numba-cuda 0.20.0__py3-none-any.whl → 0.20.1__py3-none-any.whl

numba_cuda/VERSION

@@ -1 +1 @@
-0.20.0
+0.20.1

numba_cuda/numba/cuda/testing.py

@@ -17,6 +17,7 @@ from pathlib import Path
 from typing import Iterable, Union
 from io import StringIO
 import unittest
+import numpy as np
 
 if PYVERSION >= (3, 10):
     from filecheck.matcher import Matcher
@@ -44,6 +45,8 @@ class CUDATestCase(TestCase):
     matches FileCheck checks, and is not specific to CUDADispatcher.
     """
 
+    FLOAT16_RTOL = np.finfo(np.float16).eps
+
     def setUp(self):
         self._low_occupancy_warnings = config.CUDA_LOW_OCCUPANCY_WARNINGS
         self._warn_on_implicit_copy = config.CUDA_WARN_ON_IMPLICIT_COPY

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-0.20.0.dist-info → numba_cuda-0.20.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: numba-cuda
-Version: 0.20.0
+Version: 0.20.1
 Summary: CUDA target for Numba
 Author: Anaconda Inc., NVIDIA Corporation
 License-Expression: BSD-2-Clause
@@ -16,7 +16,7 @@ License-File: LICENSE.numba
 Requires-Dist: numba>=0.60.0
 Provides-Extra: cu12
 Requires-Dist: cuda-bindings<13.0.0,>=12.9.1; extra == "cu12"
-Requires-Dist: cuda-core==0.3.*; extra == "cu12"
+Requires-Dist: cuda-core<1.0.0,>=0.3.0; extra == "cu12"
 Requires-Dist: cuda-python==12.9.*; extra == "cu12"
 Requires-Dist: nvidia-cuda-nvcc-cu12; extra == "cu12"
 Requires-Dist: nvidia-cuda-runtime-cu12; extra == "cu12"
@@ -25,7 +25,7 @@ Requires-Dist: nvidia-nvjitlink-cu12; extra == "cu12"
 Requires-Dist: nvidia-cuda-cccl-cu12; extra == "cu12"
 Provides-Extra: cu13
 Requires-Dist: cuda-bindings==13.*; extra == "cu13"
-Requires-Dist: cuda-core<0.4.0dev0,==0.3.2; extra == "cu13"
+Requires-Dist: cuda-core<1.0.0,>=0.3.2; extra == "cu13"
 Requires-Dist: cuda-python==13.*; extra == "cu13"
 Requires-Dist: nvidia-nvvm==13.*; extra == "cu13"
 Requires-Dist: nvidia-cuda-runtime==13.*; extra == "cu13"

