diff options
32 files changed, 1192 insertions, 605 deletions
diff --git a/.appveyor.yml b/.appveyor.yml deleted file mode 100644 index c8c1795c1..000000000 --- a/.appveyor.yml +++ /dev/null @@ -1,150 +0,0 @@ -# As config was originally based on an example by Olivier Grisel. Thanks! -# https://github.com/ogrisel/python-appveyor-demo/blob/master/appveyor.yml -clone_depth: 50 - -# No reason for us to restrict the number concurrent jobs -max_jobs: 100 - -cache: - - '%LOCALAPPDATA%\pip\Cache' - -environment: - global: - MINGW_32: C:\mingw-w64\i686-6.3.0-posix-dwarf-rt_v5-rev1\mingw32\bin - MINGW_64: C:\mingw-w64\x86_64-6.3.0-posix-seh-rt_v5-rev1\mingw64\bin - OPENBLAS_32: https://3f23b170c54c2533c070-1c8a9b3114517dc5fe17b7c3f8c63a43.ssl.cf2.rackcdn.com/openblas-5f998ef_gcc7_1_0_win32.zip - OPENBLAS_64: https://3f23b170c54c2533c070-1c8a9b3114517dc5fe17b7c3f8c63a43.ssl.cf2.rackcdn.com/openblas-5f998ef_gcc7_1_0_win64.zip - APPVEYOR_SAVE_CACHE_ON_ERROR: true - APPVEYOR_SKIP_FINALIZE_ON_EXIT: true - TEST_TIMEOUT: 1000 - NPY_NUM_BUILD_JOBS: 4 - - matrix: - - PYTHON: C:\Python36 - PYTHON_VERSION: 3.6 - PYTHON_ARCH: 32 - TEST_MODE: fast - - - PYTHON: C:\Python37 - PYTHON_VERSION: 3.7 - PYTHON_ARCH: 32 - TEST_MODE: fast - - - PYTHON: C:\Python36-x64 - PYTHON_VERSION: 3.6 - PYTHON_ARCH: 64 - TEST_MODE: full - - - PYTHON: C:\Python37-x64 - PYTHON_VERSION: 3.7 - PYTHON_ARCH: 64 - TEST_MODE: full - -init: - - "ECHO %PYTHON% %PYTHON_VERSION% %PYTHON_ARCH%" - - "ECHO \"%APPVEYOR_SCHEDULED_BUILD%\"" - # If there is a newer build queued for the same PR, cancel this one. - # The AppVeyor 'rollout builds' option is supposed to serve the same - # purpose but it is problematic because it tends to cancel builds pushed - # directly to master instead of just PR builds (or the converse). - # credits: JuliaLang developers. - - ps: if ($env:APPVEYOR_PULL_REQUEST_NUMBER -and $env:APPVEYOR_BUILD_NUMBER -ne ((Invoke-RestMethod ` - https://ci.appveyor.com/api/projects/$env:APPVEYOR_ACCOUNT_NAME/$env:APPVEYOR_PROJECT_SLUG/history?recordsNumber=50).builds | ` - Where-Object pullRequestId -eq $env:APPVEYOR_PULL_REQUEST_NUMBER)[0].buildNumber) { ` - raise "There are newer queued builds for this pull request, skipping build." - } - -install: - # Prepend newly installed Python to the PATH of this build (this cannot be - # done from inside the powershell script as it would require to restart - # the parent CMD process). - - SET PATH=%PYTHON%;%PYTHON%\Scripts;%PATH% - - if [%PYTHON_ARCH%]==[32] SET PATH=%MINGW_32%;%PATH% & SET OPENBLAS=%OPENBLAS_32% - - if [%PYTHON_ARCH%]==[64] SET PATH=%MINGW_64%;%PATH% & SET OPENBLAS=%OPENBLAS_64% - - # Check that we have the expected version and architecture for Python - - python --version - - >- - %CMD_IN_ENV% - python -c "import sys,platform,struct; - print(sys.platform, platform.machine(), struct.calcsize('P') * 8, )" - - # Install "openblas.a" to PYTHON\lib - # Library provided by Matthew Brett at https://github.com/matthew-brett/build-openblas - - ps: | - $clnt = new-object System.Net.WebClient - $file = "$(New-TemporaryFile).zip" - $tmpdir = New-TemporaryFile | %{ rm $_; mkdir $_ } - $destination = "$env:PYTHON\lib\openblas.a" - - echo $file - echo $tmpdir - echo $env:OPENBLAS - - $clnt.DownloadFile($env:OPENBLAS, $file) - Get-FileHash $file | Format-List - - Expand-Archive $file $tmpdir - - rm $tmpdir\$env:PYTHON_ARCH\lib\*.dll.a - $lib = ls $tmpdir\$env:PYTHON_ARCH\lib\*.a | ForEach { ls $_ } | Select-Object -first 1 - echo $lib - - cp $lib $destination - ls $destination - - # Upgrade to the latest pip. - - 'python -m pip install -U pip setuptools wheel' - - # Install the numpy test dependencies. - - 'pip install -U --timeout 5 --retries 2 -r test_requirements.txt' - -build_script: - # Here, we add MinGW to the path to be able to link an OpenBLAS.dll - # We then use the import library from the DLL to compile with MSVC - - ps: | - pip wheel -v -v -v --wheel-dir=dist . - - # For each wheel that pip has placed in the "dist" directory - # First, upload the wheel to the "artifacts" tab and then - # install the wheel. If we have only built numpy (as is the case here), - # then there will be one wheel to install. - - # This method is more representative of what will be distributed, - # because it actually tests what the built wheels will be rather than - # what 'setup.py install' will do and at it uploads the wheels so that - # they can be inspected. - - ls dist -r | Foreach-Object { - Push-AppveyorArtifact $_.FullName - pip install $_.FullName - } - -test_script: - python runtests.py -v -n -m %TEST_MODE% -- --junitxml=%cd%\junit-results.xml - -after_build: - # Remove old or huge cache files to hopefully not exceed the 1GB cache limit. - # - # If the cache limit is reached, the cache will not be updated (of not even - # created in the first run). So this is a trade of between keeping the cache - # current and having a cache at all. - # NB: This is done only `on_success` since the cache in uploaded only on - # success anyway. - - C:\cygwin\bin\find "%LOCALAPPDATA%\pip" -type f -mtime +360 -delete - - C:\cygwin\bin\find "%LOCALAPPDATA%\pip" -type f -size +10M -delete - - C:\cygwin\bin\find "%LOCALAPPDATA%\pip" -empty -delete - # Show size of cache - - C:\cygwin\bin\du -hs "%LOCALAPPDATA%\pip\Cache" - -on_finish: - # We can get a nice display of test results in the "test" tab with py.test - # For now, this does nothing. - - ps: | - If (Test-Path .\junit-results.xml) { - (new-object net.webclient).UploadFile( - "https://ci.appveyor.com/api/testresults/junit/$($env:APPVEYOR_JOB_ID)", - (Resolve-Path .\junit-results.xml) - ) - } - $LastExitCode = 0 diff --git a/.codecov.yml b/.codecov.yml index 35584a188..d92d54c9d 100644 --- a/.codecov.yml +++ b/.codecov.yml @@ -1,6 +1,4 @@ codecov: - ci: - - !appveyor notify: require_ci_to_pass: no after_n_builds: 1 diff --git a/.travis.yml b/.travis.yml index 714122957..68564d35b 100644 --- a/.travis.yml +++ b/.travis.yml @@ -18,6 +18,14 @@ cache: directories: - $HOME/.cache/pip +stage: Comprehensive tests + +stages: + # Do the style check and a single test job, don't proceed if it fails + - name: Initial tests + # Do the rest of the tests + - name: Comprehensive tests + env: global: - OpenBLAS_version=0.3.7 @@ -29,13 +37,14 @@ env: iFWt9Ka92CaqYdU7nqfWp9VImSndPmssjmCXJ1v1IjZPAM\ ahp7Qnm0rWRmA0z9SomuRUQOJQ6s684vU=" -python: - - 3.5 - - 3.6 - - 3.7 - - 3.8-dev matrix: include: + # Do all python versions without environment variables set + - python: 3.5 + - stage: Initial tests + python: 3.6 + - python: 3.7 + - python: 3.8-dev - python: 3.7 env: INSTALL_PICKLE5=1 - python: 3.6 @@ -2,8 +2,6 @@ []( https://travis-ci.org/numpy/numpy) -[]( - https://ci.appveyor.com/project/charris/numpy) []( https://dev.azure.com/numpy/numpy/_build/latest?definitionId=5) []( diff --git a/azure-pipelines.yml b/azure-pipelines.yml index ebc45ca96..633808c0b 100644 --- a/azure-pipelines.yml +++ b/azure-pipelines.yml @@ -10,212 +10,173 @@ variables: # to match numpy-wheels repo OpenBLAS_version: 0.3.7 -jobs: -- job: Linux_Python_36_32bit_full_with_asserts - pool: - vmImage: 'ubuntu-16.04' - steps: - - script: | - docker pull i386/ubuntu:bionic - docker run -v $(pwd):/numpy i386/ubuntu:bionic /bin/bash -c "cd numpy && \ - apt-get -y update && \ - apt-get -y install python3.6-dev python3-pip locales python3-certifi && \ - locale-gen fr_FR && update-locale && \ - apt-get -y install gfortran-5 wget && \ - target=\$(python3 tools/openblas_support.py) && \ - cp -r \$target/usr/local/lib/* /usr/lib && \ - cp \$target/usr/local/include/* /usr/include && \ - python3 -m pip install --user --upgrade pip setuptools && \ - python3 -m pip install --user -r test_requirements.txt && \ - python3 -m pip install . && \ - F77=gfortran-5 F90=gfortran-5 \ - CFLAGS='-UNDEBUG -std=c99' python3 runtests.py -n --debug-info --mode=full -- -rsx --junitxml=junit/test-results.xml && \ - python3 tools/openblas_support.py --check_version $(OpenBLAS_version)" - displayName: 'Run 32-bit Ubuntu Docker Build / Tests' - - task: PublishTestResults@2 - condition: succeededOrFailed() - inputs: - testResultsFiles: '**/test-*.xml' - failTaskOnFailedTests: true - testRunTitle: 'Publish test results for Python 3.6-32 bit full Linux' -- job: macOS - pool: - # NOTE: at time of writing, there is a danger - # that using an invalid vmIMage string for macOS - # image silently redirects to a Windows build on Azure; - # for now, use the only image name officially present in - # the docs even though i.e., numba uses another in their - # azure config for mac os -- Microsoft has indicated - # they will patch this issue - vmImage: macOS-10.13 - steps: - # the @0 refers to the (major) version of the *task* on Microsoft's - # end, not the order in the build matrix nor anything to do - # with version of Python selected - - task: UsePythonVersion@0 - inputs: - versionSpec: '3.6' - addToPath: true - architecture: 'x64' - # NOTE: do we have a compelling reason to use older / newer - # versions of Xcode toolchain for testing? - - script: /bin/bash -c "sudo xcode-select -s /Applications/Xcode_10.app/Contents/Developer" - displayName: 'select Xcode version' - # NOTE: might be better if we could avoid installing - # two C compilers, but with homebrew looks like we're - # now stuck getting the full gcc toolchain instead of - # just pulling in gfortran - - script: | - # same version of gfortran as the wheel builds - brew install gcc49 - # manually link critical gfortran libraries - ln -s /usr/local/Cellar/gcc@4.9/4.9.4_1/lib/gcc/4.9/libgfortran.3.dylib /usr/local/lib/libgfortran.3.dylib - ln -s /usr/local/Cellar/gcc@4.9/4.9.4_1/lib/gcc/4.9/libquadmath.0.dylib /usr/local/lib/libquadmath.0.dylib - # manually symlink gfortran-4.9 to plain gfortran - # for f2py - ln -s /usr/local/bin/gfortran-4.9 /usr/local/bin/gfortran - displayName: 'make gfortran available on mac os vm' - # use the pre-built openblas binary that most closely - # matches our MacOS wheel builds -- currently based - # primarily on file size / name details - - script: | - target=$(python tools/openblas_support.py) - # manually link to appropriate system paths - cp $target/usr/local/lib/* /usr/local/lib/ - cp $target/usr/local/include/* /usr/local/include/ - displayName: 'install pre-built openblas' - - script: python -m pip install --upgrade pip setuptools wheel - displayName: 'Install tools' - - script: | - python -m pip install -r test_requirements.txt - python -m pip install vulture docutils sphinx==2.2.0 numpydoc - displayName: 'Install dependencies; some are optional to avoid test skips' - - script: /bin/bash -c "! vulture . --min-confidence 100 --exclude doc/,numpy/distutils/ | grep 'unreachable'" - displayName: 'Check for unreachable code paths in Python modules' - # prefer usage of clang over gcc proper - # to match likely scenario on many user mac machines - - script: python setup.py build -j 