rm ar6 gp

.github/workflows/build-macos.yml

@@ -82,7 +82,6 @@ jobs:
                   USE_CUDA: 0
 
             - name: Test
-              if: (matrix.python-version != '3.12')  # matplotlib still depends on deprecated distutils in the new python 3.12
               run: |
                   python -m pip install --upgrade pytest
                   python -m pip install --upgrade pytest-cov

.github/workflows/deploy-conda.yml

@@ -72,17 +72,33 @@ jobs:
             # does not need manylinux docker, thuns, CUDA can be installed in
             # the github action's virtual machine using the Jimver's action
             # in the previous action (see above).
-            - name: Build wheels in windows with CUDA support
+            - name: Build wheels in windows with CUDA support (X86_64)
               if: matrix.os == 'windows-latest'
               uses: pypa/[email protected]
               env:
-                  CIBW_ARCHS_WINDOWS: "AMD64 x86 ARM64"
-                  CIBW_BUILD: "*-win_amd64 *-win_arm64"
+                  CIBW_ARCHS_WINDOWS: "AMD64 x86"
+                  CIBW_BUILD: "*-win_amd64"
                   CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
                   CIBW_BUILD_VERBOSITY: 1
                   CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=1 CUDA_DYNAMIC_LOADING=0"
               with:
                 output-dir: wheelhouse
+
+            # Build wheel for windows with CUDA enabled. Note that cibuildwheel
+            # does not need manylinux docker, thuns, CUDA can be installed in
+            # the github action's virtual machine using the Jimver's action
+            # in the previous action (see above).
+            - name: Build wheels in windows without CUDA support (ARM64)
+              if: matrix.os == 'windows-latest'
+              uses: pypa/[email protected]
+              env:
+                  CIBW_ARCHS_WINDOWS: "ARM64"
+                  CIBW_BUILD: "*-win_arm64"
+                  CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
+                  CIBW_BUILD_VERBOSITY: 1
+                  CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=0 CUDA_DYNAMIC_LOADING=0"
+              with:
+                output-dir: wheelhouse
 
             # Do not enable compiling with CUDA for macos since NIVDIA no
             # longer supports CUDA in the macos.
@@ -111,20 +127,48 @@ jobs:
             # anaconda cloud upload size is 1GB, the large wheel size will not
             # be a problem. The advantage of bundling cuda libraries to the
             # wheel is that the end-user does not have to install cuda toolkit.
-            - name: Build wheels in linux with CUDA support
+            - name: Build wheels in linux with CUDA support (X86_64)
               if: matrix.os == 'ubuntu-latest'
               uses: pypa/[email protected]
               env:
                   CIBW_MANYLINUX_X86_64_IMAGE: sameli/manylinux2014_x86_64_cuda_12.2
                   # CIBW_MANYLINUX_X86_64_IMAGE: manylinux2014
                   # CIBW_BEFORE_ALL_LINUX: chmod +x .github/scripts/install_cuda.sh && .github/scripts/install_cuda.sh
-                  CIBW_ARCHS_LINUX: "x86_64 aarch64"
-                  CIBW_BUILD: "*-manylinux_x86_64 *-manylinux_aarch64"
+                  CIBW_ARCHS_LINUX: "x86_64"
+                  CIBW_BUILD: "*-manylinux_x86_64"
                   CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
                   CIBW_BUILD_VERBOSITY: 1
                   CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=1 CUDA_DYNAMIC_LOADING=0 CUDA_HOME=/usr/local/cuda"
               with:
                 output-dir: wheelhouse
+
+            # Enable compiling with CUDA, only for linux. Note that cuda must
+            # be installed "inside" the manylinux docker image, not in the main
+            # github virtual machine. Also, we should use manylinux2014, but
+            # not manylinux1. Because of this, we will not install CUDA with
+            # the Jimver's action (unlike we did so for the windows wheel in
+            # the above). The ``CUDA_DYNAMIC_LOADING=0`` disables dynamic
+            # loading of CUDA library. Base of this, when the wheels are
+            # created in manylinux platform, the cuda's libraries
+            # (libcudart.so, libcublas.so, libcusparse.so) will bundle to
+            # the wheel. This is increases the wheel size, bit since the
+            # anaconda cloud upload size is 1GB, the large wheel size will not
+            # be a problem. The advantage of bundling cuda libraries to the
+            # wheel is that the end-user does not have to install cuda toolkit.
+            - name: Build wheels in linux without CUDA support (AARCH64)
+              if: matrix.os == 'ubuntu-latest'
+              uses: pypa/[email protected]
+              env:
+                  # CIBW_MANYLINUX_X86_64_IMAGE: sameli/manylinux2014_x86_64_cuda_12.2
+                  CIBW_MANYLINUX_X86_64_IMAGE: manylinux2014
+                  # CIBW_BEFORE_ALL_LINUX: chmod +x .github/scripts/install_cuda.sh && .github/scripts/install_cuda.sh
+                  CIBW_ARCHS_LINUX: "aarch64"
+                  CIBW_BUILD: "*-manylinux_aarch64"
+                  CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
+                  CIBW_BUILD_VERBOSITY: 1
+                  CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=0 CUDA_DYNAMIC_LOADING=0 CUDA_HOME=/usr/local/cuda"
+              with:
+                output-dir: wheelhouse
 
