numba-mpi 0.41__tar.gz → 0.43__tar.gz

{numba_mpi-0.41 → numba_mpi-0.43}/.github/workflows/readme_snippets.yml

@@ -23,12 +23,14 @@ jobs:
       - uses: mpi4py/setup-mpi@v1
       - run: pip install -e .
       - run: pip install pytest-codeblocks pytest matplotlib
+      - run: sudo apt-get install librsvg2-bin
       - run: python -c "import pytest_codeblocks; code=pytest_codeblocks.extract_from_file('README.md'); f=open('readme.py', 'w'); f.writelines(block.code for block in code if block.syntax=='python'); f.close()"
       - run: cat -n readme.py
       - run: mpiexec -n 4 python -We readme.py
+      - run: rsvg-convert readme_plot.svg > readme_plot.png
       - uses: actions/upload-artifact@v3
         with:
-          path: readme_plot.png
+          path: readme_plot.*
           name: readme_plot
       - if: github.ref == 'refs/heads/main'
         uses: eine/tip@master

{numba_mpi-0.41 → numba_mpi-0.43}/.github/workflows/tests+pypi.yml

@@ -88,6 +88,7 @@ jobs:
     needs: [pylint, precommit, pdoc, zenodo_json]
     strategy:
       matrix:
+        mpi-np: [2, 3]
         platform: [ubuntu-latest, macos-12, windows-latest]
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
         mpi: [ 'mpich', 'openmpi', 'msmpi', 'intelmpi']
@@ -130,7 +131,7 @@ jobs:
       - run: pip install -e .[tests]
       - run: python -We -c "import mpi4py"
       - run: python -We -c "import numba_mpi"
-      - run: mpiexec -n 2 pytest --durations=10 -p no:unraisableexception -We
+      - run: mpiexec -n ${{ matrix.mpi-np }} pytest --durations=10 -p no:unraisableexception -We
 
   dist:
     runs-on: ubuntu-latest

numba_mpi-0.43/CODE_OF_CONDUCT.md

@@ -0,0 +1,47 @@
+# Code of conduct for the numba-mpi open-source project
+
+As [contributors and maintainers of this project](https://github.com/orgs/numba-mpi/people),
+and in the interest of fostering an open and welcoming community, we pledge to respect all
+people who contribute through reporting issues, posting feature requests, updating 
+documentation, submitting pull requests or patches, and other activities.
+
+We are committed to making participation in this project a harassment-free experience for
+everyone, regardless of level of technical experience and regardless of any non-technical
+personal characteristic or identity trait (gender, religion, physicality, age, ethnicity,
+essentially **any**).
+
+Nurturing open and unemotional code reviews and community discussions, we aim at ensuring
+fruitful and enriching collaboration experience and maintaining quality engineering
+standards. This Code of Conduct applies both within project spaces and in public spaces
+when an individual is representing the project or its community.
+
+Examples of unacceptable behavior by participants include:
+
+* Breaching project security (e.g., granting GitHub access rights without ensuring
+  collaborators' consent);
+* Breaching collaborators' privacy (e.g., publishing other's private information without
+  permission);
+* Force developments (e.g., merging or releasing against expressed collaborators' comments
+  against doing it);
+* Any form of harassing language of exclusion;
+* Other unethical or unprofessional conduct (if in doubt, ask!).
+
+Project maintainers have the right and responsibility to remove, edit, or reject comments,
+commits, code, issues, and other contributions that are not aligned to this Code of Conduct,
+or to ban temporarily or permanently any contributor for other behaviors that they deem
+inappropriate, threatening, offensive, or harmful. By adopting this Code of Conduct, project
+maintainers commit themselves to fairly, consistently and collaboratively applying these
+principles to every aspect of managing this project. 
+
+Reporting actions that are violating the hereby Code of Conduct can be done publicly on the
+project GitHub space, or if needed, can be reported directly to any of the
+[project maintainers listed on GitHub](https://github.com/orgs/numba-mpi/people)
+(as of the time of writing: Sylwester Arabas, David Zwicker, Kacper Derlatka, et al.).
+Maintainers are obligated to maintain confidentiality with regard to the reporter of a
+privately reported incident.
+
+Please note that, as of time of writing, the entirety of the project team is engaged in the
+development purely on voluntary basis.
+
+--
+This Code of Conduct was inspired by [Numba Code of Conduct](https://github.com/numba/numba-governance/blob/accepted/code-of-conduct.md).