4 build_src --verbose-cfg install - displayName: 'Build NumPy' - env: - BLAS: None - LAPACK: None - ATLAS: None - ACCELERATE: None - CC: /usr/bin/clang - # wait until after dev build of NumPy to pip - # install matplotlib to avoid pip install of older numpy - - script: python -m pip install matplotlib - displayName: 'Install matplotlib before refguide run' - - script: python runtests.py -g --refguide-check - displayName: 'Run Refuide Check' - - script: python runtests.py -n --mode=full -- -rsx --junitxml=junit/test-results.xml - displayName: 'Run Full NumPy Test Suite' - - bash: python tools/openblas_support.py --check_version $(OpenBLAS_version) - displayName: 'Verify OpenBLAS version' - - task: PublishTestResults@2 - condition: succeededOrFailed() - inputs: - testResultsFiles: '**/test-*.xml' - failTaskOnFailedTests: true - testRunTitle: 'Publish test results for Python 3.6 64-bit full Mac OS' -- job: Windows - pool: - vmImage: 'VS2017-Win2016' - strategy: - maxParallel: 6 - matrix: - Python36-32bit-fast: - PYTHON_VERSION: '3.6' - PYTHON_ARCH: 'x86' - TEST_MODE: fast - BITS: 32 - Python37-32bit-fast: - PYTHON_VERSION: '3.7' - PYTHON_ARCH: 'x86' - TEST_MODE: fast - BITS: 32 - Python35-64bit-full: - PYTHON_VERSION: '3.5' - PYTHON_ARCH: 'x64' - TEST_MODE: full - BITS: 64 - Python36-64bit-full: - PYTHON_VERSION: '3.6' - PYTHON_ARCH: 'x64' - TEST_MODE: full - BITS: 64 - Python37-64bit-full: - PYTHON_VERSION: '3.7' - PYTHON_ARCH: 'x64' - TEST_MODE: full - BITS: 64 - steps: - - task: UsePythonVersion@0 - inputs: - versionSpec: $(PYTHON_VERSION) - addToPath: true - architecture: $(PYTHON_ARCH) - - script: python -m pip install --upgrade pip setuptools wheel - displayName: 'Install tools' - - script: python -m pip install -r test_requirements.txt - displayName: 'Install dependencies; some are optional to avoid test skips' - - powershell: | - $pyversion = python -c "from __future__ import print_function; import sys; print(sys.version.split()[0])" - Write-Host "Python Version: $pyversion" - $target = "C:\\hostedtoolcache\\windows\\Python\\$pyversion\\$(PYTHON_ARCH)\\lib\\openblas.a" - Write-Host "target path: $target" - $openblas = python tools/openblas_support.py - cp $openblas $target - displayName: 'Download / Install OpenBLAS' - - - powershell: | - choco install -y mingw --forcex86 --force --version=5.3.0 - displayName: 'Install 32-bit mingw for 32-bit builds' - condition: eq(variables['BITS'], 32) - # NOTE: for Windows builds it seems much more tractable to use runtests.py - # vs. manual setup.py and then runtests.py for testing only - - powershell: | - If ($(BITS) -eq 32) { - $env:CFLAGS = "-m32" - $env:LDFLAGS = "-m32" - $env:PATH = "C:\\tools\\mingw32\\bin;" + $env:PATH - refreshenv - } - python -c "from tools import openblas_support; openblas_support.make_init('numpy')" - pip wheel -v -v -v --wheel-dir=dist . - - ls dist -r | Foreach-Object { - pip install $_.FullName - } - displayName: 'Build NumPy' - - bash: | - pushd . && cd .. && target=$(python -c "import numpy, os; print(os.path.abspath(os.path.join(os.path.dirname(numpy.__file__), '.libs')))") && popd - pip download -d destination --only-binary --no-deps numpy==1.14 - cd destination && unzip numpy*.whl && cp numpy/.libs/*.dll $target - ls $target - displayName: 'Add extraneous & older DLL to numpy/.libs to probe DLL handling robustness' - condition: eq(variables['PYTHON_VERSION'], '3.6') - - script: pushd . && cd .. && python -c "from ctypes import windll; windll.kernel32.SetDefaultDllDirectories(0x00000800); import numpy" && popd - displayName: 'For gh-12667; Windows DLL resolution' - - script: python runtests.py -n --show-build-log --mode=$(TEST_MODE) -- -rsx --junitxml=junit/test-results.xml - displayName: 'Run NumPy Test Suite' - - task: PublishTestResults@2 - condition: succeededOrFailed() - inputs: - testResultsFiles: '**/test-*.xml' - failTaskOnFailedTests: true - testRunTitle: 'Publish test results for Python $(PYTHON_VERSION) $(BITS)-bit $(TEST_MODE) Windows' - -- job: Linux_PyPy3 - pool: - vmIMage: 'ubuntu-16.04' - steps: - - script: source tools/pypy-test.sh - displayName: 'Run PyPy3 Build / Tests' - - task: PublishTestResults@2 - condition: succeededOrFailed() - inputs: - testResultsFiles: '**/test-*.xml' - testRunTitle: 'Publish test results for PyPy3' - failTaskOnFailedTests: true +stages: +- stage: InitialTests + jobs: + - job: WindowsFast + pool: + vmImage: 'VS2017-Win2016' + strategy: + matrix: + Python36-64bit-fast: + PYTHON_VERSION: '3.6' + PYTHON_ARCH: 'x64' + TEST_MODE: fast + BITS: 64 + steps: + - template: azure-steps-windows.yml +- stage: ComprehensiveTests + jobs: + - job: Linux_Python_36_32bit_full_with_asserts + pool: + vmImage: 'ubuntu-16.04' + steps: + - script: | + docker pull i386/ubuntu:bionic + docker run -v $(pwd):/numpy i386/ubuntu:bionic /bin/bash -c "cd numpy && \ + apt-get -y update && \ + apt-get -y install python3.6-dev python3-pip locales python3-certifi && \ + locale-gen fr_FR && update-locale && \ + apt-get -y install gfortran-5 wget && \ + target=\$(python3 tools/openblas_support.py) && \ + cp -r \$target/usr/local/lib/* /usr/lib && \ + cp \$target/usr/local/include/* /usr/include && \ + python3 -m pip install --user --upgrade pip setuptools && \ + python3 -m pip install --user -r test_requirements.txt && \ + python3 -m pip install . && \ + F77=gfortran-5 F90=gfortran-5 \ + CFLAGS='-UNDEBUG -std=c99' python3 runtests.py -n --debug-info --mode=full -- -rsx --junitxml=junit/test-results.xml && \ + python3 tools/openblas_support.py --check_version $(OpenBLAS_version)" + displayName: 'Run 32-bit Ubuntu Docker Build / Tests' + - task: PublishTestResults@2 + condition: succeededOrFailed() + inputs: + testResultsFiles: '**/test-*.xml' + failTaskOnFailedTests: true + testRunTitle: 'Publish test results for Python 3.6-32 bit full Linux' + - job: macOS + pool: + # NOTE: at time of writing, there is a danger + # that using an invalid vmIMage string for macOS + # image silently redirects to a Windows build on Azure; + # for now, use the only image name officially present in + # the docs even though i.e., numba uses another in their + # azure config for mac os -- Microsoft has indicated + # they will patch this issue + vmImage: macOS-10.13 + steps: + # the @0 refers to the (major) version of the *task* on Microsoft's + # end, not the order in the build matrix nor anything to do + # with version of Python selected + - task: UsePythonVersion@0 + inputs: + versionSpec: '3.6' + addToPath: true + architecture: 'x64' + # NOTE: do we have a compelling reason to use older / newer + # versions of Xcode toolchain for testing? + - script: /bin/bash -c "sudo xcode-select -s /Applications/Xcode_10.app/Contents/Developer" + displayName: 'select Xcode version' + # NOTE: might be better if we could avoid installing + # two C compilers, but with homebrew looks like we're + # now stuck getting the full gcc toolchain instead of + # just pulling in gfortran + - script: | + # same version of gfortran as the wheel builds + brew install gcc49 + # manually link critical gfortran libraries + ln -s /usr/local/Cellar/gcc@4.9/4.9.4_1/lib/gcc/4.9/libgfortran.3.dylib /usr/local/lib/libgfortran.3.dylib + ln -s /usr/local/Cellar/gcc@4.9/4.9.4_1/lib/gcc/4.9/libquadmath.0.dylib /usr/local/lib/libquadmath.0.dylib + # manually symlink gfortran-4.9 to plain gfortran + # for f2py + ln -s /usr/local/bin/gfortran-4.9 /usr/local/bin/gfortran + displayName: 'make gfortran available on mac os vm' + # use the pre-built openblas binary that most closely + # matches our MacOS wheel builds -- currently based + # primarily on file size / name details + - script: | + target=$(python tools/openblas_support.py) + # manually link to appropriate system paths + cp $target/usr/local/lib/* /usr/local/lib/ + cp $target/usr/local/include/* /usr/local/include/ + displayName: 'install pre-built openblas' + - script: python -m pip install --upgrade pip setuptools wheel + displayName: 'Install tools' + - script: | + python -m pip install -r test_requirements.txt + python -m pip install vulture docutils sphinx==2.2.0 numpydoc + displayName: 'Install dependencies; some are optional to avoid test skips' + - script: /bin/bash -c "! vulture . --min-confidence 100 --exclude doc/,numpy/distutils/ | grep 'unreachable'" + displayName: 'Check for unreachable code paths in Python modules' + # prefer usage of clang over gcc proper + # to match likely scenario on many user mac machines + - script: python setup.py build -j 4 build_src --verbose-cfg install + displayName: 'Build NumPy' + env: + BLAS: None + LAPACK: None + ATLAS: None + ACCELERATE: None + CC: /usr/bin/clang + # wait until after dev build of NumPy to pip + # install matplotlib to avoid pip install of older numpy + - script: python -m pip install matplotlib + displayName: 'Install matplotlib before refguide run' + - script: python runtests.py -g --refguide-check + displayName: 'Run Refuide Check' + - script: python runtests.py -n --mode=full -- -rsx --junitxml=junit/test-results.xml + displayName: 'Run Full NumPy Test Suite' + - bash: python tools/openblas_support.py --check_version $(OpenBLAS_version) + displayName: 'Verify OpenBLAS version' + - task: PublishTestResults@2 + condition: succeededOrFailed() + inputs: + testResultsFiles: '**/test-*.xml' + failTaskOnFailedTests: true + testRunTitle: 'Publish test results for Python 3.6 64-bit full Mac OS' + - job: Windows + pool: + vmImage: 'VS2017-Win2016' + strategy: + maxParallel: 6 + matrix: + Python36-32bit-fast: + PYTHON_VERSION: '3.6' + PYTHON_ARCH: 'x86' + TEST_MODE: fast + BITS: 32 + Python37-32bit-fast: + PYTHON_VERSION: '3.7' + PYTHON_ARCH: 'x86' + TEST_MODE: fast + BITS: 32 + Python35-64bit-full: + PYTHON_VERSION: '3.5' + PYTHON_ARCH: 'x64' + TEST_MODE: full + BITS: 64 + Python36-64bit-full: + PYTHON_VERSION: '3.6' + PYTHON_ARCH: 'x64' + TEST_MODE: full + BITS: 64 + Python37-64bit-full: + PYTHON_VERSION: '3.7' + PYTHON_ARCH: 'x64' + TEST_MODE: full + BITS: 64 + steps: + - template: azure-steps-windows.yml + - job: Linux_PyPy3 + pool: + vmIMage: 'ubuntu-16.04' + steps: + - script: source tools/pypy-test.sh + displayName: 'Run PyPy3 Build / Tests' + - task: PublishTestResults@2 + condition: succeededOrFailed() + inputs: + testResultsFiles: '**/test-*.xml' + testRunTitle: 'Publish test results for PyPy3' + failTaskOnFailedTests: true diff --git a/azure-steps-windows.yml b/azure-steps-windows.yml new file mode 100644 index 000000000..26d7a667d --- /dev/null +++ b/azure-steps-windows.yml @@ -0,0 +1,56 @@ +steps: +- task: UsePythonVersion@0 + inputs: + versionSpec: $(PYTHON_VERSION) + addToPath: true + architecture: $(PYTHON_ARCH) +- script: python -m pip install --upgrade pip setuptools wheel + displayName: 'Install tools' +- script: python -m pip install -r test_requirements.txt + displayName: 'Install dependencies; some are optional to avoid test skips' +- powershell: | + $pyversion = python -c "from __future__ import print_function; import sys; print(sys.version.split()[0])" + Write-Host "Python Version: $pyversion" + $target = "C:\\hostedtoolcache\\windows\\Python\\$pyversion\\$(PYTHON_ARCH)\\lib\\openblas.a" + Write-Host "target path: $target" + $openblas = python tools/openblas_support.py + cp $openblas $target + displayName: 'Download / Install OpenBLAS' + +- powershell: | + choco install -y mingw --forcex86 --force --version=5.3.0 + displayName: 'Install 32-bit mingw for 32-bit builds' + condition: eq(variables['BITS'], 32) +# NOTE: for Windows builds it seems much more tractable to use runtests.py +# vs. manual setup.py and then runtests.py for testing only +- powershell: | + If ($(BITS) -eq 32) { + $env:CFLAGS = "-m32" + $env:LDFLAGS = "-m32" + $env:PATH = "C:\\tools\\mingw32\\bin;" + $env:PATH + refreshenv + } + python -c "from tools import openblas_support; openblas_support.make_init('numpy')" + pip wheel -v -v -v --wheel-dir=dist . + + ls dist -r | Foreach-Object { + pip install $_.FullName + } + displayName: 'Build NumPy' +- bash: | + pushd . && cd .. && target=$(python -c "import numpy, os; print(os.path.abspath(os.path.join(os.path.dirname(numpy.