Merge pull request #8392 from apmasell/release0.56.1

Release0.56.1

CHANGE_LOG

@@ -1,3 +1,25 @@
+Version 0.56.1 (1 September, 2022)
+----------------------------------
+
+This is a bugfix release that supports NumPy 1.23 and fixes CUDA function
+caching.
+
+Pull-Requests:
+
+* PR `#8239 <https://github.com/numba/numba/pull/8239>`_: Add decorator to run a test in a subprocess (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8276 <https://github.com/numba/numba/pull/8276>`_: Move Azure to use macos-11 (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8310 <https://github.com/numba/numba/pull/8310>`_: CUDA: Fix Issue #8309 - atomics don't work on complex components (`Graham Markall <https://github.com/gmarkall>`_)
+* PR `#8342 <https://github.com/numba/numba/pull/8342>`_: Upgrade to ubuntu-20.04 for azure pipeline CI (`jamesobutler <https://github.com/jamesobutler>`_)
+* PR `#8356 <https://github.com/numba/numba/pull/8356>`_: Update setup.py, buildscripts, CI and docs to require setuptools<60 (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8374 <https://github.com/numba/numba/pull/8374>`_: Don't pickle LLVM IR for CUDA code libraries (`Graham Markall <https://github.com/gmarkall>`_)
+* PR `#8377 <https://github.com/numba/numba/pull/8377>`_: Add support for NumPy 1.23 (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8384 <https://github.com/numba/numba/pull/8384>`_: Move strace() check into tests that actually need it (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8386 <https://github.com/numba/numba/pull/8386>`_: Fix the docs for numba.get_thread_id (`stuartarchibald <https://github.com/stuartarchibald>`_)
+* PR `#8407 <https://github.com/numba/numba/pull/8407>`_: Pin NumPy version to 1.18-1.24 (`Andre Masella <https://github.com/apmasell>`_)
+* PR `#8411 <https://github.com/numba/numba/pull/8411>`_: update version support table for 0.56.1 (`esc <https://github.com/esc>`_)
+* PR `#8412 <https://github.com/numba/numba/pull/8412>`_: Create changelog for 0.56.1 (`Andre Masella <https://github.com/apmasell>`_)
+* PR `#8413 <https://github.com/numba/numba/pull/8413>`_: Fix Azure CI for NumPy 1.23 and use conda-forge scipy (`Siu Kwan Lam <https://github.com/sklam>`_)
+
 Version 0.56.0 (25 July, 2022)
 ------------------------------
 

azure-pipelines.yml

@@ -10,24 +10,24 @@ jobs:
 - template: buildscripts/azure/azure-linux-macos.yml
   parameters:
     name: macOS
-    vmImage: macOS-10.15
+    vmImage: macos-11
     matrix:
       py37_np118:
         PYTHON: '3.7'
         NUMPY: '1.18'
         CONDA_ENV: 'azure_ci'
         TEST_START_INDEX: 0
-      py310_np122:
+      py310_np123:
         PYTHON: '3.10'
-        NUMPY: '1.22'
+        NUMPY: '1.23'
         CONDA_ENV: 'azure_ci'
         TEST_THREADING: 'tbb'
         TEST_START_INDEX: 1
 
 - template: buildscripts/azure/azure-linux-macos.yml
   parameters:
     name: Linux
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
     matrix:
       py37_np118_32bit:
         # 32 bit linux only has np 1.15
@@ -107,19 +107,19 @@ jobs:
         NUMPY: '1.21'
         CONDA_ENV: azure_ci
         TEST_START_INDEX: 14
-      py39_np122:
+      py39_np123:
         PYTHON: '3.9'
-        NUMPY: '1.22'
+        NUMPY: '1.23'
         CONDA_ENV: azure_ci
         TEST_START_INDEX: 15
       py310_np121:
         PYTHON: '3.10'
         NUMPY: '1.21'
         CONDA_ENV: azure_ci
         TEST_START_INDEX: 16
-      py310_np122:
+      py310_np123:
         PYTHON: '3.10'
-        NUMPY: '1.22'
+        NUMPY: '1.23'
         CONDA_ENV: azure_ci
         TEST_START_INDEX: 17
 

buildscripts/azure/azure-windows.yml

@@ -8,9 +8,9 @@ jobs:
     vmImage: ${{ parameters.vmImage }}
   strategy:
     matrix:
-      py310_np122:
+      py310_np123:
         PYTHON: '3.10'
-        NUMPY: '1.22'
+        NUMPY: '1.23'
         CONDA_ENV: 'testenv'
         TEST_START_INDEX: 18
       py37_np118:

buildscripts/condarecipe.local/meta.yaml

@@ -32,7 +32,7 @@ requirements:
   host:
     - python
     - numpy
-    - setuptools
+    - setuptools <60
     - importlib_metadata       # [py<39]
     # On channel https://anaconda.org/numba/
     - llvmlite 0.39.*
@@ -45,8 +45,8 @@ requirements:
   run:
     - python >=3.7
     # NumPy 1.22.0, 1.22.1, 1.22.2 are all broken for ufuncs, see #7756
-    - numpy >=1.18, !=1.22.0, !=1.22.1, !=1.22.2, <1.24
-    - setuptools
+    - numpy >=1.18, !=1.22.0, !=1.22.1, !=1.22.2, <1.23
+    - setuptools <60
     - importlib_metadata       # [py<39]
     # On channel https://anaconda.org/numba/
     - llvmlite 0.39.*
@@ -76,7 +76,7 @@ test:
     # temporarily disable scipy testing on ARM, need to build out more packages
     - scipy                    # [not (armv6l or armv7l)]
     - ipython                  # [not (armv6l or armv7l or aarch64)]
-    - setuptools
+    - setuptools <60
     - tbb  >=2021              # [not (armv6l or armv7l or aarch64 or linux32 or ppc64le)]
     - llvm-openmp              # [osx]
     # This is for driving gdb tests

buildscripts/gpuci/build.sh

@@ -45,14 +45,15 @@ gpuci_mamba_retry create -n numba_ci -y \
                   "cffi" \
                   "psutil" \
                   "gcc_linux-64=7" \
-                  "gxx_linux-64=7"
+                  "gxx_linux-64=7" \
+                  "setuptools<60"
 
 conda activate numba_ci
 
 if [ $NUMBA_CUDA_USE_NVIDIA_BINDING == "1" ]
 then
   gpuci_logger "Install NVIDIA CUDA Python bindings";
-  gpuci_mamba_retry install nvidia::cuda-python;
+  gpuci_mamba_retry install nvidia::cuda-python=11.7.0;
 fi;
 
 gpuci_logger "Install numba"

buildscripts/incremental/setup_conda_environment.cmd

@@ -22,9 +22,11 @@ call deactivate
 @rem Display root environment (for debugging)
 conda list
 @rem Scipy, CFFI, jinja2 and IPython are optional dependencies, but exercised in the test suite
-conda create -n %CONDA_ENV% -q -y python=%PYTHON% numpy=%NUMPY% cffi pip scipy jinja2 ipython gitpython pyyaml
+conda create -n %CONDA_ENV% -q -y python=%PYTHON% numpy=%NUMPY% cffi pip jinja2 ipython gitpython pyyaml "setuptools<60"
 
 call activate %CONDA_ENV%
+@rem Scipy comes from conda-forge for NumPy 1.23
+if %NUMPY% == "1.23" (%CONDA_INSTALL% conda-forge::scipy) else (%CONDA_INSTALL% scipy)
 @rem Install latest llvmlite build
 %CONDA_INSTALL% -c numba/label/dev llvmlite=0.39
 @rem Install required backports for older Pythons

buildscripts/incremental/setup_conda_environment.sh

@@ -43,7 +43,7 @@ conda list
 if [[ "$CONDA_SUBDIR" == "linux-32" || "$BITS32" == "yes" ]]; then
     conda create -n $CONDA_ENV -q -y ${EXTRA_CHANNELS} python=$PYTHON pip gitpython pyyaml
 else
-    conda create -n $CONDA_ENV -q -y ${EXTRA_CHANNELS} python=$PYTHON numpy=$NUMPY pip gitpython pyyaml
+    conda create -n $CONDA_ENV -q -y ${EXTRA_CHANNELS} python=$PYTHON numpy=$NUMPY pip gitpython pyyaml "setuptools<60"
 fi
 
 # Activate first
@@ -63,7 +63,11 @@ if [ "${VANILLA_INSTALL}" != "yes" ]; then
     # Only install scipy on 64bit, else it'll pull in NumPy, 32bit linux needs
     # to get scipy from pip
     if [[ "$CONDA_SUBDIR" != "linux-32" && "$BITS32" != "yes" ]] ; then
-        $CONDA_INSTALL ${EXTRA_CHANNELS} scipy
+        if [[ "$NUMPY" == "1.23" ]] ; then
+            $CONDA_INSTALL ${EXTRA_CHANNELS} conda-forge::scipy
+        else
+            $CONDA_INSTALL ${EXTRA_CHANNELS} scipy
+        fi
     fi
 fi
 

docs/environment.yml

@@ -9,7 +9,7 @@ dependencies:
   - llvmlite=0.39
   - numpy
   - numpydoc
-  - setuptools
+  - setuptools<60
   # https://stackoverflow.com/questions/67542699/readthedocs-sphinx-not-rendering-bullet-list-from-rst-fileA
   - docutils==0.16
   # The following is needed to fix RTD.