{numba_mpi-0.41 → numba_mpi-0.43}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: numba-mpi
-Version: 0.41
+Version: 0.43
 Summary: Numba @njittable MPI wrappers tested on Linux, macOS and Windows
 Home-page: https://github.com/numba-mpi/numba-mpi
 Author: https://github.com/numba-mpi/numba-mpi/graphs/contributors
@@ -8,6 +8,7 @@ License: GPL v3
 Project-URL: Tracker, https://github.com/numba-mpi/numba-mpi/issues
 Project-URL: Documentation, https://numba-mpi.github.io/numba-mpi
 Project-URL: Source, https://github.com/numba-mpi/numba-mpi
+Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: numba
@@ -79,40 +80,46 @@ hello()
 
 ### Example comparing numba-mpi vs. mpi4py performance:
 
-The example below compares Numba + mpi4py vs. Numba + numba-mpi performance.
-The sample code estimates $\pi$ by integration of $4/(1+x^2)$ between 0 and 1
+The example below compares `Numba`+`mpi4py` vs. `Numba`+`numba-mpi` performance.
+The sample code estimates $\pi$ by numerical integration of $\int_0^1 (4/(1+x^2))dx=\pi$ 
 dividing the workload into `n_intervals` handled by separate MPI processes 
-and then obtaining a sum using `allreduce`.
-The computation is carried out in a JIT-compiled function and is repeated
-`N_TIMES`, the repetitions and the MPI-handled reduction are done outside or 
-inside of the JIT-compiled block for mpi4py and numba-mpi, respectively.
+and then obtaining a sum using `allreduce` (see, e.g., analogous [Matlab docs example](https://www.mathworks.com/help/parallel-computing/numerical-estimation-of-pi-using-message-passing.html)).
+The computation is carried out in a JIT-compiled function `get_pi_part()` and is repeated
+`N_TIMES`. The repetitions and the MPI-handled reduction are done outside or 
+inside of the JIT-compiled block for `mpi4py` and `numba-mpi`, respectively.
 Timing is repeated `N_REPEAT` times and the minimum time is reported.
-The generated plot shown below depicts the speedup obtained by replacing mpi4py
-with numba_mpi as a function of `n_intervals` - the more often communication
-is needed (smaller `n_intervals`), the larger the expected speedup.
-
+The generated plot shown below depicts the speedup obtained by replacing `mpi4py`
+with `numba_mpi`, plotted as a function of `N_TIMES / n_intervals` - the number of MPI calls per 
+interval. The speedup, which stems from avoiding roundtrips between JIT-compiled
+and Python code is significant (150%-300%) in all cases. The more often communication
+is needed (smaller `n_intervals`), the larger the measured speedup. Note that nothing 