__file__), '.libs')))") && popd + pip download -d destination --only-binary --no-deps numpy==1.14 + cd destination && unzip numpy*.whl && cp numpy/.libs/*.dll $target + ls $target + displayName: 'Add extraneous & older DLL to numpy/.libs to probe DLL handling robustness' + condition: eq(variables['PYTHON_VERSION'], '3.6') +- script: pushd . && cd .. && python -c "from ctypes import windll; windll.kernel32.SetDefaultDllDirectories(0x00000800); import numpy" && popd + displayName: 'For gh-12667; Windows DLL resolution' +- script: python runtests.py -n --show-build-log --mode=$(TEST_MODE) -- -rsx --junitxml=junit/test-results.xml + displayName: 'Run NumPy Test Suite' +- task: PublishTestResults@2 + condition: succeededOrFailed() + inputs: + testResultsFiles: '**/test-*.xml' + failTaskOnFailedTests: true + testRunTitle: 'Publish test results for Python $(PYTHON_VERSION) $(BITS)-bit $(TEST_MODE) Windows'
\ No newline at end of file diff --git a/benchmarks/benchmarks/bench_avx.py b/benchmarks/benchmarks/bench_avx.py new file mode 100644 index 000000000..f7b524e43 --- /dev/null +++ b/benchmarks/benchmarks/bench_avx.py @@ -0,0 +1,34 @@ +from __future__ import absolute_import, division, print_function + +from .common import Benchmark + +import numpy as np + +avx_ufuncs = ['sqrt', + 'absolute', + 'reciprocal', + 'square', + 'rint', + 'floor', + 'ceil' , + 'trunc'] +stride = [1, 2, 4] +dtype = ['f', 'd'] + +class AVX_UFunc(Benchmark): + params = [avx_ufuncs, stride, dtype] + param_names = ['avx_based_ufunc', 'stride', 'dtype'] + timeout = 10 + + def setup(self, ufuncname, stride, dtype): + np.seterr(all='ignore') + try: + self.f = getattr(np, ufuncname) + except AttributeError: + raise NotImplementedError() + N = 10000 + self.arr = np.ones(stride*N, dtype) + + def time_ufunc(self, ufuncname, stride, dtype): + self.f(self.arr[::stride]) + diff --git a/doc/neps/nep-0029-deprecation_policy.rst b/doc/neps/nep-0029-deprecation_policy.rst index 0dea0a96f..1e5aa5b2c 100644 --- a/doc/neps/nep-0029-deprecation_policy.rst +++ b/doc/neps/nep-0029-deprecation_policy.rst @@ -164,7 +164,7 @@ the minimum version of Python supported. As a major downside, an ad-hoc approach makes it hard for downstream users to predict what the future minimum versions will be. As there is no objective threshold to when the minimum version should be dropped, it is easy for these -version support discussions to devolve into [bike shedding](https://en.wikipedia.org/wiki/Wikipedia:Avoid_Parkinson%27s_bicycle-shed_effect) and acrimony. +version support discussions to devolve into `bike shedding <https://en.wikipedia.org/wiki/Wikipedia:Avoid_Parkinson%27s_bicycle-shed_effect>`_ and acrimony. All CPython supported versions diff --git a/numpy/core/_internal.py b/numpy/core/_internal.py index b0ea603e1..5fd643505 100644 --- a/numpy/core/_internal.py +++ b/numpy/core/_internal.py @@ -247,55 +247,13 @@ class _missing_ctypes(object): self.value = ptr -class _unsafe_first_element_pointer(object): - """ - Helper to allow viewing an array as a ctypes pointer to the first element - - This avoids: - * dealing with strides - * `.view` rejecting object-containing arrays - * `memoryview` not supporting overlapping fields - """ - def __init__(self, arr): - self.base = arr - - @property - def __array_interface__(self): - i = dict( - shape=(), - typestr='|V0', - data=(self.base.__array_interface__['data'][0], False), - strides=(), - version=3, - ) - return i - - -def _get_void_ptr(arr): - """ - Get a `ctypes.c_void_p` to arr.data, that keeps a reference to the array - """ - import numpy as np - # convert to a 0d array that has a data pointer referrign to the start - # of arr. This holds a reference to arr. - simple_arr = np.asarray(_unsafe_first_element_pointer(arr)) - - # create a `char[0]` using the same memory. - c_arr = (ctypes.c_char * 0).from_buffer(simple_arr) - - # finally cast to void* - return ctypes.cast(ctypes.pointer(c_arr), ctypes.c_void_p) - - class _ctypes(object): def __init__(self, array, ptr=None): self._arr = array if ctypes: self._ctypes = ctypes - # get a void pointer to the buffer, which keeps the array alive - self._data = _get_void_ptr(array) - assert self._data.value == ptr + self._data = self._ctypes.c_void_p(ptr) else: # fake a pointer-like object that holds onto the reference self._ctypes = _missing_ctypes() @@ -317,7 +275,14 @@ class _ctypes(object): The returned pointer will keep a reference to the array. """ - return self._ctypes.cast(self._data, obj) + # _ctypes.cast function causes a circular reference of self._data in + # self._data._objects. Attributes of self._data cannot be released + # until gc.collect is called. Make a copy of the pointer first then let + # it hold the array reference. This is a workaround to circumvent the + # CPython bug https://bugs.python.org/issue12836 + ptr = self._ctypes.cast(self._data, obj) + ptr._arr = self._arr + return ptr def shape_as(self, obj): """ @@ -385,7 +350,7 @@ class _ctypes(object): Enables `c_func(some_array.ctypes)` """ - return self._data + return self.data_as(ctypes.c_void_p) # kept for compatibility get_data = data.fget diff --git a/numpy/core/arrayprint.py b/numpy/core/arrayprint.py index 233d139fd..8a7626d9d 100644 --- a/numpy/core/arrayprint.py +++ b/numpy/core/arrayprint.py @@ -1479,7 +1479,11 @@ def array_repr(arr, max_line_width=None, precision=None, suppress_small=None): arr, max_line_width, precision, suppress_small) -_guarded_str = _recursive_guard()(str) +@_recursive_guard() +def _guarded_repr_or_str(v): + if isinstance(v, bytes): + return repr(v) + return str(v) def _array_str_implementation( @@ -1497,7 +1501,7 @@ def _array_str_implementation( # obtain a scalar and call str on it, avoiding problems for subclasses # for which indexing with () returns a 0d instead of a scalar by using # ndarray's getindex. Also guard against recursive 0d object arrays. - return _guarded_str(np.ndarray.__getitem__(a, ())) + return _guarded_repr_or_str(np.ndarray.__getitem__(a, ())) return array2string(a, max_line_width, precision, suppress_small, ' ', "") diff --git a/numpy/core/code_generators/generate_umath.py b/numpy/core/code_generators/generate_umath.py index 6729fe197..0d3bbffe9 100644 --- a/numpy/core/code_generators/generate_umath.py +++ b/numpy/core/code_generators/generate_umath.py @@ -358,14 +358,14 @@ defdict = { Ufunc(1, 1, None, docstrings.get('numpy.core.umath.square'), None, - TD(ints+inexact, simd=[('avx2', ints)]), + TD(ints+inexact, simd=[('avx2', ints), ('fma', 'fd'), ('avx512f', 'fd')]), TD(O, f='Py_square'), ), 'reciprocal': Ufunc(1, 1, None, docstrings.get('numpy.core.umath.reciprocal'), None, - TD(ints+inexact, simd=[('avx2', ints)]), + TD(ints+inexact, simd=[('avx2', ints), ('fma', 'fd'), ('avx512f','fd')]), TD(O, f='Py_reciprocal'), ), # This is no longer used as numpy.ones_like, however it is @@ -395,7 +395,7 @@ defdict = { Ufunc(1, 1, None, docstrings.get('numpy.core.umath.absolute'), 'PyUFunc_AbsoluteTypeResolver', - TD(bints+flts+timedeltaonly), + TD(bints+flts+timedeltaonly, simd=[('fma', 'fd'), ('avx512f', 'fd')]), TD(cmplx, out=('f', 'd', 'g')), TD(O, f='PyNumber_Absolute'), ), @@ -762,7 +762,7 @@ defdict = { docstrings.get('numpy.core.umath.sqrt'), None, TD('e', f='sqrt', astype={'e':'f'}), - TD(inexactvec), + TD(inexactvec, simd=[('fma', 'fd'), ('avx512f', 'fd')]), TD('fdg' + cmplx, f='sqrt'), TD(P, f='sqrt'), ), @@ -777,14 +777,18 @@ defdict = { Ufunc(1, 1, None, docstrings.get('numpy.core.umath.ceil'), None, - TD(flts, f='ceil', astype={'e':'f'}), + TD('e', f='ceil', astype={'e':'f'}), + TD(inexactvec, simd=[('fma', 'fd'), ('avx512f', 'fd')]), + TD('fdg', f='ceil'), TD(O, f='npy_ObjectCeil'), ), 'trunc': Ufunc(1, 1, None, docstrings.get('numpy.core.umath.trunc'), None, - TD(flts, f='trunc', astype={'e':'f'}), + TD('e', f='trunc', astype={'e':'f'}), + TD(inexactvec, simd=[('fma', 'fd'), ('avx512f', 'fd')]), + TD('fdg', f='trunc'), TD(O, f='npy_ObjectTrunc'), ), 'fabs': @@ -798,14 +802,18 @@ defdict = { Ufunc(1, 1, None, docstrings.get('numpy.core.umath.floor'), None, - TD(flts, f='floor', astype={'e':'f'}), + TD('e', f='floor', astype={'e':'f'}), + TD(inexactvec, simd=[('fma', 'fd'), ('avx512f', 'fd')]), + TD('fdg', f='floor'), TD(O, f='npy_ObjectFloor'), ), 'rint': Ufunc(1, 1, None, docstrings.get('numpy.core.umath.rint'), None, - TD(inexact, f='rint', astype={'e':'f'}), + TD('e', f='rint', astype={'e':'f'}), + TD(inexactvec, simd=[('fma', 'fd'), ('avx512f', 'fd')]), + TD('fdg' + cmplx, f='rint'), TD(P, f='rint'), ), 'arctan2': diff --git a/numpy/core/numeric.py b/numpy/core/numeric.py index 6d25f864b..b5568fd86 100644 --- a/numpy/core/numeric.py +++ b/numpy/core/numeric.py @@ -960,6 +960,9 @@ def tensordot(a, b, axes=2): two sequences of the same length, with the first axis to sum over given first in both sequences, the second axis second, and so forth. + The shape of the result consists of the non-contracted axes of the + first tensor, followed by the non-contracted axes of the second. + Examples -------- A "traditional" example: diff --git a/numpy/core/src/multiarray/descriptor.c b/numpy/core/src/multiarray/descriptor.c index 734255a9d..23d140cf6 100644 --- a/numpy/core/src/multiarray/descriptor.c +++ b/numpy/core/src/multiarray/descriptor.c @@ -1385,7 +1385,6 @@ NPY_NO_EXPORT int PyArray_DescrConverter(PyObject *obj, PyArray_Descr **at) { int check_num = NPY_NOTYPE + 10; - PyObject *item; int elsize = 0; char endian = '='; @@ -1664,16 +1663,22 @@ finish: PyErr_Clear(); /* Now check to see if the object is registered in typeDict */ if (typeDict != NULL) { - item = PyDict_GetItem(typeDict, obj); + PyObject *item = NULL; #if defined(NPY_PY3K) - if (!item && PyBytes_Check(obj)) { + if (PyBytes_Check(obj)) { PyObject *tmp; tmp = PyUnicode_FromEncodedObject(obj, "ascii", "strict"); - if (tmp != NULL) { - item = PyDict_GetItem(typeDict, tmp); - Py_DECREF(tmp); + if (tmp == NULL) { + goto fail; } + item = PyDict_GetItem(typeDict, tmp); + Py_DECREF(tmp); + } + else { + item = PyDict_GetItem(typeDict, obj); } +#else + item = PyDict_GetItem(typeDict, obj); #endif if (item) { /* Check for a deprecated Numeric-style typecode */ diff --git a/numpy/core/src/umath/loops.c.src b/numpy/core/src/umath/loops.c.src index 5443223ab..d948e25bb 100644 --- a/numpy/core/src/umath/loops.c.src +++ b/numpy/core/src/umath/loops.c.src @@ -1635,6 +1635,30 @@ NPY_NO_EXPORT void /**end repeat**/ /**begin repeat + * #func = rint, ceil, floor, trunc# + * #scalarf = npy_rint, npy_ceil, npy_floor, npy_trunc# + */ + +/**begin repeat1 +* #TYPE = FLOAT, DOUBLE# +* #type = npy_float, npy_double# +* #typesub = f, # +*/ + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_@func@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + *(@type@ *)op1 = @scalarf@@typesub@(in1); + } +} + + +/**end repeat1**/ +/**end repeat**/ + +/**begin repeat * #func = sin, cos, exp, log# * #scalarf = npy_sinf, npy_cosf, npy_expf, npy_logf# */ @@ -1657,6 +1681,78 @@ FLOAT_@func@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSE */ /**begin repeat1 + * #TYPE = FLOAT, DOUBLE# + * #type = npy_float, npy_double# + * #typesub = f, # + */ + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_sqrt_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + if (!run_unary_@isa@_sqrt_@TYPE@(args, dimensions, steps)) { + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + *(@type@ *)op1 = npy_sqrt@typesub@(in1); + } + } +} + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_absolute_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + if (!run_unary_@isa@_absolute_@TYPE@(args, dimensions, steps)) { + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + const @type@ tmp = in1 > 0 ? in1 : -in1; + /* add 0 to clear -0.0 */ + *((@type@ *)op1) = tmp + 0; + } + } + npy_clear_floatstatus_barrier((char*)dimensions); +} + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_square_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + if (!run_unary_@isa@_square_@TYPE@(args, dimensions, steps)) { + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + *(@type@ *)op1 = in1*in1; + } + } +} + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_reciprocal_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + if (!run_unary_@isa@_reciprocal_@TYPE@(args, dimensions, steps)) { + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + *(@type@ *)op1 = 1.0f/in1; + } + } +} + +/**begin repeat2 + * #func = rint, ceil, floor, trunc# + * #scalarf = npy_rint, npy_ceil, npy_floor, npy_trunc# + */ + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_@func@_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)) +{ + if (!run_unary_@isa@_@func@_@TYPE@(args, dimensions, steps)) { + UNARY_LOOP { + const @type@ in1 = *(@type@ *)ip1; + *(@type@ *)op1 = @scalarf@@typesub@(in1); + } + } +} + +/**end repeat2**/ +/**end repeat1**/ + +/**begin repeat1 * #func = exp, log# * #scalarf = npy_expf, npy_logf# */ @@ -1706,10 +1802,9 @@ FLOAT_@func@_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY } /**end repeat1**/ - - /**end repeat**/ + /**begin repeat * Float types * #type = npy_float, npy_double, npy_longdouble, npy_float# diff --git a/numpy/core/src/umath/loops.h.src b/numpy/core/src/umath/loops.h.src index 5070ab38b..e98a1ac3c 100644 --- a/numpy/core/src/umath/loops.h.src +++ b/numpy/core/src/umath/loops.h.src @@ -175,6 +175,19 @@ NPY_NO_EXPORT void */ NPY_NO_EXPORT void @TYPE@_sqrt(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func)); + +/**begin repeat1 + * #isa = avx512f, fma# + */ + +/**begin repeat2 + * #func = sqrt, absolute, square, reciprocal# + */ +NPY_NO_EXPORT void +@TYPE@_@func@_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func)); + +/**end repeat2**/ +/**end repeat1**/ /**end repeat**/ /**begin repeat @@ -194,6 +207,26 @@ FLOAT_@func@_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY /**end repeat**/ /**begin repeat + * #func = rint, ceil, floor, trunc# + */ + +/**begin repeat1 +* #TYPE = FLOAT, DOUBLE# +*/ + +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_@func@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)); + +/**begin repeat2 + * #isa = avx512f, fma# + */ +NPY_NO_EXPORT NPY_GCC_OPT_3 void +@TYPE@_@func@_@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data)); +/**end repeat2**/ +/**end repeat1**/ +/**end repeat**/ + +/**begin repeat * Float types * #TYPE = HALF, FLOAT, DOUBLE, LONGDOUBLE# * #c = f, f, , l# diff --git a/numpy/core/src/umath/override.c b/numpy/core/src/umath/override.c index 8d67f96ac..43bed425c 100644 --- a/numpy/core/src/umath/override.c +++ b/numpy/core/src/umath/override.c @@ -494,32 +494,18 @@ PyUFunc_CheckOverride(PyUFuncObject *ufunc, char *method, } else { /* not a tuple */ - if (nout > 1 && DEPRECATE("passing a single argument to the " - "'out' keyword argument of a " - "ufunc with\n" - "more than one output will " - "result in an error in the " - "future") < 0) { - /* - * If the deprecation is removed, also remove the loop - * below setting tuple items to None (but keep this future - * error message.) - */ + if (nout > 1) { PyErr_SetString(PyExc_TypeError, "'out' must be a tuple of arguments"); goto fail; } if (out != Py_None) { /* not already a tuple and not None */ - PyObject *out_tuple = PyTuple_New(nout); + PyObject *out_tuple = PyTuple_New(1); if (out_tuple == NULL) { goto fail; } - for (i = 1; i < nout; i++) { - Py_INCREF(Py_None); - PyTuple_SET_ITEM(out_tuple, i, Py_None); - } /* out was borrowed ref; make it permanent */ Py_INCREF(out); /* steals reference */ diff --git a/numpy/core/src/umath/simd.inc.src b/numpy/core/src/umath/simd.inc.src index 88e5e1f1b..74f52cc9d 100644 --- a/numpy/core/src/umath/simd.inc.src +++ b/numpy/core/src/umath/simd.inc.src @@ -139,6 +139,37 @@ abs_ptrdiff(char *a, char *b) /* prototypes */ /**begin repeat1 + * #type = npy_float, npy_double# + * #TYPE = FLOAT, DOUBLE# + */ + +/**begin repeat2 + * #func = sqrt, absolute, square, reciprocal, rint, floor, ceil, trunc# + */ + +#if defined @CHK@ && defined NPY_HAVE_SSE2_INTRINSICS +static NPY_INLINE NPY_GCC_TARGET_@ISA@ void +@ISA@_@func@_@TYPE@(@type@ *, @type@ *, const npy_intp n, const npy_intp stride); +#endif + +static NPY_INLINE int +run_unary_@isa@_@func@_@TYPE@(char **args, npy_intp *dimensions, npy_intp *steps) +{ +#if defined @CHK@ && defined NPY_HAVE_SSE2_INTRINSICS + if (IS_OUTPUT_BLOCKABLE_UNARY(sizeof(@type@), @REGISTER_SIZE@)) { + @ISA@_@func@_@TYPE@((@type@*)args[1], (@type@*)args[0], dimensions[0], steps[0]); + return 1; + } + else + return 0; +#endif + return 0; +} + +/**end repeat2**/ +/**end repeat1**/ + +/**begin repeat1 * #func = exp, log# */ @@ -185,7 +216,6 @@ run_unary_@isa@_sincos_FLOAT(char **args, npy_intp *dimensions, npy_intp *steps, /**end repeat**/ - /**begin repeat * Float types * #type = npy_float, npy_double, npy_longdouble# @@ -1144,41 +1174,76 @@ sse2_@kind@_@TYPE@(@type@ * ip, @type@ * op, const npy_intp n) #if defined HAVE_ATTRIBUTE_TARGET_AVX2_WITH_INTRINSICS static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 -fma_get_full_load_mask(void) +fma_get_full_load_mask_ps(void) { return _mm256_set1_ps(-1.0); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256i +fma_get_full_load_mask_pd(void) +{ + return _mm256_castpd_si256(_mm256_set1_pd(-1.0)); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 -fma_get_partial_load_mask(const npy_int num_lanes, const npy_int total_elem) +fma_get_partial_load_mask_ps(const npy_int num_elem, const npy_int num_lanes) { float maskint[16] = {-1.0,-1.0,-1.0,-1.0,-1.0,-1.0,-1.0,-1.0, 1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}; - float* addr = maskint + total_elem - num_lanes; + float* addr = maskint + num_lanes - num_elem; return _mm256_loadu_ps(addr); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256i +fma_get_partial_load_mask_pd(const npy_int num_elem, const npy_int num_lanes) +{ + npy_int maskint[16] = {-1,-1,-1,-1,-1,-1,-1,-1,1,1,1,1,1,1,1,1}; + npy_int* addr = maskint + 2*num_lanes - 2*num_elem; + return _mm256_loadu_si256((__m256i*) addr); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 -fma_masked_gather(__m256 src, - npy_float* addr, - __m256i vindex, - __m256 mask) +fma_masked_gather_ps(__m256 src, + npy_float* addr, + __m256i vindex, + __m256 mask) { return _mm256_mask_i32gather_ps(src, addr, vindex, mask, 4); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256d +fma_masked_gather_pd(__m256d src, + npy_double* addr, + __m128i vindex, + __m256d mask) +{ + return _mm256_mask_i32gather_pd(src, addr, vindex, mask, 8); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 -fma_masked_load(__m256 mask, npy_float* addr) +fma_masked_load_ps(__m256 mask, npy_float* addr) { return _mm256_maskload_ps(addr, _mm256_cvtps_epi32(mask)); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256d +fma_masked_load_pd(__m256i mask, npy_double* addr) +{ + return _mm256_maskload_pd(addr, mask); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 -fma_set_masked_lanes(__m256 x, __m256 val, __m256 mask) +fma_set_masked_lanes_ps(__m256 x, __m256 val, __m256 mask) { return _mm256_blendv_ps(x, val, mask); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256d +fma_set_masked_lanes_pd(__m256d x, __m256d val, __m256d mask) +{ + return _mm256_blendv_pd(x, val, mask); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 fma_blend(__m256 x, __m256 y, __m256 ymask) { @@ -1186,6 +1251,18 @@ fma_blend(__m256 x, __m256 y, __m256 ymask) } static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 +fma_invert_mask_ps(__m256 ymask) +{ + return _mm256_andnot_ps(ymask, _mm256_set1_ps(-1.0)); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256i +fma_invert_mask_pd(__m256i ymask) +{ + return _mm256_andnot_si256(ymask, _mm256_set1_epi32(0xFFFFFFFF)); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA __m256 fma_should_calculate_sine(__m256i k, __m256i andop, __m256i cmp) { return _mm256_cvtepi32_ps( @@ -1290,42 +1367,115 @@ fma_scalef_ps(__m256 poly, __m256 quadrant) } } +/**begin repeat + * #vsub = ps, pd# + * #vtype = __m256, __m256d# + */ +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_abs_@vsub@(@vtype@ x) +{ + return _mm256_andnot_@vsub@(_mm256_set1_@vsub@(-0.0), x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_reciprocal_@vsub@(@vtype@ x) +{ + return _mm256_div_@vsub@(_mm256_set1_@vsub@(1.0f), x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_rint_@vsub@(@vtype@ x) +{ + return _mm256_round_@vsub@(x, _MM_FROUND_TO_NEAREST_INT); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_floor_@vsub@(@vtype@ x) +{ + return _mm256_round_@vsub@(x, _MM_FROUND_TO_NEG_INF); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_ceil_@vsub@(@vtype@ x) +{ + return _mm256_round_@vsub@(x, _MM_FROUND_TO_POS_INF); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_FMA @vtype@ +fma_trunc_@vsub@(@vtype@ x) +{ + return _mm256_round_@vsub@(x, _MM_FROUND_TO_ZERO); +} +/**end repeat**/ #endif #if defined HAVE_ATTRIBUTE_TARGET_AVX512F_WITH_INTRINSICS static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask16 -avx512_get_full_load_mask(void) +avx512_get_full_load_mask_ps(void) { return 0xFFFF; } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask8 +avx512_get_full_load_mask_pd(void) +{ + return 0xFF; +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask16 -avx512_get_partial_load_mask(const npy_int num_elem, const npy_int total_elem) +avx512_get_partial_load_mask_ps(const npy_int num_elem, const npy_int total_elem) { return (0x0001 << num_elem) - 0x0001; } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask8 +avx512_get_partial_load_mask_pd(const npy_int num_elem, const npy_int total_elem) +{ + return (0x01 << num_elem) - 0x01; +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512 -avx512_masked_gather(__m512 src, - npy_float* addr, - __m512i vindex, - __mmask16 kmask) +avx512_masked_gather_ps(__m512 src, + npy_float* addr, + __m512i vindex, + __mmask16 kmask) { return _mm512_mask_i32gather_ps(src, kmask, vindex, addr, 4); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512d +avx512_masked_gather_pd(__m512d src, + npy_double* addr, + __m256i vindex, + __mmask8 kmask) +{ + return _mm512_mask_i32gather_pd(src, kmask, vindex, addr, 8); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512 -avx512_masked_load(__mmask16 mask, npy_float* addr) +avx512_masked_load_ps(__mmask16 mask, npy_float* addr) { return _mm512_maskz_loadu_ps(mask, (__m512 *)addr); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512d +avx512_masked_load_pd(__mmask8 mask, npy_double* addr) +{ + return _mm512_maskz_loadu_pd(mask, (__m512d *)addr); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512 -avx512_set_masked_lanes(__m512 x, __m512 val, __mmask16 mask) +avx512_set_masked_lanes_ps(__m512 x, __m512 val, __mmask16 mask) { return _mm512_mask_blend_ps(mask, x, val); } +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512d +avx512_set_masked_lanes_pd(__m512d x, __m512d val, __mmask8 mask) +{ + return _mm512_mask_blend_pd(mask, x, val); +} + static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __m512 avx512_blend(__m512 x, __m512 y, __mmask16 ymask) { @@ -1333,6 +1483,18 @@ avx512_blend(__m512 x, __m512 y, __mmask16 ymask) } static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask16 +avx512_invert_mask_ps(__mmask16 ymask) +{ + return _mm512_knot(ymask); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask8 +avx512_invert_mask_pd(__mmask8 ymask) +{ + return _mm512_knot(ymask); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F __mmask16 avx512_should_calculate_sine(__m512i k, __m512i andop, __m512i cmp) { return _mm512_cmpeq_epi32_mask(_mm512_and_epi32(k, andop), cmp); @@ -1361,6 +1523,49 @@ avx512_scalef_ps(__m512 poly, __m512 quadrant) { return _mm512_scalef_ps(poly, quadrant); } +/**begin repeat + * #vsub = ps, pd# + * #epi_vsub = epi32, epi64# + * #vtype = __m512, __m512d# + * #and_const = 0x7fffffff, 0x7fffffffffffffffLL# + */ +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_abs_@vsub@(@vtype@ x) +{ + return (@vtype@) _mm512_and_@epi_vsub@((__m512i) x, + _mm512_set1_@epi_vsub@ (@and_const@)); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_reciprocal_@vsub@(@vtype@ x) +{ + return _mm512_div_@vsub@(_mm512_set1_@vsub@(1.0f), x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_rint_@vsub@(@vtype@ x) +{ + return _mm512_roundscale_@vsub@(x, 0x08); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_floor_@vsub@(@vtype@ x) +{ + return _mm512_roundscale_@vsub@(x, 0x09); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_ceil_@vsub@(@vtype@ x) +{ + return _mm512_roundscale_@vsub@(x, 0x0A); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_AVX512F @vtype@ +avx512_trunc_@vsub@(@vtype@ x) +{ + return _mm512_roundscale_@vsub@(x, 0x0B); +} +/**end repeat**/ #endif /**begin repeat @@ -1438,7 +1643,187 @@ static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@ sin = @fmadd@(sin, x, x); return sin; } + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@ +@isa@_sqrt_ps(@vtype@ x) +{ + return _mm@vsize@_sqrt_ps(x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@d +@isa@_sqrt_pd(@vtype@d x) +{ + return _mm@vsize@_sqrt_pd(x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@ +@isa@_square_ps(@vtype@ x) +{ + return _mm@vsize@_mul_ps(x,x); +} + +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@d +@isa@_square_pd(@vtype@d x) +{ + return _mm@vsize@_mul_pd(x,x); +} + +#endif +/**end repeat**/ + + +/**begin repeat + * #ISA = FMA, AVX512F# + * #isa = fma, avx512# + * #vsize = 256, 512# + * #BYTES = 32, 64# + * #cvtps_epi32 = _mm256_cvtps_epi32, # + * #mask = __m256, __mmask16# + * #vsub = , _mask# + * #vtype = __m256, __m512# + * #cvtps_epi32 = _mm256_cvtps_epi32, # + * #masked_store = _mm256_maskstore_ps, _mm512_mask_storeu_ps# + * #CHK = HAVE_ATTRIBUTE_TARGET_AVX2_WITH_INTRINSICS, HAVE_ATTRIBUTE_TARGET_AVX512F_WITH_INTRINSICS# + */ + +/**begin repeat1 + * #func = sqrt, absolute, square, reciprocal, rint, ceil, floor, trunc# + * #vectorf = sqrt, abs, square, reciprocal, rint, ceil, floor, trunc# + * #replace_0_with_1 = 0, 0, 0, 1, 0, 0, 0, 0# + */ + +#if defined @CHK@ +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void +@ISA@_@func@_FLOAT(npy_float* op, + npy_float* ip, + const npy_intp array_size, + const npy_intp steps) +{ + const npy_intp stride = steps/sizeof(npy_float); + const npy_int num_lanes = @BYTES@/sizeof(npy_float); + npy_intp num_remaining_elements = array_size; + @vtype@ ones_f = _mm@vsize@_set1_ps(1.0f); + @mask@ load_mask = @isa@_get_full_load_mask_ps(); +#if @replace_0_with_1@ + @mask@ inv_load_mask = @isa@_invert_mask_ps(load_mask); +#endif + npy_int indexarr[16]; + for (npy_int ii = 0; ii < 16; ii++) { + indexarr[ii] = ii*stride; + } + @vtype@i vindex = _mm@vsize@_loadu_si@vsize@((@vtype@i*)&indexarr[0]); + + while (num_remaining_elements > 0) { + if (num_remaining_elements < num_lanes) { + load_mask = @isa@_get_partial_load_mask_ps(num_remaining_elements, + num_lanes); +#if @replace_0_with_1@ + inv_load_mask = @isa@_invert_mask_ps(load_mask); +#endif + } + @vtype@ x; + if (stride == 1) { + x = @isa@_masked_load_ps(load_mask, ip); +#if @replace_0_with_1@ + /* + * Replace masked elements with 1.0f to avoid divide by zero fp + * exception in reciprocal + */ + x = @isa@_set_masked_lanes_ps(x, ones_f, inv_load_mask); +#endif + } + else { + x = @isa@_masked_gather_ps(ones_f, ip, vindex, load_mask); + } + @vtype@ out = @isa@_@vectorf@_ps(x); + @masked_store@(op, @cvtps_epi32@(load_mask), out); + + ip += num_lanes*stride; + op += num_lanes; + num_remaining_elements -= num_lanes; + } +} +#endif +/**end repeat1**/ +/**end repeat**/ + +/**begin repeat + * #ISA = FMA, AVX512F# + * #isa = fma, avx512# + * #vsize = 256, 512# + * #BYTES = 32, 64# + * #cvtps_epi32 = _mm256_cvtps_epi32, # + * #mask = __m256i, __mmask8# + * #vsub = , _mask# + * #vtype = __m256d, __m512d# + * #vindextype = __m128i, __m256i# + * #vindexsize = 128, 256# + * #vindexload = _mm_loadu_si128, _mm256_loadu_si256# + * #cvtps_epi32 = _mm256_cvtpd_epi32, # + * #castmask = _mm256_castsi256_pd, # + * #masked_store = _mm256_maskstore_pd, _mm512_mask_storeu_pd# + * #CHK = HAVE_ATTRIBUTE_TARGET_AVX2_WITH_INTRINSICS, HAVE_ATTRIBUTE_TARGET_AVX512F_WITH_INTRINSICS# + */ + +/**begin repeat1 + * #func = sqrt, absolute, square, reciprocal, rint, ceil, floor, trunc# + * #vectorf = sqrt, abs, square, reciprocal, rint, ceil, floor, trunc# + * #replace_0_with_1 = 0, 0, 0, 1, 0, 0, 0, 0# + */ + +#if defined @CHK@ +static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void +@ISA@_@func@_DOUBLE(npy_double* op, + npy_double* ip, + const npy_intp array_size, + const npy_intp steps) +{ + const npy_intp stride = steps/sizeof(npy_double); + const npy_int num_lanes = @BYTES@/sizeof(npy_double); + npy_intp num_remaining_elements = array_size; + @mask@ load_mask = @isa@_get_full_load_mask_pd(); +#if @replace_0_with_1@ + @mask@ inv_load_mask = @isa@_invert_mask_pd(load_mask); +#endif + @vtype@ ones_d = _mm@vsize@_set1_pd(1.0f); + npy_int indexarr[8]; + for (npy_int ii = 0; ii < 8; ii++) { + indexarr[ii] = ii*stride; + } + @vindextype@ vindex = @vindexload@((@vindextype@*)&indexarr[0]); + + while (num_remaining_elements > 0) { + if (num_remaining_elements < num_lanes) { + load_mask = @isa@_get_partial_load_mask_pd(num_remaining_elements, + num_lanes); +#if @replace_0_with_1@ + inv_load_mask = @isa@_invert_mask_pd(load_mask); #endif + } + @vtype@ x; + if (stride == 1) { + x = @isa@_masked_load_pd(load_mask, ip); +#if @replace_0_with_1@ + /* + * Replace masked elements with 1.0f to avoid divide by zero fp + * exception in reciprocal + */ + x = @isa@_set_masked_lanes_pd(x, ones_d, @castmask@(inv_load_mask)); +#endif + } + else { + x = @isa@_masked_gather_pd(ones_d, ip, vindex, @castmask@(load_mask)); + } + @vtype@ out = @isa@_@vectorf@_pd(x); + @masked_store@(op, load_mask, out); + + ip += num_lanes*stride; + op += num_lanes; + num_remaining_elements -= num_lanes; + } +} +#endif +/**end repeat1**/ /**end repeat**/ /**begin repeat @@ -1460,7 +1845,6 @@ static NPY_INLINE NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ @vtype@ * #CHK = HAVE_ATTRIBUTE_TARGET_AVX2_WITH_INTRINSICS, HAVE_ATTRIBUTE_TARGET_AVX512F_WITH_INTRINSICS# */ - /* * Vectorized approximate sine/cosine algorithms: The following code is a * vectorized version of the algorithm presented here: @@ -1519,7 +1903,7 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void @vtype@ quadrant, reduced_x, reduced_x2, cos, sin; @vtype@i iquadrant; @mask@ nan_mask, glibc_mask, sine_mask, negate_mask; - @mask@ load_mask = @isa@_get_full_load_mask(); + @mask@ load_mask = @isa@_get_full_load_mask_ps(); npy_intp num_remaining_elements = array_size; npy_int indexarr[16]; for (npy_int ii = 0; ii < 16; ii++) { @@ -1530,16 +1914,16 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void while (num_remaining_elements > 0) { if (num_remaining_elements < num_lanes) { - load_mask = @isa@_get_partial_load_mask(num_remaining_elements, + load_mask = @isa@_get_partial_load_mask_ps(num_remaining_elements, num_lanes); } @vtype@ x; if (stride == 1) { - x = @isa@_masked_load(load_mask, ip); + x = @isa@_masked_load_ps(load_mask, ip); } else { - x = @isa@_masked_gather(zero_f, ip, vindex, load_mask); + x = @isa@_masked_gather_ps(zero_f, ip, vindex, load_mask); } /* @@ -1551,7 +1935,7 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void glibc_mask = @isa@_in_range_mask(x, large_number,-large_number); glibc_mask = @and_masks@(load_mask, glibc_mask); nan_mask = _mm@vsize@_cmp_ps@vsub@(x, x, _CMP_NEQ_UQ); - x = @isa@_set_masked_lanes(x, zero_f, @or_masks@(nan_mask, glibc_mask)); + x = @isa@_set_masked_lanes_ps(x, zero_f, @or_masks@(nan_mask, glibc_mask)); npy_int iglibc_mask = @mask_to_int@(glibc_mask); if (iglibc_mask != @full_mask@) { @@ -1584,7 +1968,7 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void /* multiply by -1 for appropriate elements */ negate_mask = @isa@_should_negate(iquadrant, twos, twos); cos = @isa@_blend(cos, _mm@vsize@_sub_ps(zero_f, cos), negate_mask); - cos = @isa@_set_masked_lanes(cos, _mm@vsize@_set1_ps(NPY_NANF), nan_mask); + cos = @isa@_set_masked_lanes_ps(cos, _mm@vsize@_set1_ps(NPY_NANF), nan_mask); @masked_store@(op, @cvtps_epi32@(load_mask), cos); } @@ -1662,27 +2046,27 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void @vtype@i vindex = _mm@vsize@_loadu_si@vsize@((@vtype@i*)&indexarr[0]); @mask@ xmax_mask, xmin_mask, nan_mask, inf_mask; - @mask@ overflow_mask = @isa@_get_partial_load_mask(0, num_lanes); - @mask@ load_mask = @isa@_get_full_load_mask(); + @mask@ overflow_mask = @isa@_get_partial_load_mask_ps(0, num_lanes); + @mask@ load_mask = @isa@_get_full_load_mask_ps(); npy_intp num_remaining_elements = array_size; while (num_remaining_elements > 0) { if (num_remaining_elements < num_lanes) { - load_mask = @isa@_get_partial_load_mask(num_remaining_elements, - num_lanes); + load_mask = @isa@_get_partial_load_mask_ps(num_remaining_elements, + num_lanes); } @vtype@ x; if (stride == 1) { - x = @isa@_masked_load(load_mask, ip); + x = @isa@_masked_load_ps(load_mask, ip); } else { - x = @isa@_masked_gather(zeros_f, ip, vindex, load_mask); + x = @isa@_masked_gather_ps(zeros_f, ip, vindex, load_mask); } nan_mask = _mm@vsize@_cmp_ps@vsub@(x, x, _CMP_NEQ_UQ); - x = @isa@_set_masked_lanes(x, zeros_f, nan_mask); + x = @isa@_set_masked_lanes_ps(x, zeros_f, nan_mask); xmax_mask = _mm@vsize@_cmp_ps@vsub@(x, _mm@vsize@_set1_ps(xmax), _CMP_GE_OQ); xmin_mask = _mm@vsize@_cmp_ps@vsub@(x, _mm@vsize@_set1_ps(xmin), _CMP_LE_OQ); @@ -1690,7 +2074,7 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void overflow_mask = @or_masks@(overflow_mask, @xor_masks@(xmax_mask, inf_mask)); - x = @isa@_set_masked_lanes(x, zeros_f, @or_masks@( + x = @isa@_set_masked_lanes_ps(x, zeros_f, @or_masks@( @or_masks@(nan_mask, xmin_mask), xmax_mask)); quadrant = _mm@vsize@_mul_ps(x, log2e); @@ -1723,9 +2107,9 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void * elem < xmin; return 0.0f * elem = +/- nan, return nan */ - poly = @isa@_set_masked_lanes(poly, _mm@vsize@_set1_ps(NPY_NANF), nan_mask); - poly = @isa@_set_masked_lanes(poly, inf, xmax_mask); - poly = @isa@_set_masked_lanes(poly, zeros_f, xmin_mask); + poly = @isa@_set_masked_lanes_ps(poly, _mm@vsize@_set1_ps(NPY_NANF), nan_mask); + poly = @isa@_set_masked_lanes_ps(poly, inf, xmax_mask); + poly = @isa@_set_masked_lanes_ps(poly, zeros_f, xmin_mask); @masked_store@(op, @cvtps_epi32@(load_mask), poly); @@ -1790,24 +2174,24 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void @vtype@ poly, num_poly, denom_poly, exponent; @mask@ inf_mask, nan_mask, sqrt2_mask, zero_mask, negx_mask; - @mask@ invalid_mask = @isa@_get_partial_load_mask(0, num_lanes); + @mask@ invalid_mask = @isa@_get_partial_load_mask_ps(0, num_lanes); @mask@ divide_by_zero_mask = invalid_mask; - @mask@ load_mask = @isa@_get_full_load_mask(); + @mask@ load_mask = @isa@_get_full_load_mask_ps(); npy_intp num_remaining_elements = array_size; while (num_remaining_elements > 0) { if (num_remaining_elements < num_lanes) { - load_mask = @isa@_get_partial_load_mask(num_remaining_elements, - num_lanes); + load_mask = @isa@_get_partial_load_mask_ps(num_remaining_elements, + num_lanes); } @vtype@ x_in; if (stride == 1) { - x_in = @isa@_masked_load(load_mask, ip); + x_in = @isa@_masked_load_ps(load_mask, ip); } else { - x_in = @isa@_masked_gather(zeros_f, ip, vindex, load_mask); + x_in = @isa@_masked_gather_ps(zeros_f, ip, vindex, load_mask); } negx_mask = _mm@vsize@_cmp_ps@vsub@(x_in, zeros_f, _CMP_LT_OQ); @@ -1818,7 +2202,7 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void @and_masks@(zero_mask, load_mask)); invalid_mask = @or_masks@(invalid_mask, negx_mask); - @vtype@ x = @isa@_set_masked_lanes(x_in, zeros_f, negx_mask); + @vtype@ x = @isa@_set_masked_lanes_ps(x_in, zeros_f, negx_mask); /* set x = normalized mantissa */ exponent = @isa@_get_exponent(x); @@ -1852,10 +2236,10 @@ static NPY_GCC_OPT_3 NPY_GCC_TARGET_@ISA@ void * x = +/- NAN; return NAN * x = 0.0f; return -INF */ - poly = @isa@_set_masked_lanes(poly, nan, nan_mask); - poly = @isa@_set_masked_lanes(poly, neg_nan, negx_mask); - poly = @isa@_set_masked_lanes(poly, neg_inf, zero_mask); - poly = @isa@_set_masked_lanes(poly, inf, inf_mask); + poly = @isa@_set_masked_lanes_ps(poly, nan, nan_mask); + poly = @isa@_set_masked_lanes_ps(poly, neg_nan, negx_mask); + poly = @isa@_set_masked_lanes_ps(poly, neg_inf, zero_mask); + poly = @isa@_set_masked_lanes_ps(poly, inf, inf_mask); @masked_store@(op, @cvtps_epi32@(load_mask), poly); diff --git a/numpy/core/tests/test_multiarray.py b/numpy/core/tests/test_multiarray.py index 9b124f603..66e3e3c60 100644 --- a/numpy/core/tests/test_multiarray.py +++ b/numpy/core/tests/test_multiarray.py @@ -3602,10 +3602,10 @@ class TestBinop(object): assert_equal(np.modf(dummy, out=(None, a)), (1,)) assert_equal(np.modf(dummy, out=(dummy, a)), (1,)) assert_equal(np.modf(a, out=(dummy, a)), 0) - with warnings.catch_warnings(record=True) as w: - warnings.filterwarnings('always', '', DeprecationWarning) - assert_equal(np.modf(dummy, out=a), (0,)) - assert_(w[0].category is DeprecationWarning) + with assert_raises(TypeError): + # Out argument must be tuple, since there are multiple outputs + np.modf(dummy, out=a) + assert_raises(ValueError, np.modf, dummy, out=(a,)) # 2 inputs, 1 output @@ -7975,6 +7975,8 @@ class TestFormat(object): dst = object.__format__(a, '30') assert_equal(res, dst) +from numpy.testing import IS_PYPY + class TestCTypes(object): def test_ctypes_is_available(self): @@ -8041,7 +8043,29 @@ class TestCTypes(object): # but when the `ctypes_ptr` object dies, so should `arr` del ctypes_ptr + if IS_PYPY: + # Pypy does not recycle arr objects immediately. Trigger gc to + # release arr. Cpython uses refcounts. An explicit call to gc + # should not be needed here. + break_cycles() + assert_(arr_ref() is None, "unknowable whether ctypes pointer holds a reference") + + def test_ctypes_as_parameter_holds_reference(self): + arr = np.array([None]).copy() + + arr_ref = weakref.ref(arr) + + ctypes_ptr = arr.ctypes._as_parameter_ + + # `ctypes_ptr` should hold onto `arr` + del arr break_cycles() + assert_(arr_ref() is not None, "ctypes pointer did not hold onto a reference") + + # but when the `ctypes_ptr` object dies, so should `arr` + del ctypes_ptr + if IS_PYPY: + break_cycles() assert_(arr_ref() is None, "unknowable whether ctypes pointer holds a reference") diff --git a/numpy/core/tests/test_umath.py b/numpy/core/tests/test_umath.py index a646e5e45..9b4ce9e47 100644 --- a/numpy/core/tests/test_umath.py +++ b/numpy/core/tests/test_umath.py @@ -687,8 +687,96 @@ class TestSpecialFloats(object): assert_raises(FloatingPointError, np.cos, np.float32(-np.inf)) assert_raises(FloatingPointError, np.cos, np.float32(np.inf)) + def test_sqrt_values(self): + with np.errstate(all='ignore'): + x = [np.nan, np.nan, np.inf, np.nan, 0.] + y = [np.nan, -np.nan, np.inf, -np.inf, 0.] + for dt in ['f', 'd', 'g']: + xf = np.array(x, dtype=dt) + yf = np.array(y, dtype=dt) + assert_equal(np.sqrt(yf), xf) + + #with np.errstate(invalid='raise'): + # for dt in ['f', 'd', 'g']: + # assert_raises(FloatingPointError, np.sqrt, np.array(-100., dtype=dt)) + + def test_abs_values(self): + x = [np.nan, np.nan, np.inf, np.inf, 0., 0., 1.0, 1.0] + y = [np.nan, -np.nan, np.inf, -np.inf, 0., -0., -1.0, 1.0] + for dt in ['f', 'd', 'g']: + xf = np.array(x, dtype=dt) + yf = np.array(y, dtype=dt) + assert_equal(np.abs(yf), xf) + + def test_square_values(self): + x = [np.nan, np.nan, np.inf, np.inf] + y = [np.nan, -np.nan, np.inf, -np.inf] + with np.errstate(all='ignore'): + for dt in ['f', 'd', 'g']: + xf = np.array(x, dtype=dt) + yf = np.array(y, dtype=dt) + assert_equal(np.square(yf), xf) + + with np.errstate(over='raise'): + assert_raises(FloatingPointError, np.square, np.array(1E32, dtype='f')) + assert_raises(FloatingPointError, np.square, np.array(1E200, dtype='d')) -class TestSIMDFloat32(object): + def test_reciprocal_values(self): + with np.errstate(all='ignore'): + x = [np.nan, np.nan, 0.0, -0.0, np.inf, -np.inf] + y = [np.nan, -np.nan, np.inf, -np.inf, 0., -0.] + for dt in ['f', 'd', 'g']: + xf = np.array(x, dtype=dt) + yf = np.array(y, dtype=dt) + assert_equal(np.reciprocal(yf), xf) + + with np.errstate(divide='raise'): + for dt in ['f', 'd', 'g']: + assert_raises(FloatingPointError, np.reciprocal, np.array(-0.0, dtype=dt)) + +# func : [maxulperror, low, high] +avx_ufuncs = {'sqrt' :[1, 0., 100.], + 'absolute' :[0, -100., 100.], + 'reciprocal' :[1, 1., 100.], + 'square' :[1, -100., 100.], + 'rint' :[0, -100., 100.], + 'floor' :[0, -100., 100.], + 'ceil' :[0, -100., 100.], + 'trunc' :[0, -100., 100.]} + +class TestAVXUfuncs(object): + def test_avx_based_ufunc(self): + strides = np.array([-4,-3,-2,-1,1,2,3,4]) + np.random.seed(42) + for func, prop in avx_ufuncs.items(): + maxulperr = prop[0] + minval = prop[1] + maxval = prop[2] + # various array sizes to ensure masking in AVX is tested + for size in range(1,32): + myfunc = getattr(np, func) + x_f32 = np.float32(np.random.uniform(low=minval, high=maxval, + size=size)) + x_f64 = np.float64(x_f32) + x_f128 = np.longdouble(x_f32) + y_true128 = myfunc(x_f128) + if maxulperr == 0: + assert_equal(myfunc(x_f32), np.float32(y_true128)) + assert_equal(myfunc(x_f64), np.float64(y_true128)) + else: + assert_array_max_ulp(myfunc(x_f32), np.float32(y_true128), + maxulp=maxulperr) + assert_array_max_ulp(myfunc(x_f64), np.float64(y_true128), + maxulp=maxulperr) + # various strides to test gather instruction + if size > 1: + y_true32 = myfunc(x_f32) + y_true64 = myfunc(x_f64) + for jj in strides: + assert_equal(myfunc(x_f64[::jj]), y_true64[::jj]) + assert_equal(myfunc(x_f32[::jj]), y_true32[::jj]) + +class TestAVXFloat32Transcendental(object): def test_exp_float32(self): np.random.seed(42) x_f32 = np.float32(np.random.uniform(low=0.0,high=88.1,size=1000000)) @@ -715,8 +803,8 @@ class TestSIMDFloat32(object): def test_strided_float32(self): np.random.seed(42) - strides = np.random.randint(low=-100, high=100, size=100) - sizes = np.random.randint(low=1, high=2000, size=100) + strides = np.array([-4,-3,-2,-1,1,2,3,4]) + sizes = np.arange(2,100) for ii in sizes: x_f32 = np.float32(np.random.uniform(low=0.01,high=88.1,size=ii)) exp_true = np.exp(x_f32) diff --git a/numpy/f2py/__init__.py b/numpy/f2py/__init__.py index d146739bb..42e3632fd 