numba-cuda 0.16.0__py3-none-any.whl → 0.18.0__py3-none-any.whl

numba_cuda/numba/cuda/tests/cudapy/test_intrinsics.py

@@ -3,7 +3,6 @@ import numpy as np
 import operator
 import re
 from numba import cuda, int64
-from numba.cuda import compile_ptx
 from numba.core.errors import TypingError
 from numba.core.types import f2
 from numba.cuda.testing import (
@@ -11,6 +10,7 @@ from numba.cuda.testing import (
     CUDATestCase,
     skip_on_cudasim,
     skip_unless_cc_53,
+    skip_if_nvjitlink_missing,
 )
 
 
@@ -174,27 +174,27 @@ def hlt_func_2(x, y):
     return cuda.fp16.hlt(x, y)
 
 
-def test_multiple_hcmp_1(r, a, b, c):
+def multiple_hcmp_1(r, a, b, c):
     # float16 predicates used in two separate functions
     r[0] = hlt_func_1(a, b) and hlt_func_2(b, c)
 
 
-def test_multiple_hcmp_2(r, a, b, c):
+def multiple_hcmp_2(r, a, b, c):
     # The same float16 predicate used in the caller and callee
     r[0] = hlt_func_1(a, b) and cuda.fp16.hlt(b, c)
 
 
-def test_multiple_hcmp_3(r, a, b, c):
+def multiple_hcmp_3(r, a, b, c):
     # Different float16 predicates used in the caller and callee
     r[0] = hlt_func_1(a, b) and cuda.fp16.hge(c, b)
 
 
-def test_multiple_hcmp_4(r, a, b, c):
+def multiple_hcmp_4(r, a, b, c):
     # The same float16 predicates used twice in a function
     r[0] = cuda.fp16.hlt(a, b) and cuda.fp16.hlt(b, c)
 
 
-def test_multiple_hcmp_5(r, a, b, c):
+def multiple_hcmp_5(r, a, b, c):
     # Different float16 predicates used in a function
     r[0] = cuda.fp16.hlt(a, b) and cuda.fp16.hge(c, b)
 
@@ -561,13 +561,13 @@ class TestCudaIntrinsic(CUDATestCase):
 
     def test_popc_u1(self):
         compiled = cuda.jit("void(int32[:], uint8)")(simple_popc)
-        ary = np.zeros(1, dtype=np.int8)
+        ary = np.zeros(1, dtype=np.int32)
         compiled[1, 1](ary, np.uint8(0xFF))
         self.assertEqual(ary[0], 8)
 
     def test_popc_u2(self):
         compiled = cuda.jit("void(int32[:], uint16)")(simple_popc)
-        ary = np.zeros(1, dtype=np.int16)
+        ary = np.zeros(1, dtype=np.int32)
         compiled[1, 1](ary, np.uint16(0xFFFF))
         self.assertEqual(ary[0], 16)
 
@@ -585,13 +585,13 @@ class TestCudaIntrinsic(CUDATestCase):
 
     def test_bit_count_u1(self):
         compiled = cuda.jit("void(int32[:], uint8)")(simple_bit_count)
-        ary = np.zeros(1, dtype=np.int8)
+        ary = np.zeros(1, dtype=np.int32)
         compiled[1, 1](ary, np.uint8(0xFF))
         self.assertEqual(ary[0], 8)
 
     def test_bit_count_u2(self):
         compiled = cuda.jit("void(int32[:], uint16)")(simple_bit_count)
-        ary = np.zeros(1, dtype=np.int16)
+        ary = np.zeros(1, dtype=np.int32)
         compiled[1, 1](ary, np.uint16(0xFFFF))
         self.assertEqual(ary[0], 16)
 
@@ -639,9 +639,11 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_hadd_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2, f2)", lto=True)(simple_hadd_scalar)
         args = (f2[:], f2, f2)
-        ptx, _ = compile_ptx(simple_hadd_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("add.f16", ptx)
 