+in the actual number crunching (within the `get_pi_part()` function) or in the employed communication logic
+(handled by the same MPI library) differs between the `mpi4py` or `numba-mpi` solutions.
+These are the overhead of `mpi4py` higher-level abstractions and the overhead of 
+repeatedly entering and leaving the JIT-compiled block if using `mpi4py`, which can be
+eliminated by using `numba-mpi`, and which the measured differences in execution time
+stem from.
 ```python
 import timeit, mpi4py, numba, numpy as np, numba_mpi
 
 N_TIMES = 10000
-N_REPEAT = 10
 RTOL = 1e-3
 
-@numba.njit
-def get_pi_part(out, n_intervals, rank, size):
+@numba.jit
+def get_pi_part(n_intervals=1000000, rank=0, size=1):
     h = 1 / n_intervals
     partial_sum = 0.0
     for i in range(rank + 1, n_intervals, size):
         x = h * (i - 0.5)
         partial_sum += 4 / (1 + x**2)
-    out[0] = h * partial_sum
+    return h * partial_sum
 
-@numba.njit
+@numba.jit
 def pi_numba_mpi(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, numba_mpi.rank(), numba_mpi.size())
+        part[0] = get_pi_part(n_intervals, numba_mpi.rank(), numba_mpi.size())
         numba_mpi.allreduce(part, pi, numba_mpi.Operator.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
@@ -120,30 +127,30 @@ def pi_mpi4py(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
+        part[0] = get_pi_part(n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
         mpi4py.MPI.COMM_WORLD.Allreduce(part, (pi, mpi4py.MPI.DOUBLE), op=mpi4py.MPI.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
-plot_x = [1000 * k for k in range(1, 11)]
+plot_x = [x for x in range(1, 11)]
 plot_y = {'numba_mpi': [], 'mpi4py': []}
-for n_intervals in plot_x:
+for x in plot_x:
     for impl in plot_y:
         plot_y[impl].append(min(timeit.repeat(
-            f"pi_{impl}({n_intervals})",
+            f"pi_{impl}(n_intervals={N_TIMES // x})",
             globals=locals(),
             number=1,
-            repeat=N_REPEAT
+            repeat=10
         )))
 
 if numba_mpi.rank() == 0:
     from matplotlib import pyplot
     pyplot.figure(figsize=(8.3, 3.5), tight_layout=True)
     pyplot.plot(plot_x, np.array(plot_y['mpi4py'])/np.array(plot_y['numba_mpi']), marker='o')
-    pyplot.xlabel('n_intervals (workload in between communication)')
-    pyplot.ylabel('wall time ratio (mpi4py / numba_mpi)')
+    pyplot.xlabel('number of MPI calls per interval')
+    pyplot.ylabel('mpi4py/numba-mpi wall-time ratio')
     pyplot.title(f'mpiexec -np {numba_mpi.size()}')
     pyplot.grid()
-    pyplot.savefig('readme_plot.png')
+    pyplot.savefig('readme_plot.svg')
 ```
 
