numba-cuda 0.18.1__py3-none-any.whl → 0.19.0__py3-none-any.whl

numba_cuda/VERSION

@@ -1 +1 @@
-0.18.1
+0.19.0

numba_cuda/numba/cuda/__init__.py

@@ -50,7 +50,7 @@ if config.CUDA_USE_NVIDIA_BINDING:
         )
 
 if config.CUDA_ENABLE_PYNVJITLINK:
-    if USE_NV_BINDING:
+    if USE_NV_BINDING and not pynvjitlink_auto_enabled:
         warnings.warn(
             "Explicitly enabling pynvjitlink is no longer necessary. "
             "NVIDIA bindings are enabled. cuda.core will be used "

numba_cuda/numba/cuda/_internal/cuda_bf16.py

@@ -23,8 +23,8 @@ from numba.core.extending import (
     make_attribute_wrapper,
     register_model,
 )
-from numba.core.typing import signature
-from numba.core.typing.templates import AttributeTemplate, ConcreteTemplate
+from numba.cuda.typing import signature
+from numba.cuda.typing.templates import AttributeTemplate, ConcreteTemplate
 from numba.cuda import CUSource, declare_device
 from numba.cuda.cudadecl import register, register_attr, register_global
 from numba.cuda.cudaimpl import lower

numba_cuda/numba/cuda/_internal/cuda_fp16.py

@@ -26,7 +26,7 @@ from numba.core.extending import (
 from numba.core.imputils import Registry as TargetRegistry, lower_cast
 import numba.core.typeconv
 from numba.core.typing import signature
-from numba.core.typing.templates import (
+from numba.cuda.typing.templates import (
     Registry as TypingRegistry,
     AbstractTemplate,
     AttributeTemplate,

numba_cuda/numba/cuda/api.py

@@ -10,7 +10,6 @@ import numpy as np
 from .cudadrv import devicearray, devices, driver
 from numba.core import config
 from numba.cuda.api_util import prepare_shape_strides_dtype
-from numba.cuda.cudadrv.runtime import get_version
 
 # NDarray device helper
 
@@ -99,13 +98,9 @@ def is_float16_supported():
 def is_bfloat16_supported():
     """Whether bfloat16 are supported.
 
-    bfloat16 are only supported on devices with compute capability >= 8.0 and cuda version >= 12.0
+    bfloat16 is only supported on devices with compute capability >= 8.0
     """
-    cuda_version = get_version()
-    return current_context().device.supports_bfloat16 and cuda_version >= (
-        12,
-        0,
-    )
+    return current_context().device.supports_bfloat16
 
 
 @require_context

numba_cuda/numba/cuda/compiler.py

@@ -1,14 +1,12 @@
 from llvmlite import ir
 from collections import namedtuple
 from numba.core import ir as numba_ir
-from numba.cuda import cgutils
+from numba.cuda import cgutils, typing
 from numba.core import (
     types,
-    typing,
     funcdesc,
     config,
     compiler,
-    sigutils,
 )
 from numba.core.compiler import (
     sanitize_compile_result_entries,
@@ -22,7 +20,6 @@ from numba.core.compiler_machinery import (
     PassManager,
     register_pass,
 )
-from numba.core.interpreter import Interpreter
 from numba.core.errors import NumbaInvalidConfigWarning
 from numba.core.untyped_passes import TranslateByteCode
 from numba.core.typed_passes import (
@@ -34,12 +31,18 @@ from numba.cuda import nvvmutils
 from numba.cuda.api import get_current_device
 from numba.cuda.codegen import ExternalCodeLibrary
 from numba.cuda.core.typed_passes import BaseNativeLowering
+from numba.cuda.core import sigutils
 from numba.cuda.cudadrv import nvvm, nvrtc
 from numba.cuda.descriptor import cuda_target
 from numba.cuda.flags import CUDAFlags
 from numba.cuda.target import CUDACABICallConv
 from numba.cuda import lowering, utils
+from numba.core.utils import PYVERSION
 
+if PYVERSION < (3, 10):
+    from numba.core.interpreter import Interpreter
+else:
+    from numba.cuda.core.interpreter import Interpreter
 
 # The CUDACompileResult (CCR) has a specially-defined entry point equal to its
 # id.  This is because the entry point is used as a key into a dict of