     @skip_unless_cc_53
@@ -666,9 +668,13 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_hfma_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2, f2, f2)", lto=True)(
+            simple_hfma_scalar
+        )
         args = (f2[:], f2, f2, f2)
-        ptx, _ = compile_ptx(simple_hfma_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("fma.rn.f16", ptx)
 
     @skip_unless_cc_53
@@ -691,14 +697,16 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_hsub_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2, f2)", lto=True)(simple_hsub_scalar)
         args = (f2[:], f2, f2)
-        ptx, _ = compile_ptx(simple_hsub_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("sub.f16", ptx)
 
     @skip_unless_cc_53
     def test_hmul(self):
-        compiled = cuda.jit()(simple_hmul)
+        compiled = cuda.jit(simple_hmul)
         ary = np.zeros(1, dtype=np.float16)
         arg1 = np.array([3.0], dtype=np.float16)
         arg2 = np.array([4.0], dtype=np.float16)
@@ -716,9 +724,11 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_hmul_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2, f2)", lto=True)(simple_hmul_scalar)
         args = (f2[:], f2, f2)
-        ptx, _ = compile_ptx(simple_hmul_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("mul.f16", ptx)
 
     @skip_unless_cc_53
@@ -761,14 +771,16 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_hneg_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2)", lto=True)(simple_hneg_scalar)
         args = (f2[:], f2)
-        ptx, _ = compile_ptx(simple_hneg_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("neg.f16", ptx)
 
     @skip_unless_cc_53
     def test_habs(self):
-        compiled = cuda.jit()(simple_habs)
+        compiled = cuda.jit(simple_habs)
         ary = np.zeros(1, dtype=np.float16)
         arg1 = np.array([-3.0], dtype=np.float16)
         compiled[1, 1](ary, arg1)
@@ -784,9 +796,11 @@ class TestCudaIntrinsic(CUDATestCase):
         np.testing.assert_allclose(ary[0], ref)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_habs_ptx(self):
+        compiled = cuda.jit("void(f2[:], f2)", lto=True)(simple_habs_scalar)
         args = (f2[:], f2)
-        ptx, _ = compile_ptx(simple_habs_scalar, args)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("abs.f16", ptx)
 
     @skip_unless_cc_53
@@ -908,11 +922,11 @@ class TestCudaIntrinsic(CUDATestCase):
     @skip_unless_cc_53
     def test_multiple_float16_comparisons(self):
         functions = (
-            test_multiple_hcmp_1,
-            test_multiple_hcmp_2,
-            test_multiple_hcmp_3,
-            test_multiple_hcmp_4,
-            test_multiple_hcmp_5,
+            multiple_hcmp_1,
+            multiple_hcmp_2,
+            multiple_hcmp_3,
+            multiple_hcmp_4,
+            multiple_hcmp_5,
         )
         for fn in functions:
             with self.subTest(fn=fn):

numba_cuda/numba/cuda/tests/cudapy/test_operator.py

@@ -4,10 +4,12 @@ from numba.cuda.testing import (
     CUDATestCase,
     skip_unless_cc_53,
     skip_on_cudasim,
+    skip_if_nvjitlink_missing,
 )
 from numba import cuda
+from numba.core import types
 from numba.core.types import f2, b1
-from numba.cuda import compile_ptx
+from numba.core.typing import signature
 import operator
 import itertools
 from numba.np.numpy_support import from_dtype
@@ -87,27 +89,27 @@ def hlt_func_2(x, y):
     return x < y
 
 
-def test_multiple_hcmp_1(r, a, b, c):
+def multiple_hcmp_1(r, a, b, c):
     # float16 predicates used in two separate functions
     r[0] = hlt_func_1(a, b) and hlt_func_2(b, c)
 
 
-def test_multiple_hcmp_2(r, a, b, c):
+def multiple_hcmp_2(r, a, b, c):
     # The same float16 predicate used in the caller and callee
     r[0] = hlt_func_1(a, b) and b < c
 