 ![plot](https://github.com/numba-mpi/numba-mpi/releases/download/tip/readme_plot.png)

{numba_mpi-0.41 → numba_mpi-0.43}/README.md

@@ -60,40 +60,46 @@ hello()
 
 ### Example comparing numba-mpi vs. mpi4py performance:
 
-The example below compares Numba + mpi4py vs. Numba + numba-mpi performance.
-The sample code estimates $\pi$ by integration of $4/(1+x^2)$ between 0 and 1
+The example below compares `Numba`+`mpi4py` vs. `Numba`+`numba-mpi` performance.
+The sample code estimates $\pi$ by numerical integration of $\int_0^1 (4/(1+x^2))dx=\pi$ 
 dividing the workload into `n_intervals` handled by separate MPI processes 
-and then obtaining a sum using `allreduce`.
-The computation is carried out in a JIT-compiled function and is repeated
-`N_TIMES`, the repetitions and the MPI-handled reduction are done outside or 
-inside of the JIT-compiled block for mpi4py and numba-mpi, respectively.
+and then obtaining a sum using `allreduce` (see, e.g., analogous [Matlab docs example](https://www.mathworks.com/help/parallel-computing/numerical-estimation-of-pi-using-message-passing.html)).
+The computation is carried out in a JIT-compiled function `get_pi_part()` and is repeated
+`N_TIMES`. The repetitions and the MPI-handled reduction are done outside or 
+inside of the JIT-compiled block for `mpi4py` and `numba-mpi`, respectively.
 Timing is repeated `N_REPEAT` times and the minimum time is reported.
-The generated plot shown below depicts the speedup obtained by replacing mpi4py
-with numba_mpi as a function of `n_intervals` - the more often communication
-is needed (smaller `n_intervals`), the larger the expected speedup.
-
+The generated plot shown below depicts the speedup obtained by replacing `mpi4py`
+with `numba_mpi`, plotted as a function of `N_TIMES / n_intervals` - the number of MPI calls per 
+interval. The speedup, which stems from avoiding roundtrips between JIT-compiled
+and Python code is significant (150%-300%) in all cases. The more often communication
+is needed (smaller `n_intervals`), the larger the measured speedup. Note that nothing 
+in the actual number crunching (within the `get_pi_part()` function) or in the employed communication logic
+(handled by the same MPI library) differs between the `mpi4py` or `numba-mpi` solutions.
+These are the overhead of `mpi4py` higher-level abstractions and the overhead of 
+repeatedly entering and leaving the JIT-compiled block if using `mpi4py`, which can be
+eliminated by using `numba-mpi`, and which the measured differences in execution time
+stem from.
 ```python
 import timeit, mpi4py, numba, numpy as np, numba_mpi
 
 N_TIMES = 10000
-N_REPEAT = 10
 RTOL = 1e-3
 
-@numba.njit
-def get_pi_part(out, n_intervals, rank, size):
+@numba.jit
+def get_pi_part(n_intervals=1000000, rank=0, size=1):
     h = 1 / n_intervals
     partial_sum = 0.0
     for i in range(rank + 1, n_intervals, size):
         x = h * (i - 0.5)
         partial_sum += 4 / (1 + x**2)
-    out[0] = h * partial_sum
+    return h * partial_sum
 
-@numba.njit
+@numba.jit
 def pi_numba_mpi(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, numba_mpi.rank(), numba_mpi.size())
+        part[0] = get_pi_part(n_intervals, numba_mpi.rank(), numba_mpi.size())
         numba_mpi.allreduce(part, pi, numba_mpi.Operator.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
@@ -101,30 +107,30 @@ def pi_mpi4py(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
+        part[0] = get_pi_part(n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
         mpi4py.MPI.COMM_WORLD.Allreduce(part, (pi, mpi4py.MPI.DOUBLE), op=mpi4py.MPI.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
-plot_x = [1000 * k for k in range(1, 11)]
+plot_x = [x for x in range(1, 11)]
 plot_y = {'numba_mpi': [], 'mpi4py': []}
-for n_intervals in plot_x:
+for x in plot_x:
     for impl in plot_y:
         plot_y[impl].append(min(timeit.repeat(
-            f"pi_{impl}({n_intervals})",
+            f"pi_{impl}(n_intervals={N_TIMES // x})",
             globals=locals(),
             number=1,
-            repeat=N_REPEAT
+            repeat=10
         )))
 
 if numba_mpi.rank() == 0:
     from matplotlib import pyplot
     pyplot.figure(figsize=(8.3, 3.5), tight_layout=True)
     pyplot.plot(plot_x, np.array(plot_y['mpi4py'])/np.array(plot_y['numba_mpi']), marker='o')
-    pyplot.xlabel('n_intervals (workload in between communication)')
-    pyplot.ylabel('wall time ratio (mpi4py / numba_mpi)')
+    pyplot.xlabel('number of MPI calls per interval')
+    pyplot.ylabel('mpi4py/numba-mpi wall-time ratio')
     pyplot.title(f'mpiexec -np {numba_mpi.size()}')
     pyplot.grid()
-    pyplot.savefig('readme_plot.png')
+    pyplot.savefig('readme_plot.svg')
 ```
 
 ![plot](https://github.com/numba-mpi/numba-mpi/releases/download/tip/readme_plot.png)

