numba-cuda 0.19.1__py3-none-any.whl → 0.20.1__py3-none-any.whl

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

@@ -629,7 +629,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.array([3.0], dtype=np.float16)
         arg2 = np.array([4.0], dtype=np.float16)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg1 + arg2)
+        np.testing.assert_allclose(ary[0], arg1 + arg2, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hadd_scalar(self):
@@ -639,7 +639,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg2 = np.float16(3.0)
         compiled[1, 1](ary, arg1, arg2)
         ref = arg1 + arg2
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -657,7 +657,9 @@ class TestCudaIntrinsic(CUDATestCase):
         arg2 = np.array([3.0], dtype=np.float16)
         arg3 = np.array([4.0], dtype=np.float16)
         compiled[1, 1](ary, arg1, arg2, arg3)
-        np.testing.assert_allclose(ary[0], arg1 * arg2 + arg3)
+        np.testing.assert_allclose(
+            ary[0], arg1 * arg2 + arg3, rtol=self.FLOAT16_RTOL
+        )
 
     @skip_unless_cc_53
     def test_hfma_scalar(self):
@@ -668,7 +670,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg3 = np.float16(4.0)
         compiled[1, 1](ary, arg1, arg2, arg3)
         ref = arg1 * arg2 + arg3
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -687,7 +689,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.array([3.0], dtype=np.float16)
         arg2 = np.array([4.0], dtype=np.float16)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg1 - arg2)
+        np.testing.assert_allclose(ary[0], arg1 - arg2, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hsub_scalar(self):
@@ -697,7 +699,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg2 = np.float16(1.57)
         compiled[1, 1](ary, arg1, arg2)
         ref = arg1 - arg2
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -714,7 +716,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.array([3.0], dtype=np.float16)
         arg2 = np.array([4.0], dtype=np.float16)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg1 * arg2)
+        np.testing.assert_allclose(ary[0], arg1 * arg2, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hmul_scalar(self):
@@ -724,7 +726,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg2 = np.float16(1.57)
         compiled[1, 1](ary, arg1, arg2)
         ref = arg1 * arg2
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -743,7 +745,7 @@ class TestCudaIntrinsic(CUDATestCase):
 
         compiled[1, 1](ary, arg1, arg2)
         ref = arg1 / arg2
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hdiv(self):
@@ -754,7 +756,7 @@ class TestCudaIntrinsic(CUDATestCase):
 
         compiled.forall(ary.size)(ary, arry1, arry2)
         ref = arry1 / arry2
-        np.testing.assert_allclose(ary, ref)
+        np.testing.assert_allclose(ary, ref, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hneg(self):
@@ -762,7 +764,7 @@ class TestCudaIntrinsic(CUDATestCase):
         ary = np.zeros(1, dtype=np.float16)
         arg1 = np.array([3.0], dtype=np.float16)
         compiled[1, 1](ary, arg1)
-        np.testing.assert_allclose(ary[0], -arg1)
+        np.testing.assert_allclose(ary[0], -arg1, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hneg_scalar(self):
@@ -771,7 +773,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.float16(3.1415926)
         compiled[1, 1](ary, arg1)
         ref = -arg1
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -787,7 +789,7 @@ class TestCudaIntrinsic(CUDATestCase):
         ary = np.zeros(1, dtype=np.float16)
         arg1 = np.array([-3.0], dtype=np.float16)
         compiled[1, 1](ary, arg1)
-        np.testing.assert_allclose(ary[0], abs(arg1))
+        np.testing.assert_allclose(ary[0], abs(arg1), rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_habs_scalar(self):
@@ -796,7 +798,7 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.float16(-3.1415926)
         compiled[1, 1](ary, arg1)
         ref = abs(arg1)
-        np.testing.assert_allclose(ary[0], ref)
+        np.testing.assert_allclose(ary[0], ref, rtol=self.FLOAT16_RTOL)
 