 
-def test_multiple_hcmp_3(r, a, b, c):
+def multiple_hcmp_3(r, a, b, c):
     # Different float16 predicates used in the caller and callee
     r[0] = hlt_func_1(a, b) and c >= b
 
 
-def test_multiple_hcmp_4(r, a, b, c):
+def multiple_hcmp_4(r, a, b, c):
     # The same float16 predicates used twice in a function
     r[0] = a < b and b < c
 
 
-def test_multiple_hcmp_5(r, a, b, c):
+def multiple_hcmp_5(r, a, b, c):
     # Different float16 predicates used in a function
     r[0] = a < b and c >= b
 
@@ -172,16 +174,19 @@ class TestOperatorModule(CUDATestCase):
                 np.testing.assert_allclose(got, expected)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_binary_ptx(self):
         functions = (simple_fp16add, simple_fp16sub, simple_fp16mul)
         instrs = ("add.f16", "sub.f16", "mul.f16")
         args = (f2[:], f2, f2)
         for fn, instr in zip(functions, instrs):
             with self.subTest(instr=instr):
-                ptx, _ = compile_ptx(fn, args)
+                compiled = cuda.jit("void(f2[:], f2, f2)", lto=True)(fn)
+                ptx = compiled.inspect_lto_ptx(args)
                 self.assertIn(instr, ptx)
 
     @skip_unless_cc_53
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_mixed_fp16_binary_arithmetic(self):
         functions = (
             simple_fp16add,
@@ -193,7 +198,7 @@ class TestOperatorModule(CUDATestCase):
         types = (np.int8, np.int16, np.int32, np.int64, np.float32, np.float64)
         for (fn, op), ty in itertools.product(zip(functions, ops), types):
             with self.subTest(op=op, ty=ty):
-                kernel = cuda.jit(fn)
+                kernel = cuda.jit(fn, lto=True)
 
                 arg1 = np.random.random(1).astype(np.float16)
                 arg2 = (np.random.random(1) * 100).astype(ty)
@@ -205,6 +210,7 @@ class TestOperatorModule(CUDATestCase):
                 np.testing.assert_allclose(got, expected)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_inplace_binary_ptx(self):
         functions = (simple_fp16_iadd, simple_fp16_isub, simple_fp16_imul)
         instrs = ("add.f16", "sub.f16", "mul.f16")
@@ -212,7 +218,8 @@ class TestOperatorModule(CUDATestCase):
 
         for fn, instr in zip(functions, instrs):
             with self.subTest(instr=instr):
-                ptx, _ = compile_ptx(fn, args)
+                compiled = cuda.jit("void(f2[:], f2)", lto=True)(fn)
+                ptx = compiled.inspect_lto_ptx(args)
                 self.assertIn(instr, ptx)
 
     @skip_unless_cc_53
@@ -253,16 +260,19 @@ class TestOperatorModule(CUDATestCase):
                 np.testing.assert_allclose(got, expected)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_neg_ptx(self):
         args = (f2[:], f2)
-        ptx, _ = compile_ptx(simple_fp16neg, args)
+        compiled = cuda.jit("void(f2[:], f2)", lto=True)(simple_fp16neg)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("neg.f16", ptx)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_abs_ptx(self):
         args = (f2[:], f2)
-        ptx, _ = compile_ptx(simple_fp16abs, args)
-
+        compiled = cuda.jit("void(f2[:], f2)", lto=True)(simple_fp16abs)
+        ptx = compiled.inspect_lto_ptx(args)
         self.assertIn("abs.f16", ptx)
 