{numba_mpi-0.41 → numba_mpi-0.43}/numba_mpi/__init__.py

@@ -1,4 +1,6 @@
-""" Numba @njittable MPI wrappers tested on Linux, macOS and Windows """
+"""
+.. include::../README.md
+"""
 
 from importlib.metadata import PackageNotFoundError, version
 

{numba_mpi-0.41 → numba_mpi-0.43}/numba_mpi/api/requests.py

@@ -34,6 +34,9 @@ def wait(request):
     """Wrapper for MPI_Wait. Returns integer status code (0 == MPI_SUCCESS).
     Status is currently not handled. Requires 'request' parameter to be a
     c-style pointer to MPI_Request (such as returned by 'isend'/'irecv').
+
+    Uninitialized contents of 'request' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
 
     status_buffer = create_status_buffer()
@@ -64,6 +67,9 @@ def waitall(requests):
     """Wrapper for MPI_Waitall. Returns integer status code (0 == MPI_SUCCESS).
     Status is currently not handled. Requires 'requests' parameter to be an
     array or tuple of MPI_Request objects.
+
+    Uninitialized contents of 'requests' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
     if isinstance(requests, np.ndarray):
         return _waitall_array_impl(requests)
@@ -123,6 +129,9 @@ def waitany(requests):
     status; second - the index of request that was completed. Status is
     currently not handled. Requires 'requests' parameter to be an array
     or tuple of MPI_Request objects.
+
+    Uninitialized contents of 'requests' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
 
     if isinstance(requests, np.ndarray):
@@ -167,6 +176,9 @@ def test(request):
     flag that indicates whether given request is completed. Status is currently
     not handled. Requires 'request' parameter to be a c-style pointer to
     MPI_Request (such as returned by 'isend'/'irecv').
+
+    Uninitialized contents of 'request' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
 
     status_buffer = create_status_buffer()
@@ -203,6 +215,9 @@ def testall(requests):
     flag that indicates whether given request is completed. Status is currently
     not handled. Requires 'requests' parameter to be an array or tuple of
     MPI_Request objects.
+
+    Uninitialized contents of 'requests' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
     if isinstance(requests, np.ndarray):
         return _testall_array_impl(requests)
@@ -269,6 +284,9 @@ def testany(requests):
     that indicates whether any of requests is completed, and index of request
     that is guaranteed to be completed. Requires 'requests' parameter to be an
     array or tuple of MPI_Request objects.
+
+    Uninitialized contents of 'requests' (e.g., from numpy.empty()) may
+    cause invalid pointer dereference and segmentation faults.
     """
 
     if isinstance(requests, np.ndarray):

{numba_mpi-0.41 → numba_mpi-0.43}/numba_mpi.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: numba-mpi
-Version: 0.41
+Version: 0.43
 Summary: Numba @njittable MPI wrappers tested on Linux, macOS and Windows
 Home-page: https://github.com/numba-mpi/numba-mpi
 Author: https://github.com/numba-mpi/numba-mpi/graphs/contributors
@@ -8,6 +8,7 @@ License: GPL v3
 Project-URL: Tracker, https://github.com/numba-mpi/numba-mpi/issues
 Project-URL: Documentation, https://numba-mpi.github.io/numba-mpi
 Project-URL: Source, https://github.com/numba-mpi/numba-mpi
+Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: numba
@@ -79,40 +80,46 @@ hello()
 