     @skip_on_cudasim("Compilation unsupported in the simulator")
     @skip_if_nvjitlink_missing("Numbast generated bindings")
@@ -849,7 +851,7 @@ class TestCudaIntrinsic(CUDATestCase):
                 kernel = cuda.jit("void(f2[:], f2[:])")(kernel)
                 kernel[1, N](r, x)
                 expected = fn(x, dtype=np.float16)
-                np.testing.assert_allclose(r, expected)
+                np.testing.assert_allclose(r, expected, rtol=self.FLOAT16_RTOL)
 
         x2 = np.random.randint(1, 10, size=N).astype(np.float16)
         for kernel, fn in zip(exp_kernels, expected_exp_functions):
@@ -857,7 +859,7 @@ class TestCudaIntrinsic(CUDATestCase):
                 kernel = cuda.jit("void(f2[:], f2[:])")(kernel)
                 kernel[1, N](r, x2)
                 expected = fn(x2, dtype=np.float16)
-                np.testing.assert_allclose(r, expected)
+                np.testing.assert_allclose(r, expected, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hexp10(self):
@@ -876,7 +878,7 @@ class TestCudaIntrinsic(CUDATestCase):
 
         # Run the kernel
         hexp10_vectors[1, N](r, x)
-        np.testing.assert_allclose(r, 10**x)
+        np.testing.assert_allclose(r, 10**x, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_fp16_comparison(self):
@@ -948,10 +950,10 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.float16(3.0)
         arg2 = np.float16(4.0)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg2)
+        np.testing.assert_allclose(ary[0], arg2, rtol=self.FLOAT16_RTOL)
         arg1 = np.float16(5.0)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg1)
+        np.testing.assert_allclose(ary[0], arg1, rtol=self.FLOAT16_RTOL)
 
     @skip_unless_cc_53
     def test_hmin(self):
@@ -960,10 +962,10 @@ class TestCudaIntrinsic(CUDATestCase):
         arg1 = np.float16(3.0)
         arg2 = np.float16(4.0)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg1)
+        np.testing.assert_allclose(ary[0], arg1, rtol=self.FLOAT16_RTOL)
         arg1 = np.float16(5.0)
         compiled[1, 1](ary, arg1, arg2)
-        np.testing.assert_allclose(ary[0], arg2)
+        np.testing.assert_allclose(ary[0], arg2, rtol=self.FLOAT16_RTOL)
 
     def test_cbrt_f32(self):
         compiled = cuda.jit("void(float32[:], float32)")(simple_cbrt)

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

@@ -2,63 +2,34 @@
 # SPDX-License-Identifier: BSD-2-Clause
 
 from numba.cuda.testing import CUDATestCase
-import numba
-from numba.core.registry import cpu_target
-from numba.core.compiler import CompilerBase, Flags
-from numba.core.compiler_machinery import PassManager
+from numba.cuda.core.compiler import CompilerBase
+from numba.cuda.flags import Flags
+from numba.cuda.core.compiler_machinery import PassManager
 from numba.cuda.core import ir_utils
-from numba.core import types, ir, bytecode, compiler, registry
-from numba.core.untyped_passes import (
+from numba.core import types, ir, bytecode, registry
+from numba.cuda import compiler
+from numba.cuda.core.untyped_passes import (
     ExtractByteCode,
     TranslateByteCode,
     FixupArgs,
     IRProcessing,
 )
-from numba.experimental import jitclass
-from numba.core.typed_passes import (
+from numba.cuda.core.typed_passes import (
     NopythonTypeInference,
-    type_inference_stage,
     DeadCodeElimination,
 )
+from numba.cuda.testing import skip_on_cudasim
 
 
 # global constant for testing find_const
 GLOBAL_B = 11
 
 
-@jitclass([("val", numba.core.types.List(numba.intp))])
-class Dummy(object):
-    def __init__(self, val):
-        self.val = val
-
-
 class TestIrUtils(CUDATestCase):
     """
     Tests ir handling utility functions like find_callname.
     """
 
-    def test_obj_func_match(self):
-        """Test matching of an object method (other than Array see #3449)"""
-
-        def test_func():
-            d = Dummy([1])
-            d.val.append(2)
-
-        test_ir = compiler.run_frontend(test_func)
-        typingctx = cpu_target.typing_context
-        targetctx = cpu_target.target_context
-        typing_res = type_inference_stage(
-            typingctx, targetctx, test_ir, (), None
-        )
-        matched_call = ir_utils.find_callname(
-            test_ir, test_ir.blocks[0].body[7].value, typing_res.typemap
-        )
-        self.assertTrue(
-            isinstance(matched_call, tuple)
-            and len(matched_call) == 2
-            and matched_call[0] == "append"
-        )
-
     def test_dead_code_elimination(self):
         class Tester(CompilerBase):
             @classmethod
@@ -177,6 +148,7 @@ class TestIrUtils(CUDATestCase):
             len(no_dce.blocks[0].body) - len(removed), len(w_dce.blocks[0].body)
         )
 