     @skip_unless_cc_53
@@ -331,11 +341,11 @@ class TestOperatorModule(CUDATestCase):
     @skip_unless_cc_53
     def test_multiple_float16_comparisons(self):
         functions = (
-            test_multiple_hcmp_1,
-            test_multiple_hcmp_2,
-            test_multiple_hcmp_3,
-            test_multiple_hcmp_4,
-            test_multiple_hcmp_5,
+            multiple_hcmp_1,
+            multiple_hcmp_2,
+            multiple_hcmp_3,
+            multiple_hcmp_4,
+            multiple_hcmp_5,
         )
         for fn in functions:
             with self.subTest(fn=fn):
@@ -350,11 +360,11 @@ class TestOperatorModule(CUDATestCase):
     @skip_unless_cc_53
     def test_multiple_float16_comparisons_false(self):
         functions = (
-            test_multiple_hcmp_1,
-            test_multiple_hcmp_2,
-            test_multiple_hcmp_3,
-            test_multiple_hcmp_4,
-            test_multiple_hcmp_5,
+            multiple_hcmp_1,
+            multiple_hcmp_2,
+            multiple_hcmp_3,
+            multiple_hcmp_4,
+            multiple_hcmp_5,
         )
         for fn in functions:
             with self.subTest(fn=fn):
@@ -367,6 +377,7 @@ class TestOperatorModule(CUDATestCase):
                 self.assertFalse(ary[0])
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_comparison_ptx(self):
         functions = (
             simple_fp16_gt,
@@ -390,16 +401,18 @@ class TestOperatorModule(CUDATestCase):
             "setp.lt.f16",
             "setp.le.f16",
             "setp.eq.f16",
-            "setp.ne.f16",
+            "setp.neu.f16",
         )
         args = (b1[:], f2, f2)
 
         for fn, op, s in zip(functions, ops, opstring):
             with self.subTest(op=op):
-                ptx, _ = compile_ptx(fn, args)
+                compiled = cuda.jit("void(b1[:], f2, f2)", lto=True)(fn)
+                ptx = compiled.inspect_lto_ptx(args)
                 self.assertIn(s, ptx)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_fp16_int8_comparison_ptx(self):
         # Test that int8 can be safely converted to fp16
         # in a comparison
@@ -426,15 +439,17 @@ class TestOperatorModule(CUDATestCase):
             operator.lt: "setp.lt.f16",
             operator.le: "setp.le.f16",
             operator.eq: "setp.eq.f16",
-            operator.ne: "setp.ne.f16",
+            operator.ne: "setp.neu.f16",
         }
         for fn, op in zip(functions, ops):
             with self.subTest(op=op):
                 args = (b1[:], f2, from_dtype(np.int8))
-                ptx, _ = compile_ptx(fn, args)
+                compiled = cuda.jit(signature(types.void, *args), lto=True)(fn)
+                ptx = compiled.inspect_lto_ptx(args)
                 self.assertIn(opstring[op], ptx)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
+    @skip_if_nvjitlink_missing("Numbast generated bindings")
     def test_mixed_fp16_comparison_promotion_ptx(self):
         functions = (
             simple_fp16_gt,
@@ -475,7 +490,8 @@ class TestOperatorModule(CUDATestCase):
             with self.subTest(op=op, ty=ty):
                 arg2_ty = np.result_type(np.float16, ty)
                 args = (b1[:], f2, from_dtype(arg2_ty))
-                ptx, _ = compile_ptx(fn, args)
+                compiled = cuda.jit(signature(types.void, *args), lto=True)(fn)
+                ptx = compiled.inspect_lto_ptx(args)
 
                 ops = opstring[op] + opsuffix[arg2_ty]
                 self.assertIn(ops, ptx)

numba_cuda/numba/cuda/tests/cudapy/test_stream_api.py

@@ -4,7 +4,7 @@ from numba.cuda.testing import (
     unittest,
     CUDATestCase,
 )
-from numba import config, cuda
+from numba import cuda
 
 # Basic tests that stream APIs execute on the hardware and in the simulator.
 #
@@ -38,10 +38,7 @@ class TestStreamAPI(CUDATestCase):
         # We don't test synchronization on the stream because it's not a real
         # stream - we used a dummy pointer for testing the API, so we just
         # ensure that the stream handle matches the external stream pointer.
-        if config.CUDA_USE_NVIDIA_BINDING:
-            value = int(s.handle)
-        else:
-            value = s.handle.value
+        value = s.handle.value
         self.assertEqual(ptr, value)
 