 ### Example comparing numba-mpi vs. mpi4py performance:
 
-The example below compares Numba + mpi4py vs. Numba + numba-mpi performance.
-The sample code estimates $\pi$ by integration of $4/(1+x^2)$ between 0 and 1
+The example below compares `Numba`+`mpi4py` vs. `Numba`+`numba-mpi` performance.
+The sample code estimates $\pi$ by numerical integration of $\int_0^1 (4/(1+x^2))dx=\pi$ 
 dividing the workload into `n_intervals` handled by separate MPI processes 
-and then obtaining a sum using `allreduce`.
-The computation is carried out in a JIT-compiled function and is repeated
-`N_TIMES`, the repetitions and the MPI-handled reduction are done outside or 
-inside of the JIT-compiled block for mpi4py and numba-mpi, respectively.
+and then obtaining a sum using `allreduce` (see, e.g., analogous [Matlab docs example](https://www.mathworks.com/help/parallel-computing/numerical-estimation-of-pi-using-message-passing.html)).
+The computation is carried out in a JIT-compiled function `get_pi_part()` and is repeated
+`N_TIMES`. The repetitions and the MPI-handled reduction are done outside or 
+inside of the JIT-compiled block for `mpi4py` and `numba-mpi`, respectively.
 Timing is repeated `N_REPEAT` times and the minimum time is reported.
-The generated plot shown below depicts the speedup obtained by replacing mpi4py
-with numba_mpi as a function of `n_intervals` - the more often communication
-is needed (smaller `n_intervals`), the larger the expected speedup.
-
+The generated plot shown below depicts the speedup obtained by replacing `mpi4py`
+with `numba_mpi`, plotted as a function of `N_TIMES / n_intervals` - the number of MPI calls per 
+interval. The speedup, which stems from avoiding roundtrips between JIT-compiled
+and Python code is significant (150%-300%) in all cases. The more often communication
+is needed (smaller `n_intervals`), the larger the measured speedup. Note that nothing 
+in the actual number crunching (within the `get_pi_part()` function) or in the employed communication logic
+(handled by the same MPI library) differs between the `mpi4py` or `numba-mpi` solutions.
+These are the overhead of `mpi4py` higher-level abstractions and the overhead of 
+repeatedly entering and leaving the JIT-compiled block if using `mpi4py`, which can be
+eliminated by using `numba-mpi`, and which the measured differences in execution time
+stem from.
 ```python
 import timeit, mpi4py, numba, numpy as np, numba_mpi
 
 N_TIMES = 10000
-N_REPEAT = 10
 RTOL = 1e-3
 
-@numba.njit
-def get_pi_part(out, n_intervals, rank, size):
+@numba.jit
+def get_pi_part(n_intervals=1000000, rank=0, size=1):
     h = 1 / n_intervals
     partial_sum = 0.0
     for i in range(rank + 1, n_intervals, size):
         x = h * (i - 0.5)
         partial_sum += 4 / (1 + x**2)
-    out[0] = h * partial_sum
+    return h * partial_sum
 
-@numba.njit
+@numba.jit
 def pi_numba_mpi(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, numba_mpi.rank(), numba_mpi.size())
+        part[0] = get_pi_part(n_intervals, numba_mpi.rank(), numba_mpi.size())
         numba_mpi.allreduce(part, pi, numba_mpi.Operator.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
@@ -120,30 +127,30 @@ def pi_mpi4py(n_intervals):
     pi = np.array([0.])
     part = np.empty_like(pi)
     for _ in range(N_TIMES):
-        get_pi_part(part, n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
+        part[0] = get_pi_part(n_intervals, mpi4py.MPI.COMM_WORLD.rank, mpi4py.MPI.COMM_WORLD.size)
         mpi4py.MPI.COMM_WORLD.Allreduce(part, (pi, mpi4py.MPI.DOUBLE), op=mpi4py.MPI.SUM)
         assert abs(pi[0] - np.pi) / np.pi < RTOL
 
-plot_x = [1000 * k for k in range(1, 11)]
+plot_x = [x for x in range(1, 11)]
 plot_y = {'numba_mpi': [], 'mpi4py': []}
-for n_intervals in plot_x:
+for x in plot_x:
     for impl in plot_y:
         plot_y[impl].append(min(timeit.repeat(
-            f"pi_{impl}({n_intervals})",
+            f"pi_{impl}(n_intervals={N_TIMES // x})",
             globals=locals(),
             number=1,
-            repeat=N_REPEAT
+            repeat=10
         )))
 
 if numba_mpi.rank() == 0:
     from matplotlib import pyplot
     pyplot.figure(figsize=(8.3, 3.5), tight_layout=True)
     pyplot.plot(plot_x, np.array(plot_y['mpi4py'])/np.array(plot_y['numba_mpi']), marker='o')
-    pyplot.xlabel('n_intervals (workload in between communication)')
-    pyplot.ylabel('wall time ratio (mpi4py / numba_mpi)')
+    pyplot.xlabel('number of MPI calls per interval')
+    pyplot.ylabel('mpi4py/numba-mpi wall-time ratio')
     pyplot.title(f'mpiexec -np {numba_mpi.size()}')
     pyplot.grid()
-    pyplot.savefig('readme_plot.png')
+    pyplot.savefig('readme_plot.svg')
 ```
 