+    @skip_on_cudasim("Skipping ir utils tests on CUDA simulator")
     def test_find_const_global(self):
         """
         Test find_const() for values in globals (ir.Global) and freevars
@@ -203,6 +175,7 @@ class TestIrUtils(CUDATestCase):
         self.assertEqual(const_b, GLOBAL_B)
         self.assertEqual(const_c, FREEVAR_C)
 
+    @skip_on_cudasim("Skipping ir utils tests on CUDA simulator")
     def test_flatten_labels(self):
         """tests flatten_labels"""
 

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

@@ -4,7 +4,7 @@
 import numpy as np
 from numba import cuda, float64, void
 from numba.cuda.testing import unittest, CUDATestCase
-from numba.core import config
+from numba.cuda.core import config
 
 # NOTE: CUDA kernel does not return any value
 

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

@@ -8,7 +8,7 @@ from numba.cuda.testing import (
     CUDATestCase,
     skip_on_cudasim,
 )
-from numba.np import numpy_support
+from numba.cuda.np import numpy_support
 from numba import cuda, float32, float64, int32, vectorize, void, int64
 import math
 

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

@@ -5,7 +5,7 @@ import numpy as np
 
 from numba import cuda, float32, void
 from numba.cuda.testing import unittest, CUDATestCase
-from numba.core import config
+from numba.cuda.core import config
 
 # Ensure the test takes a reasonable amount of time in the simulator
 if config.ENABLE_CUDASIM:

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

@@ -15,7 +15,7 @@ from numba.core.types import f2, b1
 from numba.cuda.typing import signature
 import operator
 import itertools
-from numba.np.numpy_support import from_dtype
+from numba.cuda.np.numpy_support import from_dtype
 
 
 def simple_fp16_div_scalar(ary, a, b):

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

@@ -102,6 +102,21 @@ print_too_many[1, 1](np.arange(33))
 cuda.synchronize()
 """
 
+print_bfloat16_usecase = """\
+from numba import cuda
+from numba.cuda import config
+
+@cuda.jit
+def print_bfloat16():
+    # 0.9375 is a dyadic rational, it's integer significand can expand within 7 digits.
+    # printing this should not give any rounding error.
+    a = cuda.types.bfloat16(0.9375)
+    print(a, a, a)
+
+print_bfloat16[1, 1]()
+cuda.synchronize()
+"""
+
 
 class TestPrint(CUDATestCase):
     # Note that in these tests we generally strip the output to avoid dealing
@@ -148,6 +163,11 @@ class TestPrint(CUDATestCase):
         expected = [str(i) for i in np.ndindex(2, 2, 2)]
         self.assertEqual(sorted(lines), expected)
 
+    @skip_on_cudasim("bfloat16 on host is not yet supported.")
+    def test_bfloat16(self):
+        output, _ = self.run_code(print_bfloat16_usecase)
+        self.assertEqual(output.strip(), "0.937500 0.937500 0.937500")
+
     @skip_on_cudasim("cudasim can print unlimited output")
     def test_too_many_args(self):
         # Tests that we emit the format string and warn when there are more

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

@@ -6,7 +6,7 @@ from numba import cuda
 from numba.core import types
 from numba.cuda.testing import skip_on_cudasim, CUDATestCase
 import unittest
-from numba.np import numpy_support
+from numba.cuda.np import numpy_support
 
 
 def set_a(ary, i, v):

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

@@ -3,7 +3,7 @@
 
 import numpy as np
 from numba import cuda
-from numba.core.config import ENABLE_CUDASIM
+from numba.cuda.core.config import ENABLE_CUDASIM
 from numba.cuda.testing import CUDATestCase
 import unittest
 

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

@@ -7,7 +7,7 @@ from numba import cuda, vectorize
 from numba.core import types
 from numba.cuda.testing import skip_on_cudasim, CUDATestCase
 import unittest
-from numba.np import numpy_support
+from numba.cuda.np import numpy_support
 
 
 @skip_on_cudasim("pickling not supported in CUDASIM")

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

@@ -7,7 +7,7 @@ from numba.core import types
 from numba.cuda.testing import unittest, CUDATestCase, skip_on_cudasim
 
 import numpy as np
-from numba.np import numpy_support as nps
+from numba.cuda.np import numpy_support as nps
 
 from .extensions_usecases import test_struct_model_type, TestStruct