100644 --- a/numpy/f2py/__init__.py +++ b/numpy/f2py/__init__.py @@ -109,6 +109,7 @@ def compile(source, output = '' else: status = 0 + output = output.decode() if verbose: print(output) finally: diff --git a/numpy/f2py/cfuncs.py b/numpy/f2py/cfuncs.py index 17f3861ca..ccb7b3a32 100644 --- a/numpy/f2py/cfuncs.py +++ b/numpy/f2py/cfuncs.py @@ -1049,8 +1049,10 @@ static int create_cb_arglist(PyObject* fun,PyTupleObject* xa,const int maxnofarg CFUNCSMESS(\"create_cb_arglist\\n\"); tot=opt=ext=siz=0; /* Get the total number of arguments */ - if (PyFunction_Check(fun)) + if (PyFunction_Check(fun)) { tmp_fun = fun; + Py_INCREF(tmp_fun); + } else { di = 1; if (PyObject_HasAttrString(fun,\"im_func\")) { @@ -1062,6 +1064,7 @@ static int create_cb_arglist(PyObject* fun,PyTupleObject* xa,const int maxnofarg tmp_fun = PyObject_GetAttrString(tmp,\"im_func\"); else { tmp_fun = fun; /* built-in function */ + Py_INCREF(tmp_fun); tot = maxnofargs; if (xa != NULL) tot += PyTuple_Size((PyObject *)xa); @@ -1073,6 +1076,7 @@ static int create_cb_arglist(PyObject* fun,PyTupleObject* xa,const int maxnofarg if (xa != NULL) tot += PyTuple_Size((PyObject *)xa); tmp_fun = fun; + Py_INCREF(tmp_fun); } else if (F2PyCapsule_Check(fun)) { tot = maxnofargs; @@ -1083,6 +1087,7 @@ static int create_cb_arglist(PyObject* fun,PyTupleObject* xa,const int maxnofarg goto capi_fail; } tmp_fun = fun; + Py_INCREF(tmp_fun); } } if (tmp_fun==NULL) { @@ -1091,13 +1096,19 @@ goto capi_fail; } #if PY_VERSION_HEX >= 0x03000000 if (PyObject_HasAttrString(tmp_fun,\"__code__\")) { - if (PyObject_HasAttrString(tmp = PyObject_GetAttrString(tmp_fun,\"__code__\"),\"co_argcount\")) + if (PyObject_HasAttrString(tmp = PyObject_GetAttrString(tmp_fun,\"__code__\"),\"co_argcount\")) { #else if (PyObject_HasAttrString(tmp_fun,\"func_code\")) { - if (PyObject_HasAttrString(tmp = PyObject_GetAttrString(tmp_fun,\"func_code\"),\"co_argcount\")) + if (PyObject_HasAttrString(tmp = PyObject_GetAttrString(tmp_fun,\"func_code\"),\"co_argcount\")) { #endif - tot = PyInt_AsLong(PyObject_GetAttrString(tmp,\"co_argcount\")) - di; - Py_XDECREF(tmp); + PyObject *tmp_argcount = PyObject_GetAttrString(tmp,\"co_argcount\"); + Py_DECREF(tmp); + if (tmp_argcount == NULL) { + goto capi_fail; + } + tot = PyInt_AsLong(tmp_argcount) - di; + Py_DECREF(tmp_argcount); + } } /* Get the number of optional arguments */ #if PY_VERSION_HEX >= 0x03000000 @@ -1136,10 +1147,12 @@ goto capi_fail; PyTuple_SET_ITEM(*args,i,tmp); } CFUNCSMESS(\"create_cb_arglist-end\\n\"); + Py_DECREF(tmp_fun); return 1; capi_fail: if ((PyErr_Occurred())==NULL) PyErr_SetString(#modulename#_error,errmess); + Py_XDECREF(tmp_fun); return 0; } """ diff --git a/numpy/f2py/common_rules.py b/numpy/f2py/common_rules.py index 62c1ba207..f61d8810a 100644 --- a/numpy/f2py/common_rules.py +++ b/numpy/f2py/common_rules.py @@ -124,8 +124,9 @@ def buildhooks(m): cadd('\t%s(f2pyinit%s,F2PYINIT%s)(f2py_setup_%s);' % (F_FUNC, lower_name, name.upper(), name)) cadd('}\n') - iadd('\tF2PyDict_SetItemString(d, \"%s\", PyFortranObject_New(f2py_%s_def,f2py_init_%s));' % ( - name, name, name)) + iadd('\ttmp = PyFortranObject_New(f2py_%s_def,f2py_init_%s);' % (name, name)) + iadd('\tF2PyDict_SetItemString(d, \"%s\", tmp);' % name) + iadd('\tPy_DECREF(tmp);') tname = name.replace('_', '\\_') dadd('\\subsection{Common block \\texttt{%s}}\n' % (tname)) dadd('\\begin{description}') diff --git a/numpy/f2py/rules.py b/numpy/f2py/rules.py index 1b41498ea..f2f713bde 100755 --- a/numpy/f2py/rules.py +++ b/numpy/f2py/rules.py @@ -215,6 +215,7 @@ PyMODINIT_FUNC init#modulename#(void) { \td = PyModule_GetDict(m); \ts = PyString_FromString(\"$R""" + """evision: $\"); \tPyDict_SetItemString(d, \"__version__\", s); +\tPy_DECREF(s); #if PY_VERSION_HEX >= 0x03000000 \ts = PyUnicode_FromString( #else @@ -222,8 +223,14 @@ PyMODINIT_FUNC init#modulename#(void) { #endif \t\t\"This module '#modulename#' is auto-generated with f2py (version:#f2py_version#).\\nFunctions:\\n\"\n#docs#\".\"); \tPyDict_SetItemString(d, \"__doc__\", s); -\t#modulename#_error = PyErr_NewException (\"#modulename#.error\", NULL, NULL); \tPy_DECREF(s); +\t#modulename#_error = PyErr_NewException (\"#modulename#.error\", NULL, NULL); +\t/* +\t * Store the error object inside the dict, so that it could get deallocated. +\t * (in practice, this is a module, so it likely will not and cannot.) +\t */ +\tPyDict_SetItemString(d, \"_#modulename#_error\", #modulename#_error); +\tPy_DECREF(#modulename#_error); \tfor(i=0;f2py_routine_defs[i].name!=NULL;i++) { \t\ttmp = PyFortranObject_NewAsAttr(&f2py_routine_defs[i]); \t\tPyDict_SetItemString(d, f2py_routine_defs[i].name, tmp); @@ -238,7 +245,6 @@ PyMODINIT_FUNC init#modulename#(void) { \tif (! PyErr_Occurred()) \t\ton_exit(f2py_report_on_exit,(void*)\"#modulename#\"); #endif - \treturn RETVAL; } #ifdef __cplusplus @@ -439,12 +445,16 @@ rout_rules = [ { extern #ctype# #F_FUNC#(#name_lower#,#NAME#)(void); PyObject* o = PyDict_GetItemString(d,"#name#"); - PyObject_SetAttrString(o,"_cpointer", F2PyCapsule_FromVoidPtr((void*)#F_FUNC#(#name_lower#,#NAME#),NULL)); + tmp = F2PyCapsule_FromVoidPtr((void*)#F_FUNC#(#name_lower#,#NAME#),NULL); + PyObject_SetAttrString(o,"_cpointer", tmp); + Py_DECREF(tmp); #if PY_VERSION_HEX >= 0x03000000 - PyObject_SetAttrString(o,"__name__", PyUnicode_FromString("#name#")); + s = PyUnicode_FromString("#name#"); #else - PyObject_SetAttrString(o,"__name__", PyString_FromString("#name#")); + s = PyString_FromString("#name#"); #endif + PyObject_SetAttrString(o,"__name__", s); + Py_DECREF(s); } '''}, 'need': {l_not(l_or(ismoduleroutine, isdummyroutine)): ['F_WRAPPEDFUNC', 'F_FUNC']}, @@ -477,12 +487,16 @@ rout_rules = [ { extern void #F_FUNC#(#name_lower#,#NAME#)(void); PyObject* o = PyDict_GetItemString(d,"#name#"); - PyObject_SetAttrString(o,"_cpointer", F2PyCapsule_FromVoidPtr((void*)#F_FUNC#(#name_lower#,#NAME#),NULL)); + tmp = F2PyCapsule_FromVoidPtr((void*)#F_FUNC#(#name_lower#,#NAME#),NULL); + PyObject_SetAttrString(o,"_cpointer", tmp); + Py_DECREF(tmp); #if PY_VERSION_HEX >= 0x03000000 - PyObject_SetAttrString(o,"__name__", PyUnicode_FromString("#name#")); + s = PyUnicode_FromString("#name#"); #else - PyObject_SetAttrString(o,"__name__", PyString_FromString("#name#")); + s = PyString_FromString("#name#"); #endif + PyObject_SetAttrString(o,"__name__", s); + Py_DECREF(s); } '''}, 'need': {l_not(l_or(ismoduleroutine, isdummyroutine)): ['F_WRAPPEDFUNC', 'F_FUNC']}, @@ -794,10 +808,13 @@ if (#varname#_capi==Py_None) { if (#varname#_xa_capi==NULL) { if (PyObject_HasAttrString(#modulename#_module,\"#varname#_extra_args\")) { PyObject* capi_tmp = PyObject_GetAttrString(#modulename#_module,\"#varname#_extra_args\"); - if (capi_tmp) + if (capi_tmp) { #varname#_xa_capi = (PyTupleObject *)PySequence_Tuple(capi_tmp); - else + Py_DECREF(capi_tmp); + } + else { #varname#_xa_capi = (PyTupleObject *)Py_BuildValue(\"()\"); + } if (#varname#_xa_capi==NULL) { PyErr_SetString(#modulename#_error,\"Failed to convert #modulename#.#varname#_extra_args to tuple.\\n\"); return NULL; diff --git a/numpy/f2py/src/fortranobject.c b/numpy/f2py/src/fortranobject.c index b55385b50..8aa55555d 100644 --- a/numpy/f2py/src/fortranobject.c +++ b/numpy/f2py/src/fortranobject.c @@ -39,19 +39,33 @@ PyFortranObject_New(FortranDataDef* defs, f2py_void_func init) { int i; PyFortranObject *fp = NULL; PyObject *v = NULL; - if (init!=NULL) /* Initialize F90 module objects */ + if (init!=NULL) { /* Initialize F90 module objects */ (*(init))(); - if ((fp = PyObject_New(PyFortranObject, &PyFortran_Type))==NULL) return NULL; - if ((fp->dict = PyDict_New())==NULL) return NULL; + } + fp = PyObject_New(PyFortranObject, &PyFortran_Type); + if (fp == NULL) { + return NULL; + } + if ((fp->dict = PyDict_New()) == NULL) { + Py_DECREF(fp); + return NULL; + } fp->len = 0; - while (defs[fp->len].name != NULL) fp->len++; - if (fp->len == 0) goto fail; + while (defs[fp->len].name != NULL) { + fp->len++; + } + if (fp->len == 0) { + goto fail; + } fp->defs = defs; - for (i=0;i<fp->len;i++) + for (i=0;i<fp->len;i++) { if (fp->defs[i].rank == -1) { /* Is Fortran routine */ v = PyFortranObject_NewAsAttr(&(fp->defs[i])); - if (v==NULL) return NULL; + if (v==NULL) { + goto fail; + } PyDict_SetItemString(fp->dict,fp->defs[i].name,v); + Py_XDECREF(v); } else if ((fp->defs[i].data)!=NULL) { /* Is Fortran variable or array (not allocatable) */ if (fp->defs[i].type == NPY_STRING) { @@ -65,13 +79,16 @@ PyFortranObject_New(FortranDataDef* defs, f2py_void_func init) { fp->defs[i].type, NULL, fp->defs[i].data, 0, NPY_ARRAY_FARRAY, NULL); } - if (v==NULL) return NULL; + if (v==NULL) { + goto fail; + } PyDict_SetItemString(fp->dict,fp->defs[i].name,v); + Py_XDECREF(v); } - Py_XDECREF(v); + } return (PyObject *)fp; fail: - Py_XDECREF(v); + Py_XDECREF(fp); return NULL; } diff --git a/numpy/f2py/src/test/foomodule.c b/numpy/f2py/src/test/foomodule.c index 733fab0be..caf3590d4 100644 --- a/numpy/f2py/src/test/foomodule.c +++ b/numpy/f2py/src/test/foomodule.c @@ -115,7 +115,7 @@ static PyMethodDef foo_module_methods[] = { void initfoo() { int i; - PyObject *m, *d, *s; + PyObject *m, *d, *s, *tmp; import_array(); m = Py_InitModule("foo", foo_module_methods); @@ -125,11 +125,17 @@ void initfoo() { PyDict_SetItemString(d, "__doc__", s); /* Fortran objects: */ - PyDict_SetItemString(d, "mod", PyFortranObject_New(f2py_mod_def,f2py_init_mod)); - PyDict_SetItemString(d, "foodata", PyFortranObject_New(f2py_foodata_def,f2py_init_foodata)); - for(i=0;f2py_routines_def[i].name!=NULL;i++) - PyDict_SetItemString(d, f2py_routines_def[i].name, - PyFortranObject_NewAsAttr(&f2py_routines_def[i])); + tmp = PyFortranObject_New(f2py_mod_def,f2py_init_mod); + PyDict_SetItemString(d, "mod", tmp); + Py_DECREF(tmp); + tmp = PyFortranObject_New(f2py_foodata_def,f2py_init_foodata); + PyDict_SetItemString(d, "foodata", tmp); + Py_DECREF(tmp); + for(i=0;f2py_routines_def[i].name!