             - name: Upload artifacts
               uses: actions/upload-artifact@v3
@@ -147,7 +191,7 @@ jobs:
         # To publish to pypi on each GitHub Release, use the following:
         # if: github.event_name == 'release' && github.event.action == 'published'
         steps:
-            
+
             - name: Clean runner space
               if: matrix.os == 'ubuntu-latest'
               run: |

.github/workflows/deploy-docs.yml

@@ -19,7 +19,7 @@ jobs:
             - name: Set up Python
               uses: actions/setup-python@v4
               with:
-                  python-version: '3.11'
+                python-version: '3.11'
 
             # Build doxygen files in /docs/doxygen/output/html
             - name: Build doxygen

.github/workflows/deploy-pypi.yml

@@ -1,9 +1,9 @@
 name: deploy-pypi
 
 on:
-    # push:
-    #     branches:
-    #         - main
+    push:
+        branches:
+            - main
     release:
         types:
             - published
@@ -79,17 +79,33 @@ jobs:
             # does not need manylinux docker, thuns, CUDA can be installed in
             # the github action's virtual machine using the Jimver's action
             # in the previous action (see above).
-            - name: Build wheels in windows with CUDA support
+            - name: Build wheels in windows with CUDA support (X86_64)
               if: matrix.os == 'windows-latest'
               uses: pypa/[email protected]
               env:
-                  CIBW_ARCHS_WINDOWS: "AMD64 x86 ARM64"
-                  CIBW_BUILD: "*-win_amd64 *-win_arm64"
+                  CIBW_ARCHS_WINDOWS: "AMD64 x86"
+                  CIBW_BUILD: "*-win_amd64"
                   CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
                   CIBW_BUILD_VERBOSITY: 1
                   CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=1 CUDA_DYNAMIC_LOADING=1"
               with:
                 output-dir: wheelhouse
+
+            # Build wheel for windows with CUDA enabled. Note that cibuildwheel
+            # does not need manylinux docker, thuns, CUDA can be installed in
+            # the github action's virtual machine using the Jimver's action
+            # in the previous action (see above).
+            - name: Build wheels in windows without CUDA support (ARM64)
+              if: matrix.os == 'windows-latest'
+              uses: pypa/[email protected]
+              env:
+                  CIBW_ARCHS_WINDOWS: "ARM64"
+                  CIBW_BUILD: "*-win_arm64"
+                  CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
+                  CIBW_BUILD_VERBOSITY: 1
+                  CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=0 CUDA_DYNAMIC_LOADING=1"
+              with:
+                output-dir: wheelhouse
 
