ -89,12 +89,20 @@ def test_matmul_simple(dtype):
     skip_if_dtype_not_supported(dtype, q)
 
     n, m = 235, 17
-    m1 = dpt.ones((m, n), dtype=dtype)
-    m2 = dpt.ones((n, m), dtype=dtype)
+    m1 = dpt.zeros((m, n), dtype=dtype)
+    m2 = dpt.zeros((n, m), dtype=dtype)
+
+    dt = m1.dtype
+    if dt.kind in "ui":
+        n1 = min(n, dpt.iinfo(dt).max)
+    else:
+        n1 = n
+    m1[:, :n1] = dpt.ones((m, n1), dtype=dt)
+    m2[:n1, :] = dpt.ones((n1, m), dtype=dt)
 
     for k in [1, 2, 3, 4, 7, 8, 9, 15, 16, 17]:
         r = dpt.matmul(m1[:k, :], m2[:, :k])
-        assert dpt.all(r == dpt.full((k, k), n, dtype=dtype))
+        assert dpt.all(r == dpt.full((k, k), fill_value=n1, dtype=dt))
 
 
 @pytest.mark.parametrize("dtype", _numeric_types)