docs/source/user/installing.rst

@@ -194,14 +194,14 @@ vary with target operating system and hardware. The following lists them all
 
 * Required build time:
 
-  * ``setuptools``
+  * ``setuptools<60``
   * ``numpy``
   * ``llvmlite``
   * Compiler toolchain mentioned above
 
 * Required run time:
 
-  * ``setuptools``
+  * ``setuptools<60``
   * ``numpy``
   * ``llvmlite``
 
@@ -278,7 +278,9 @@ information.
 +===========+==============+===========================+============================+==============================+===================+=============================+
 | 0.57.x    | TBC          | 3.8.x <= version < 3.12   | 1.19 <= version < 1.24     | 0.40.x                       | 11.x              | 2021.x                      |
 +-----------+--------------+---------------------------+----------------------------+------------------------------+-------------------+-----------------------------+
-| 0.56.x    | 2022-07-25   | 3.7.x <= version < 3.11   | 1.18 <= version < 1.23     | 0.39.x                       | 11.x              | 2021.x                      |
+| 0.56.1    | TBC          | 3.7.x <= version < 3.11   | 1.18 <= version < 1.24     | 0.39.x                       | 11.x              | 2021.x                      |
++-----------+--------------+---------------------------+----------------------------+------------------------------+-------------------+-----------------------------+
+| 0.56.0    | 2022-07-25   | 3.7.x <= version < 3.11   | 1.18 <= version < 1.23     | 0.39.x                       | 11.x              | 2021.x                      |
 +-----------+--------------+---------------------------+----------------------------+------------------------------+-------------------+-----------------------------+
 | 0.55.2    | 2022-05-25   | 3.7.x <= version < 3.11   | 1.18 <= version < 1.23     | 0.38.x                       | 11.x              | 2021.x                      |
 +-----------+--------------+---------------------------+----------------------------+------------------------------+-------------------+-----------------------------+

docs/source/user/threading-layer.rst

@@ -276,6 +276,18 @@ the parallel code will only execute on 2 threads. However, we can later call
 size. And we do not have to worry about setting it before Numba gets imported.
 It only needs to be called before the parallel function is run.
 