             # Do not enable compiling with CUDA for macos since NIVDIA no
             # longer supports CUDA in the macos.
@@ -118,21 +134,50 @@ jobs:
             # increase (cuda libs are large), and the  manylinux wheel can be
             # uploaded to pypi (pypi has 100MB upload limit). On the down side,
             # the user has to install cuda library themselves.
-            - name: Build wheels in linux with CUDA support
+            - name: Build wheels in linux with CUDA support (X86_64)
               if: matrix.os == 'ubuntu-latest'
               uses: pypa/[email protected]
               env:
                   CIBW_MANYLINUX_X86_64_IMAGE: sameli/manylinux2014_x86_64_cuda_12.2
                   # CIBW_MANYLINUX_X86_64_IMAGE: manylinux2014
                   # CIBW_BEFORE_ALL_LINUX: chmod +x .github/scripts/install_cuda.sh && .github/scripts/install_cuda.sh
-                  CIBW_ARCHS_LINUX: "x86_64 aarch64"
-                  CIBW_BUILD: "*-manylinux_x86_64 *-manylinux_aarch64"
+                  CIBW_ARCHS_LINUX: "x86_64"
+                  CIBW_BUILD: "*-manylinux_x86_64"
                   CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
                   CIBW_BUILD_VERBOSITY: 1
                   CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=1 CUDA_DYNAMIC_LOADING=1 CUDA_HOME=/usr/local/cuda"
                   # CIBW_BEFORE_BUILD_LINUX: yum update; yum install gcc-gfortran openblas-devel.x86_64 lapack-devel.x86_64 -y;
               with:
                 output-dir: wheelhouse
+
+            # Enable compiling with CUDA, only for linux. Note that cuda must
+            # be installed "inside" the manylinux docker image, not in the main
+            # github virtual machine. Also, we should use manylinux2014, but
+            # not manylinux1. Because of this, we will not install CUDA with
+            # the Jimver's action (unlike we did so for the windows wheel in
+            # the above). The ``CUDA_DYNAMIC_LOADING=1`` enables dynamic
+            # loading of CUDA library. Base of this, when the wheels are
+            # created in manylinux platform, the cuda's libraries
+            # (libcudart.so, libcublas.so, libcusparse.so) will NOT bundle to
+            # the wheel. This is very good since the size of the wheel does not
+            # increase (cuda libs are large), and the  manylinux wheel can be
+            # uploaded to pypi (pypi has 100MB upload limit). On the down side,
+            # the user has to install cuda library themselves.
+            - name: Build wheels in linux without CUDA support (AARCH64)
+              if: matrix.os == 'ubuntu-latest'
+              uses: pypa/[email protected]
+              env:
+                  # CIBW_MANYLINUX_X86_64_IMAGE: sameli/manylinux2014_x86_64_cuda_12.2
+                  CIBW_MANYLINUX_X86_64_IMAGE: manylinux2014
+                  # CIBW_BEFORE_ALL_LINUX: chmod +x .github/scripts/install_cuda.sh && .github/scripts/install_cuda.sh
+                  CIBW_ARCHS_LINUX: "aarch64"
+                  CIBW_BUILD: "*-manylinux_aarch64"
+                  CIBW_SKIP: "pp* cp36-* cp37-* cp38-*"
+                  CIBW_BUILD_VERBOSITY: 1
+                  CIBW_ENVIRONMENT: "USE_LONG_INT=0 USE_UNSIGNED_LONG_INT=0 USE_CBLAS=0 USE_CUDA=0 CUDA_DYNAMIC_LOADING=1 CUDA_HOME=/usr/local/cuda"
+                  # CIBW_BEFORE_BUILD_LINUX: yum update; yum install gcc-gfortran openblas-devel.x86_64 lapack-devel.x86_64 -y;
+              with:
+                output-dir: wheelhouse
 
             - name: Upload artifacts
               uses: actions/upload-artifact@v3

README.rst

@@ -62,7 +62,7 @@ Successful installation and tests performed on the following operating systems,
 +          +-------------------+--------+-------+-------+-------+-------+-------+-------+-------+                 +
 |          | AARCH-64          | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |                 |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
-|          |                   | GPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |                 |
+|          |                   | GPU    |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |                 |
 +----------+-------------------+--------+-------+-------+-------+-------+-------+-------+-------+-----------------+
 | macOS    | X86-64            | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  | |build-macos|   |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
@@ -78,7 +78,7 @@ Successful installation and tests performed on the following operating systems,
 +          +-------------------+--------+-------+-------+-------+-------+-------+-------+-------+                 +
 |          | ARM-64            | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  |                 |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
-|          |                   | GPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  |                 |
+|          |                   | GPU    |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |                 |
 +----------+-------------------+--------+-------+-------+-------+-------+-------+-------+-------+-----------------+
 