     @skip_unless_cudasim("External streams are usable with hardware")

numba_cuda/numba/cuda/tests/cudapy/test_ufuncs.py

@@ -17,16 +17,16 @@ def _make_ufunc_usecase(ufunc):
     return fn
 
 
-# This test would also be a CUDATestCase, but to avoid a confusing and
-# potentially dangerous inheritance diamond with setUp methods that modify
-# global state, we implement the necessary parts of CUDATestCase within this
-# class instead. These are:
+# This class provides common functionality for UFunc tests. The UFunc tests
+# are quite long-running in comparison to other tests, so we break the tests up
+# into multiple test classes for distribution across workers.
 #
-# - Disable parallel testing with _numba_parallel_test_.
-# - Disabling CUDA performance warnings for the duration of tests.
-class TestUFuncs(BasicUFuncTest, TestCase):
-    _numba_parallel_test_ = False
-
+# This class would also be a CUDATestCase, but to avoid a confusing and
+# potentially dangerous inheritance diamond with setUp methods that modify
+# global state, we implement the necessary part of CUDATestCase within this
+# class instead. This disables CUDA performance warnings for the duration of
+# tests.
+class CUDAUFuncTestBase(BasicUFuncTest, TestCase):
     def setUp(self):
         BasicUFuncTest.setUp(self)
 
@@ -146,6 +146,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     ############################################################################
     # Trigonometric Functions
 
+
+class TestBasicTrigUFuncs(CUDAUFuncTestBase):
     def test_sin_ufunc(self):
         self.basic_ufunc_test(np.sin, kinds="cf")
 
@@ -167,6 +169,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     def test_arctan2_ufunc(self):
         self.basic_ufunc_test(np.arctan2, kinds="f")
 
+
+class TestHypTrigUFuncs(CUDAUFuncTestBase):
     def test_hypot_ufunc(self):
         self.basic_ufunc_test(np.hypot, kinds="f")
 
@@ -207,6 +211,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
 
         self.basic_ufunc_test(np.arctanh, skip_inputs=to_skip, kinds="cf")
 
+
+class TestConversionUFuncs(CUDAUFuncTestBase):
     def test_deg2rad_ufunc(self):
         self.basic_ufunc_test(np.deg2rad, kinds="f")
 
@@ -221,6 +227,9 @@ class TestUFuncs(BasicUFuncTest, TestCase):
 
     ############################################################################
     # Comparison functions
+
+
+class TestComparisonUFuncs1(CUDAUFuncTestBase):
     def test_greater_ufunc(self):
         self.signed_unsigned_cmp_test(np.greater)
 
@@ -239,6 +248,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     def test_equal_ufunc(self):
         self.signed_unsigned_cmp_test(np.equal)
 
+
+class TestLogicalUFuncs(CUDAUFuncTestBase):
     def test_logical_and_ufunc(self):
         self.basic_ufunc_test(np.logical_and)
 
@@ -251,6 +262,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     def test_logical_not_ufunc(self):
         self.basic_ufunc_test(np.logical_not)
 
+
+class TestMinmaxUFuncs(CUDAUFuncTestBase):
     def test_maximum_ufunc(self):
         self.basic_ufunc_test(np.maximum)
 
@@ -263,6 +276,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     def test_fmin_ufunc(self):
         self.basic_ufunc_test(np.fmin)
 
+
+class TestBitwiseUFuncs(CUDAUFuncTestBase):
     def test_bitwise_and_ufunc(self):
         self.basic_int_ufunc_test(np.bitwise_and)
 
@@ -286,6 +301,8 @@ class TestUFuncs(BasicUFuncTest, TestCase):
     ############################################################################
     # Mathematical Functions
 