+.. _numba-threading-layer-thread-id:
+
+Getting a Thread ID
+-------------------
+
+In some cases it may be beneficial to have access to a unique identifier for the
+current thread that is executing a parallel region in Numba. For that purpose,
+Numba provides the :func:`numba.get_thread_id` function. This function is the
+corollary of OpenMP's function ``omp_get_thread_num`` and returns an integer
+between 0 (inclusive) and the number of configured threads as described above
+(exclusive).
+
 API Reference
 ~~~~~~~~~~~~~
 
@@ -298,13 +310,4 @@ API Reference
 
 .. autofunction:: numba.get_num_threads
 
-Getting a Thread ID
--------------------
-
-In some cases it may be beneficial to have access to a unique identifier
-for the current thread that is executing a parallel region in Numba.
-For that purpose,
-Numba provides the ``get_thread_id()`` function.  This function is the corollary of
-OpenMP's function ``omp_get_thread_num`` and returns an integer between 0 (inclusive)
-and the number of configured threads as described above (exclusive).
-
+.. autofunction:: numba.get_thread_id

numba/core/typing/arraydecl.py

@@ -280,7 +280,23 @@ def resolve_T(self, ary):
             retty = ary.copy(layout=layout)
         return retty
 
-    # NOTE: .imag and .real are implemented with @overload_attribute
+    def resolve_real(self, ary):
+        return self._resolve_real_imag(ary, attr='real')
+
+    def resolve_imag(self, ary):
+        return self._resolve_real_imag(ary, attr='imag')
+
+    def _resolve_real_imag(self, ary, attr):
+        if ary.dtype in types.complex_domain:
+            return ary.copy(dtype=ary.dtype.underlying_float, layout='A')
+        elif ary.dtype in types.number_domain:
+            res = ary.copy(dtype=ary.dtype)
+            if attr == 'imag':
+                res = res.copy(readonly=True)
+            return res
+        else:
+            msg = "cannot access .{} of array of {}"
+            raise TypingError(msg.format(attr, ary.dtype))
 
     @bound_function("array.transpose")
     def resolve_transpose(self, ary, args, kws):

numba/cuda/codegen.py

@@ -80,6 +80,8 @@ def __init__(self, codegen, name, entry_name=None, max_registers=None,
         # Driver API at link time.
         self._linking_files = set()
 
+        # Cache the LLVM IR string
+        self._llvm_strs = None
         # Maps CC -> PTX string
         self._ptx_cache = {}
         # Maps CC -> cubin
@@ -95,8 +97,14 @@ def __init__(self, codegen, name, entry_name=None, max_registers=None,
         self._nvvm_options = nvvm_options
         self._entry_name = entry_name
 
+    @property
+    def llvm_strs(self):
+        if self._llvm_strs is None:
+            self._llvm_strs = [str(mod) for mod in self.modules]
+        return self._llvm_strs
+
     def get_llvm_str(self):
-        return str(self._module)
+        return self.llvm_strs[0]
 
     def get_asm_str(self, cc=None):
         return self._join_ptxes(self._get_ptxes(cc=cc))
@@ -126,7 +134,7 @@ def _get_ptxes(self, cc=None):
                     warn(NumbaInvalidConfigWarning(msg))
             options['debug'] = False
 
-        irs = [str(mod) for mod in self.modules]
+        irs = self.llvm_strs
 
         if options.get('debug', False):
             # If we're compiling with debug, we need to compile modules with
@@ -298,15 +306,15 @@ def _reduce_states(self):
         after deserialization.
         """
         if self._linking_files:
-            msg = ('cannot pickle CUDACodeLibrary function with additional '
-                   'libraries to link against')
+            msg = 'Cannot pickle CUDACodeLibrary with linking files'
             raise RuntimeError(msg)
+        if not self._finalized:
+            raise RuntimeError('Cannot pickle unfinalized CUDACodeLibrary')
         return dict(
             codegen=None,
             name=self.name,
             entry_name=self._entry_name,
-            module=self._module,
-            linking_libraries=self._linking_libraries,
+            llvm_strs=self.llvm_strs,
             ptx_cache=self._ptx_cache,
             cubin_cache=self._cubin_cache,
             linkerinfo_cache=self._linkerinfo_cache,
@@ -315,28 +323,23 @@ def _reduce_states(self):
         )
 
     @classmethod
-    def _rebuild(cls, codegen, name, entry_name, module, linking_libraries,
-                 ptx_cache, cubin_cache, linkerinfo_cache, max_registers,
-                 nvvm_options):
+    def _rebuild(cls, codegen, name, entry_name, llvm_strs, ptx_cache,
+                 cubin_cache, linkerinfo_cache, max_registers, nvvm_options):
         """
         Rebuild an instance.
         """
-        instance = cls.__new__(cls)
-        super(cls, instance).__init__(codegen, name)
-        instance._entry_name = entry_name
-
-        instance._module = module
-        instance._linking_libraries = linking_libraries
-        instance._linking_files = set()
+        instance = cls(codegen, name, entry_name=entry_name)
 
+        instance._llvm_strs = llvm_strs
         instance._ptx_cache = ptx_cache
         instance._cubin_cache = cubin_cache
         instance._linkerinfo_cache = linkerinfo_cache
-        instance._cufunc_cache = {}
 
         instance._max_registers = max_registers
         instance._nvvm_options = nvvm_options
 
+        instance._finalized = True
+
         return instance
 
 

numba/cuda/cudadrv/devicearray.py

@@ -812,7 +812,7 @@ def array_core(ary):
     a view without the repeated dimensions.
 
     """
-    if not ary.strides:
+    if not ary.strides or not ary.size:
         return ary
     core_index = []
     for stride in ary.strides:

numba/cuda/simulator/cudadrv/devicearray.py

@@ -233,7 +233,7 @@ def array_core(ary):
     a view without the repeated dimensions.
 
     """
-    if not ary.strides:
+    if not ary.strides or not ary.size:
         return ary
     core_index = []
     for stride in ary.strides:

numba/cuda/tests/cudadrv/test_cuda_ndarray.py

@@ -172,10 +172,14 @@ def test_devicearray_view_bad_not_c_contig(self):
         array = cuda.to_device(original)[:, ::2]
         with self.assertRaises(ValueError) as e:
             array.view("i4")
-        self.assertEqual(
-            "To change to a dtype of a different size,"
-            " the array must be C-contiguous",
-            str(e.exception))
+
+        msg = str(e.exception)
+        self.assertIn('To change to a dtype of a different size,', msg)
+
+        contiguous_pre_np123 = 'the array must be C-contiguous' in msg
+        contiguous_post_np123 = 'the last axis must be contiguous' in msg
+        self.assertTrue(contiguous_pre_np123 or contiguous_post_np123,
+                        'Expected message to mention contiguity')
 
     def test_devicearray_view_bad_itemsize(self):
         original = np.array(np.arange(12), dtype="i2").reshape(4, 3)