 ![plot](https://github.com/numba-mpi/numba-mpi/releases/download/tip/readme_plot.png)

{numba_mpi-0.41 → numba_mpi-0.43}/numba_mpi.egg-info/SOURCES.txt

@@ -1,6 +1,7 @@
 .gitignore
 .pre-commit-config.yaml
 .zenodo.json
+CODE_OF_CONDUCT.md
 LICENSE
 README.md
 setup.py

{numba_mpi-0.41 → numba_mpi-0.43}/setup.py

@@ -18,6 +18,7 @@ setup(
         "local_scheme": lambda _: "",
         "version_scheme": "post-release",
     },
+    python_requires=">=3.8",
     setup_requires=["setuptools_scm"],
     license="GPL v3",
     description="Numba @njittable MPI wrappers tested on Linux, macOS and Windows",

{numba_mpi-0.41 → numba_mpi-0.43}/tests/api/test_isend_irecv.py

@@ -176,7 +176,7 @@ def test_isend_irecv_waitall(isnd, ircv, wall, data_type):
     dst1 = np.empty_like(src1)
     dst2 = np.empty_like(src2)
 
-    reqs = np.empty((2,), dtype=mpi.RequestType)
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
     if mpi.rank() == 0:
         status, reqs[0:1] = isnd(src1, dest=1, tag=11)
         assert status == MPI_SUCCESS
@@ -245,7 +245,7 @@ def test_isend_irecv_waitall_exchange(isnd, ircv, wall):
     src = get_random_array((5,))
     dst = np.empty_like(src)
 
-    reqs = np.empty((2,), dtype=mpi.RequestType)
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
     if mpi.rank() == 0:
         status, reqs[0:1] = isnd(src, dest=1, tag=11)
         assert status == MPI_SUCCESS
@@ -263,6 +263,24 @@ def test_isend_irecv_waitall_exchange(isnd, ircv, wall):
     np.testing.assert_equal(dst, src)
 
 
+@pytest.mark.parametrize(
+    "fun",
+    (
+        jit_waitany.py_func,
+        jit_waitall.py_func,
+        jit_testany.py_func,
+        jit_testall.py_func,
+        jit_waitany,
+        jit_waitall,
+        jit_testany,
+        jit_testall,
+    ),
+)
+def test_wall_segfault(fun):
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
+    fun(reqs)
+
+
 @pytest.mark.parametrize(
     "isnd, ircv, wany, wall",
     [
@@ -282,7 +300,7 @@ def test_isend_irecv_waitany(isnd, ircv, wany, wall, data_type):
     dst1 = np.empty_like(src1)
     dst2 = np.empty_like(src2)
 
-    reqs = np.empty((2,), dtype=mpi.RequestType)
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
     if mpi.rank() == 0:
         status, reqs[0:1] = isnd(src1, dest=1, tag=11)
         assert status == MPI_SUCCESS
@@ -356,7 +374,7 @@ def test_isend_irecv_testall(isnd, ircv, tall, wall):
     dst1 = np.empty_like(src1)
     dst2 = np.empty_like(src2)
 
-    reqs = np.empty((2,), dtype=mpi.RequestType)
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
     if mpi.rank() == 0:
         time.sleep(TEST_WAIT_FULL_IN_SECONDS)
 
@@ -402,7 +420,7 @@ def test_isend_irecv_testany(isnd, ircv, tany, wall):
     dst1 = np.empty_like(src1)
     dst2 = np.empty_like(src2)
 
-    reqs = np.empty((2,), dtype=mpi.RequestType)
+    reqs = np.zeros((2,), dtype=mpi.RequestType)
     if mpi.rank() == 0:
         time.sleep(TEST_WAIT_FULL_IN_SECONDS)