+
+class TestLogUFuncs(CUDAUFuncTestBase):
     def test_log_ufunc(self):
         self.basic_ufunc_test(np.log, kinds="cf")
 

numba_cuda/numba/cuda/tests/cudapy/test_warning.py

@@ -1,7 +1,12 @@
 import numpy as np
 from numba import cuda
-from numba.cuda.testing import unittest, CUDATestCase, skip_on_cudasim
-from numba.tests.support import linux_only, override_config
+from numba.cuda.cudadrv import driver
+from numba.cuda.testing import (
+    unittest,
+    CUDATestCase,
+    skip_on_cudasim,
+)
+from numba.tests.support import linux_only, override_config, run_in_subprocess
 from numba.core.errors import NumbaPerformanceWarning
 from numba.core import config
 import warnings
@@ -9,6 +14,26 @@ import warnings
 
 @skip_on_cudasim("cudasim does not raise performance warnings")
 class TestWarnings(CUDATestCase):
+    def test_float16_warn_if_lto_missing(self):
+        fp16_kernel_invocation = """
+import math
+from numba import cuda, core
+
+@cuda.jit
+def kernel():
+    x = core.types.float16(1.0)
+    y = math.sin(x)
+
+kernel[1,1]()
+kernel[1,1]()
+"""
+        performance_warning = "float16 relies on LTO for performance"
+        expected_warning_count = 0 if driver._have_nvjitlink() else 1
+        _, err = run_in_subprocess(fp16_kernel_invocation)
+        self.assertEqual(
+            err.decode().count(performance_warning), expected_warning_count
+        )
+
     def test_inefficient_launch_configuration(self):
         @cuda.jit
         def kernel():

numba_cuda/numba/cuda/tests/enum_usecases.py

@@ -0,0 +1,56 @@
+from enum import Enum, IntEnum
+
+
+class Color(Enum):
+    red = 1
+    green = 2
+    blue = 3
+
+
+class Shake(Enum):
+    vanilla = 7
+    chocolate = 4
+    cookies = 9
+    # Same as Color.blue
+    mint = 3
+
+
+class Planet(Enum):
+    MERCURY = (3.303e23, 2.4397e6)
+    VENUS = (4.869e24, 6.0518e6)
+    EARTH = (5.976e24, 6.37814e6)
+    MARS = (6.421e23, 3.3972e6)
+    JUPITER = (1.9e27, 7.1492e7)
+    SATURN = (5.688e26, 6.0268e7)
+    URANUS = (8.686e25, 2.5559e7)
+    NEPTUNE = (1.024e26, 2.4746e7)
+
+
+class HeterogeneousEnum(Enum):
+    red = 1.0
+    green = 2.0
+    blue = 3j
+
+
+class Shape(IntEnum):
+    # Same as Color.green
+    circle = 2
+    # Same as RequestError.internal_error
+    square = 500
+
+
+class RequestError(IntEnum):
+    dummy = 2
+    not_found = 404
+    internal_error = 500
+
+
+class IntEnumWithNegatives(IntEnum):
+    # Used for testing of hash, need to make sure -1 -> -2 to comply with CPy
+    one = 1
+    two = 2
+    too = 2
+    three = 3
+    negone = -1
+    negtwo = -2
+    negthree = -3

numba_cuda/numba/cuda/tests/nocuda/test_library_lookup.py

@@ -9,7 +9,6 @@ from numba.cuda.cudadrv import nvvm
 from numba.cuda.testing import (
     unittest,
     skip_on_cudasim,
-    SerialMixin,
     skip_unless_conda_cudatoolkit,
 )
 from numba.cuda.cuda_paths import (
@@ -24,7 +23,7 @@ has_cuda = nvvm.is_available()
 has_mp_get_context = hasattr(mp, "get_context")
 
 
-class LibraryLookupBase(SerialMixin, unittest.TestCase):
+class LibraryLookupBase(unittest.TestCase):
     def setUp(self):
         ctx = mp.get_context("spawn")
 

numba_cuda/numba/cuda/tests/nocuda/test_nvvm.py

@@ -1,6 +1,6 @@
 from numba.cuda.cudadrv import nvvm
 from numba.cuda.testing import skip_on_cudasim
-from numba.core import utils
+from numba.cuda import utils
 
 from llvmlite import ir
 from llvmlite import binding as llvm