{numba_cuda-0.20.0.dist-info → numba_cuda-0.20.1.dist-info}/RECORD

@@ -1,6 +1,6 @@
 _numba_cuda_redirector.pth,sha256=Il6zlJxEoQiz1AdLkQhk8m_7WIoS50MmOMLVgXBGcN4,170
 _numba_cuda_redirector.py,sha256=hJphzrGT9qp3-NtdQ51_Jv94lqlmrwf6S5Sh_UFvaEE,2799
-numba_cuda/VERSION,sha256=jaai7lw2uydw3mRcvNaJIJ9ZNbOA1ocp4QfQJQpjO1M,7
+numba_cuda/VERSION,sha256=NWycnwmdp_mzgMb-3FEHfumN4iQCs3i8D0Mf3Feqx1c,7
 numba_cuda/__init__.py,sha256=4P6_SMBIgPf3E7O2rxPXSz_MbSAHNqx0Rng-nsgyGQY,211
 numba_cuda/_version.py,sha256=4SQp70c0OEm9jbZUU0iI72YiAs79LOoMQiF0tTMEg2I,285
 numba_cuda/numba/cuda/__init__.py,sha256=jXQGekoQK5wxXH7z8cUoxML1rOr4idJHU7evV2G_0DU,3531
@@ -46,7 +46,7 @@ numba_cuda/numba/cuda/serialize.py,sha256=Zx2a2_G27zfbxFJKTJkJO2AThc3MBracTAjV-s
 numba_cuda/numba/cuda/simulator_init.py,sha256=FT9Max5WAS4M0JLGomoZxLKstXeLi0Iuq-1pqxUQyC8,590
 numba_cuda/numba/cuda/stubs.py,sha256=2UALzKfJYa3IlwpYxR-6nJSxQYaCnkioQLd48JP2KGc,16667
 numba_cuda/numba/cuda/target.py,sha256=Rgi5MHG-m0k42_x0Q_9cTZgDtdGCq68MO-eYmJHmtoM,15423
-numba_cuda/numba/cuda/testing.py,sha256=iYWK_y4OxzUEgLiRkz5oPGKCTYJY_bGDX8mCg2TfWLo,11792
+numba_cuda/numba/cuda/testing.py,sha256=rewVOq7240JyrffueHPudjFNV87wU1mWCQ28HzsUUP8,11856
 numba_cuda/numba/cuda/types.py,sha256=5QsQHbw4Vj0FCxHlltMcAbSEjPPCgchR6uQgZ1fD5No,2967
 numba_cuda/numba/cuda/ufuncs.py,sha256=lT2hoM3U4OuNrfUDuIt7lrfOBEA_pUKpdDKU_Oe_XF8,23714
 numba_cuda/numba/cuda/utils.py,sha256=6dA0uW4eX4ms3TRZWJQk3Nb70ncG0A1S0NwYhGR4d1Q,19259
@@ -253,7 +253,7 @@ numba_cuda/numba/cuda/tests/cudapy/test_gufunc_scheduling.py,sha256=oLP8RUmgh53J
 numba_cuda/numba/cuda/tests/cudapy/test_idiv.py,sha256=M89feOfFGRiva1FpOm5bA3OpyNjEFugT5RzjsKFCNdI,1194
 numba_cuda/numba/cuda/tests/cudapy/test_inline.py,sha256=R-yq6VlRYQRrKaOa-n0OJE7lBq8dMrO-OOBcrjlD-9U,5048
 numba_cuda/numba/cuda/tests/cudapy/test_inspect.py,sha256=idteoOYUoSidwCkQ395GQq8GcbcLIWTsegAoQCAOSfU,7540
-numba_cuda/numba/cuda/tests/cudapy/test_intrinsics.py,sha256=_tHm47rDfosrg3yXm6Hr1m1I4xgqzwIk7MKRjk6tnmU,37554
+numba_cuda/numba/cuda/tests/cudapy/test_intrinsics.py,sha256=CgRu8tvy_H8HC2KbFPYEvM7Mjiil_6dru01cBrJJSM0,38080
 numba_cuda/numba/cuda/tests/cudapy/test_ipc.py,sha256=Z6CC24xKU2rhdJRK-UAPeorYoE1x-0ZnS4V0YjV5Yy4,10698
 numba_cuda/numba/cuda/tests/cudapy/test_ir_utils.py,sha256=6qBuuabtsxa5bITbZ-KH6_AR17ZoG8fFyMHG4vnZ1bs,9044
 numba_cuda/numba/cuda/tests/cudapy/test_iterators.py,sha256=iAF15K202ylm9NdctQFBp7Kr4NIXoA9y4gYJ8L19dFU,2483
@@ -349,9 +349,9 @@ numba_cuda/numba/cuda/typing/context.py,sha256=Smca11cA_d2T6RlIkK6ERu30x72P0k8Te
 numba_cuda/numba/cuda/typing/enumdecl.py,sha256=ptGYmkVIzvG_hWa-h0Se-8IqK4tE5N9PZqfTsFy8p34,1652
 numba_cuda/numba/cuda/typing/npydecl.py,sha256=Fe-tl9Tsp3-etcTSEFVy50lm6DbrsJgdlcaGWtCIWxM,20057
 numba_cuda/numba/cuda/typing/templates.py,sha256=4xQaKFeXJSuZT8YvPYWGecJvjbBV0k_idKwhfx6G74E,51619
-numba_cuda-0.20.0.dist-info/licenses/LICENSE,sha256=qeS8IOtEoJl4qHh9S9LIPPJm77GQlJld2Rc3RNMhgk8,1327
-numba_cuda-0.20.0.dist-info/licenses/LICENSE.numba,sha256=OnDgohytAxRyRvY2PQe_dvriFscneXB_SuXbBrys8Gc,1286
-numba_cuda-0.20.0.dist-info/METADATA,sha256=ukJGGzojlcEaHHPRirauk9IPyUVdGOSh5Ech8YndLUk,3684
-numba_cuda-0.20.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-numba_cuda-0.20.0.dist-info/top_level.txt,sha256=C50SsH-8tXDmt7I0Y3nlJYhS5s6pqWflCPdobe9vx2M,11
-numba_cuda-0.20.0.dist-info/RECORD,,
+numba_cuda-0.20.1.dist-info/licenses/LICENSE,sha256=qeS8IOtEoJl4qHh9S9LIPPJm77GQlJld2Rc3RNMhgk8,1327
+numba_cuda-0.20.1.dist-info/licenses/LICENSE.numba,sha256=OnDgohytAxRyRvY2PQe_dvriFscneXB_SuXbBrys8Gc,1286
+numba_cuda-0.20.1.dist-info/METADATA,sha256=cJv8o_FMsnYhX3WU9MPU0VS8mVfWEXpD6PnCCEyDqWg,3687
+numba_cuda-0.20.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+numba_cuda-0.20.1.dist-info/top_level.txt,sha256=C50SsH-8tXDmt7I0Y3nlJYhS5s6pqWflCPdobe9vx2M,11
+numba_cuda-0.20.1.dist-info/RECORD,,