 .. |build-linux| image:: https://img.shields.io/github/actions/workflow/status/ameli/imate/build-linux.yml

conda-recipe/conda_build_config.yaml

@@ -2,3 +2,4 @@ python:
     - 3.9
     - 3.10
     - 3.11
+    - 3.12

docs/notes.rst

@@ -132,9 +132,6 @@ Fast Trace Estimator
 TODO
 ====
 
-* Python 3.12 completely removes distutils. The distutils package should be
-  replaced with setuptools.
-
 * Implement ``keep`` functionality for slq method.
 * Hutchinson method can be implemented in C++ and also in CUDA on GPU.
 * Other functions (besides traceinv and logdet)

docs/source/index.rst

@@ -93,7 +93,7 @@ Successful installation and tests performed on the following operating systems,
 +          +-------------------+--------+-------+-------+-------+-------+-------+-------+-------+                 +
 |          | AARCH-64          | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |                 |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
-|          |                   | GPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |  |y|  |                 |
+|          |                   | GPU    |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |                 |
 +----------+-------------------+--------+-------+-------+-------+-------+-------+-------+-------+-----------------+
 | macOS    | X86-64            | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  | |build-macos|   |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
@@ -109,7 +109,7 @@ Successful installation and tests performed on the following operating systems,
 +          +-------------------+--------+-------+-------+-------+-------+-------+-------+-------+                 +
 |          | ARM-64            | CPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  |                 |
 +          +                   +--------+-------+-------+-------+-------+-------+-------+-------+                 +
-|          |                   | GPU    |  |y|  |  |y|  |  |y|  |  |y|  |  |n|  |  |n|  |  |n|  |                 |
+|          |                   | GPU    |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |  |n|  |                 |
 +----------+-------------------+--------+-------+-------+-------+-------+-------+-------+-------+-----------------+
 
 .. |build-linux| image:: https://img.shields.io/github/actions/workflow/status/ameli/imate/build-linux.yml

docs/source/install/dependencies.rst

@@ -20,7 +20,7 @@ OpenMP (`Required`)
 
     For **macOS** users:
 
-        It's crucial to note that OpenMP is not part of the default Apple Xcode's LLVM compiler. Even if you have Apple Xcode LLVM compiler readily installed on macOS, you will still need to install OpenMP separately via `libomp` Homebrew package (see below) or as part of the *open source* `LLVM compiler <https://llvm.org/>`__, via `llvm` Homebrew package.
+        It's crucial to note that OpenMP is not part of the default Apple Xcode's LLVM compiler. Even if you have Apple Xcode LLVM compiler readily installed on macOS, you will still need to install OpenMP separately via ``libomp`` Homebrew package (see below) or as part of the *open source* `LLVM compiler <https://llvm.org/>`__, via ``llvm`` Homebrew package.
 
     For **Windows** users:
 

docs/source/install/troubleshooting.rst

@@ -17,7 +17,7 @@ Upon importing |project|, you may encounter this error:
 	.../site-packages/imate/logdet/../.dylibs/libomp.dylib:
 	cannot load 'libomp.dylib' (load command 0x80000034 is unknown)
 
-This message indicates that OpenMP is not installed on your system, a requirement for the |project|. To resolve this issue, refer to `OpenMP installation instructions <dependencies_openmp>`.
+This message indicates that OpenMP is not installed on your system, a requirement for the |project|. To resolve this issue, refer to :ref:`OpenMP installation instructions <dependencies_openmp>`.
 
 Issue of Initializing ``libomp``
 ================================

examples/_utilities/plot_utilities.py

@@ -21,7 +21,7 @@
 from mpl_toolkits.axes_grid1.inset_locator import inset_axes       # noqa: F401
 from mpl_toolkits.axes_grid1.inset_locator import InsetPosition    # noqa: F401
 from mpl_toolkits.axes_grid1.inset_locator import mark_inset       # noqa: F401
-from distutils.spawn import find_executable
+import shutil
 from .display_utilities import is_notebook
 
 # Check DISPLAY
@@ -71,7 +71,7 @@ def load_plot_settings(font_scale=None):
     # sns.set()
 
     # LaTeX
-    if find_executable('latex'):
+    if shutil.which('latex'):
         try:
             # plt.rc('text', usetex=True)
             matplotlib.rcParams['text.usetex'] = True

imate/__version__.py

@@ -1 +1 @@
-__version__ = "0.20.12"
+__version__ = "0.20.13"

imate/_utilities/plot_utilities.py

@@ -24,7 +24,7 @@
 from matplotlib.ticker import FormatStrFormatter, FuncFormatter    # noqa: F401
 from mpl_toolkits.axes_grid1 import make_axes_locatable            # noqa: F401
 
-from distutils.spawn import find_executable
+import shutil
 from .display_utilities import is_notebook
 import logging
 import warnings
@@ -72,7 +72,7 @@ def load_plot_settings():
     # sns.set()
 
     # LaTeX
-    if find_executable('latex'):
+    if shutil.which('latex'):
         try:
             # plt.rc('text',usetex=True)
             matplotlib.rcParams['text.usetex'] = True