=NULL;i++) { + tmp = PyFortranObject_NewAsAttr(&f2py_routines_def[i]); + PyDict_SetItemString(d, f2py_routines_def[i].name, tmp); + Py_DECREF(tmp); + } Py_DECREF(s); diff --git a/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c b/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c index 7f46303b0..978db4e69 100644 --- a/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c +++ b/numpy/f2py/tests/src/array_from_pyobj/wrapmodule.c @@ -49,9 +49,18 @@ static PyObject *f2py_rout_wrap_call(PyObject *capi_self, return NULL; rank = PySequence_Length(dims_capi); dims = malloc(rank*sizeof(npy_intp)); - for (i=0;i<rank;++i) - dims[i] = (npy_intp)PyInt_AsLong(PySequence_GetItem(dims_capi,i)); - + for (i=0;i<rank;++i) { + PyObject *tmp; + tmp = PySequence_GetItem(dims_capi, i); + if (tmp == NULL) { + goto fail; + } + dims[i] = (npy_intp)PyInt_AsLong(tmp); + Py_DECREF(tmp); + if (dims[i] == -1 && PyErr_Occurred()) { + goto fail; + } + } capi_arr_tmp = array_from_pyobj(type_num,dims,rank,intent|F2PY_INTENT_OUT,arr_capi); if (capi_arr_tmp == NULL) { free(dims); @@ -60,6 +69,10 @@ static PyObject *f2py_rout_wrap_call(PyObject *capi_self, capi_buildvalue = Py_BuildValue("N",capi_arr_tmp); free(dims); return capi_buildvalue; + +fail: + free(dims); + return NULL; } static char doc_f2py_rout_wrap_attrs[] = "\ @@ -97,7 +110,7 @@ static PyObject *f2py_rout_wrap_attrs(PyObject *capi_self, PyTuple_SetItem(dimensions,i,PyInt_FromLong(PyArray_DIM(arr,i))); PyTuple_SetItem(strides,i,PyInt_FromLong(PyArray_STRIDE(arr,i))); } - return Py_BuildValue("siOOO(cciii)ii",s,PyArray_NDIM(arr), + return Py_BuildValue("siNNO(cciii)ii",s,PyArray_NDIM(arr), dimensions,strides, (PyArray_BASE(arr)==NULL?Py_None:PyArray_BASE(arr)), PyArray_DESCR(arr)->kind, @@ -154,61 +167,69 @@ PyMODINIT_FUNC inittest_array_from_pyobj_ext(void) { PyDict_SetItemString(d, "__doc__", s); wrap_error = PyErr_NewException ("wrap.error", NULL, NULL); Py_DECREF(s); - PyDict_SetItemString(d, "F2PY_INTENT_IN", PyInt_FromLong(F2PY_INTENT_IN)); - PyDict_SetItemString(d, "F2PY_INTENT_INOUT", PyInt_FromLong(F2PY_INTENT_INOUT)); - PyDict_SetItemString(d, "F2PY_INTENT_OUT", PyInt_FromLong(F2PY_INTENT_OUT)); - PyDict_SetItemString(d, "F2PY_INTENT_HIDE", PyInt_FromLong(F2PY_INTENT_HIDE)); - PyDict_SetItemString(d, "F2PY_INTENT_CACHE", PyInt_FromLong(F2PY_INTENT_CACHE)); - PyDict_SetItemString(d, "F2PY_INTENT_COPY", PyInt_FromLong(F2PY_INTENT_COPY)); - PyDict_SetItemString(d, "F2PY_INTENT_C", PyInt_FromLong(F2PY_INTENT_C)); - PyDict_SetItemString(d, "F2PY_OPTIONAL", PyInt_FromLong(F2PY_OPTIONAL)); - PyDict_SetItemString(d, "F2PY_INTENT_INPLACE", PyInt_FromLong(F2PY_INTENT_INPLACE)); - PyDict_SetItemString(d, "NPY_BOOL", PyInt_FromLong(NPY_BOOL)); - PyDict_SetItemString(d, "NPY_BYTE", PyInt_FromLong(NPY_BYTE)); - PyDict_SetItemString(d, "NPY_UBYTE", PyInt_FromLong(NPY_UBYTE)); - PyDict_SetItemString(d, "NPY_SHORT", PyInt_FromLong(NPY_SHORT)); - PyDict_SetItemString(d, "NPY_USHORT", PyInt_FromLong(NPY_USHORT)); - PyDict_SetItemString(d, "NPY_INT", PyInt_FromLong(NPY_INT)); - PyDict_SetItemString(d, "NPY_UINT", PyInt_FromLong(NPY_UINT)); - PyDict_SetItemString(d, "NPY_INTP", PyInt_FromLong(NPY_INTP)); - PyDict_SetItemString(d, "NPY_UINTP", PyInt_FromLong(NPY_UINTP)); - PyDict_SetItemString(d, "NPY_LONG", PyInt_FromLong(NPY_LONG)); - PyDict_SetItemString(d, "NPY_ULONG", PyInt_FromLong(NPY_ULONG)); - PyDict_SetItemString(d, "NPY_LONGLONG", PyInt_FromLong(NPY_LONGLONG)); - PyDict_SetItemString(d, "NPY_ULONGLONG", PyInt_FromLong(NPY_ULONGLONG)); - PyDict_SetItemString(d, "NPY_FLOAT", PyInt_FromLong(NPY_FLOAT)); - PyDict_SetItemString(d, "NPY_DOUBLE", PyInt_FromLong(NPY_DOUBLE)); - PyDict_SetItemString(d, "NPY_LONGDOUBLE", PyInt_FromLong(NPY_LONGDOUBLE)); - PyDict_SetItemString(d, "NPY_CFLOAT", PyInt_FromLong(NPY_CFLOAT)); - PyDict_SetItemString(d, "NPY_CDOUBLE", PyInt_FromLong(NPY_CDOUBLE)); - PyDict_SetItemString(d, "NPY_CLONGDOUBLE", PyInt_FromLong(NPY_CLONGDOUBLE)); - PyDict_SetItemString(d, "NPY_OBJECT", PyInt_FromLong(NPY_OBJECT)); - PyDict_SetItemString(d, "NPY_STRING", PyInt_FromLong(NPY_STRING)); - PyDict_SetItemString(d, "NPY_UNICODE", PyInt_FromLong(NPY_UNICODE)); - PyDict_SetItemString(d, "NPY_VOID", PyInt_FromLong(NPY_VOID)); - PyDict_SetItemString(d, "NPY_NTYPES", PyInt_FromLong(NPY_NTYPES)); - PyDict_SetItemString(d, "NPY_NOTYPE", PyInt_FromLong(NPY_NOTYPE)); - PyDict_SetItemString(d, "NPY_USERDEF", PyInt_FromLong(NPY_USERDEF)); - - PyDict_SetItemString(d, "CONTIGUOUS", PyInt_FromLong(NPY_ARRAY_C_CONTIGUOUS)); - PyDict_SetItemString(d, "FORTRAN", PyInt_FromLong(NPY_ARRAY_F_CONTIGUOUS)); - PyDict_SetItemString(d, "OWNDATA", PyInt_FromLong(NPY_ARRAY_OWNDATA)); - PyDict_SetItemString(d, "FORCECAST", PyInt_FromLong(NPY_ARRAY_FORCECAST)); - PyDict_SetItemString(d, "ENSURECOPY", PyInt_FromLong(NPY_ARRAY_ENSURECOPY)); - PyDict_SetItemString(d, "ENSUREARRAY", PyInt_FromLong(NPY_ARRAY_ENSUREARRAY)); - PyDict_SetItemString(d, "ALIGNED", PyInt_FromLong(NPY_ARRAY_ALIGNED)); - PyDict_SetItemString(d, "WRITEABLE", PyInt_FromLong(NPY_ARRAY_WRITEABLE)); - PyDict_SetItemString(d, "UPDATEIFCOPY", PyInt_FromLong(NPY_ARRAY_UPDATEIFCOPY)); - PyDict_SetItemString(d, "WRITEBACKIFCOPY", PyInt_FromLong(NPY_ARRAY_WRITEBACKIFCOPY)); - - PyDict_SetItemString(d, "BEHAVED", PyInt_FromLong(NPY_ARRAY_BEHAVED)); - PyDict_SetItemString(d, "BEHAVED_NS", PyInt_FromLong(NPY_ARRAY_BEHAVED_NS)); - PyDict_SetItemString(d, "CARRAY", PyInt_FromLong(NPY_ARRAY_CARRAY)); - PyDict_SetItemString(d, "FARRAY", PyInt_FromLong(NPY_ARRAY_FARRAY)); - PyDict_SetItemString(d, "CARRAY_RO", PyInt_FromLong(NPY_ARRAY_CARRAY_RO)); - PyDict_SetItemString(d, "FARRAY_RO", PyInt_FromLong(NPY_ARRAY_FARRAY_RO)); - PyDict_SetItemString(d, "DEFAULT", PyInt_FromLong(NPY_ARRAY_DEFAULT)); - PyDict_SetItemString(d, "UPDATE_ALL", PyInt_FromLong(NPY_ARRAY_UPDATE_ALL)); + +#define ADDCONST(NAME, CONST) \ + s = PyInt_FromLong(CONST); \ + PyDict_SetItemString(d, NAME, s); \ + Py_DECREF(s) + + ADDCONST("F2PY_INTENT_IN", F2PY_INTENT_IN); + ADDCONST("F2PY_INTENT_INOUT", F2PY_INTENT_INOUT); + ADDCONST("F2PY_INTENT_OUT", F2PY_INTENT_OUT); + ADDCONST("F2PY_INTENT_HIDE", F2PY_INTENT_HIDE); + ADDCONST("F2PY_INTENT_CACHE", F2PY_INTENT_CACHE); + ADDCONST("F2PY_INTENT_COPY", F2PY_INTENT_COPY); + ADDCONST("F2PY_INTENT_C", F2PY_INTENT_C); + ADDCONST("F2PY_OPTIONAL", F2PY_OPTIONAL); + ADDCONST("F2PY_INTENT_INPLACE", F2PY_INTENT_INPLACE); + ADDCONST("NPY_BOOL", NPY_BOOL); + ADDCONST("NPY_BYTE", NPY_BYTE); + ADDCONST("NPY_UBYTE", NPY_UBYTE); + ADDCONST("NPY_SHORT", NPY_SHORT); + ADDCONST("NPY_USHORT", NPY_USHORT); + ADDCONST("NPY_INT", NPY_INT); + ADDCONST("NPY_UINT", NPY_UINT); + ADDCONST("NPY_INTP", NPY_INTP); + ADDCONST("NPY_UINTP", NPY_UINTP); + ADDCONST("NPY_LONG", NPY_LONG); + ADDCONST("NPY_ULONG", NPY_ULONG); + ADDCONST("NPY_LONGLONG", NPY_LONGLONG); + ADDCONST("NPY_ULONGLONG", NPY_ULONGLONG); + ADDCONST("NPY_FLOAT", NPY_FLOAT); + ADDCONST("NPY_DOUBLE", NPY_DOUBLE); + ADDCONST("NPY_LONGDOUBLE", NPY_LONGDOUBLE); + ADDCONST("NPY_CFLOAT", NPY_CFLOAT); + ADDCONST("NPY_CDOUBLE", NPY_CDOUBLE); + ADDCONST("NPY_CLONGDOUBLE", NPY_CLONGDOUBLE); + ADDCONST("NPY_OBJECT", NPY_OBJECT); + ADDCONST("NPY_STRING", NPY_STRING); + ADDCONST("NPY_UNICODE", NPY_UNICODE); + ADDCONST("NPY_VOID", NPY_VOID); + ADDCONST("NPY_NTYPES", NPY_NTYPES); + ADDCONST("NPY_NOTYPE", NPY_NOTYPE); + ADDCONST("NPY_USERDEF", NPY_USERDEF); + + ADDCONST("CONTIGUOUS", NPY_ARRAY_C_CONTIGUOUS); + ADDCONST("FORTRAN", NPY_ARRAY_F_CONTIGUOUS); + ADDCONST("OWNDATA", NPY_ARRAY_OWNDATA); + ADDCONST("FORCECAST", NPY_ARRAY_FORCECAST); + ADDCONST("ENSURECOPY", NPY_ARRAY_ENSURECOPY); + ADDCONST("ENSUREARRAY", NPY_ARRAY_ENSUREARRAY); + ADDCONST("ALIGNED", NPY_ARRAY_ALIGNED); + ADDCONST("WRITEABLE", NPY_ARRAY_WRITEABLE); + ADDCONST("UPDATEIFCOPY", NPY_ARRAY_UPDATEIFCOPY); + ADDCONST("WRITEBACKIFCOPY", NPY_ARRAY_WRITEBACKIFCOPY); + + ADDCONST("BEHAVED", NPY_ARRAY_BEHAVED); + ADDCONST("BEHAVED_NS", NPY_ARRAY_BEHAVED_NS); + ADDCONST("CARRAY", NPY_ARRAY_CARRAY); + ADDCONST("FARRAY", NPY_ARRAY_FARRAY); + ADDCONST("CARRAY_RO", NPY_ARRAY_CARRAY_RO); + ADDCONST("FARRAY_RO", NPY_ARRAY_FARRAY_RO); + ADDCONST("DEFAULT", NPY_ARRAY_DEFAULT); + ADDCONST("UPDATE_ALL", NPY_ARRAY_UPDATE_ALL); + +#undef ADDCONST( if (PyErr_Occurred()) Py_FatalError("can't initialize module wrap"); diff --git a/numpy/f2py/tests/test_compile_function.py b/numpy/f2py/tests/test_compile_function.py index 36abf05f9..40ea7997f 100644 --- a/numpy/f2py/tests/test_compile_function.py +++ b/numpy/f2py/tests/test_compile_function.py @@ -29,6 +29,7 @@ def setup_module(): @pytest.mark.parametrize( "extra_args", [['--noopt', '--debug'], '--noopt --debug', ''] ) +@pytest.mark.leaks_references(reason="Imported module seems never deleted.") def test_f2py_init_compile(extra_args): # flush through the f2py __init__ compile() function code path as a # crude test for input handling following migration from @@ -81,6 +82,9 @@ def test_f2py_init_compile(extra_args): return_check = import_module(modname) calc_result = return_check.foo() assert_equal(calc_result, 15) + # Removal from sys.modules, is not as such necessary. Even with + # removal, the module (dict) stays alive. + del sys.modules[modname] def test_f2py_init_compile_failure(): diff --git a/numpy/f2py/tests/util.py b/numpy/f2py/tests/util.py index d20dc5908..77cb612d0 100644 --- a/numpy/f2py/tests/util.py +++ b/numpy/f2py/tests/util.py @@ -31,6 +31,7 @@ except ImportError: # _module_dir = None +_module_num = 5403 def _cleanup(): @@ -59,13 +60,14 @@ def get_module_dir(): def get_temp_module_name(): # Assume single-threaded, and the module dir usable only by this thread + global _module_num d = get_module_dir() - for j in range(5403, 9999999): - name = "_test_ext_module_%d" % j - fn = os.path.join(d, name) - if name not in sys.modules and not os.path.isfile(fn + '.py'): - return name - raise RuntimeError("Failed to create a temporary module name") + name = "_test_ext_module_%d" % _module_num + _module_num += 1 + if name in sys.modules: + # this should not be possible, but check anyway + raise RuntimeError("Temporary module name already in use.") + return name def _memoize(func): diff --git a/numpy/lib/function_base.py b/numpy/lib/function_base.py index 46950bc95..c39c2eea1 100644 --- a/numpy/lib/function_base.py +++ b/numpy/lib/function_base.py @@ -1314,9 +1314,13 @@ def interp(x, xp, fp, left=None, right=None, period=None): Notes ----- - Does not check that the x-coordinate sequence `xp` is increasing. - If `xp` is not increasing, the results are nonsense. - A simple check for increasing is:: + The x-coordinate sequence is expected to be increasing, but this is not + explicitly enforced. However, if the sequence `xp` is non-increasing, + interpolation results are meaningless. + + Note that, since NaN is unsortable, `xp` also cannot contain NaNs. + + A simple check for `xp` being strictly increasing is:: np.all(np.diff(xp) > 0) diff --git a/numpy/lib/tests/test_io.py b/numpy/lib/tests/test_io.py index 6ee17c830..1181fe986 100644 --- a/numpy/lib/tests/test_io.py +++ b/numpy/lib/tests/test_io.py @@ -1871,7 +1871,7 @@ M 33 21.99 data = ["1, 1, 1, 1, -1.1"] * 50 mdata = TextIO("\n".join(data)) - converters = {4: lambda x: "(%s)" % x} + converters = {4: lambda x: "(%s)" % x.decode()} kwargs = dict(delimiter=",", converters=converters, dtype=[(_, int) for _ in 'abcde'],) assert_raises(ValueError, np.genfromtxt, mdata, **kwargs) diff --git a/tools/travis-test.sh b/tools/travis-test.sh index 6baa55817..6094f0ee6 100755 --- a/tools/travis-test.sh +++ b/tools/travis-test.sh @@ -88,7 +88,7 @@ run_test() if [ -n "$RUN_FULL_TESTS" ]; then export PYTHONWARNINGS="ignore::DeprecationWarning:virtualenv" - $PYTHON ../runtests.py -n -v --durations 10 --mode=full $COVERAGE_FLAG + $PYTHON -b ../runtests.py -n -v --durations 10 --mode=full $COVERAGE_FLAG else # disable --durations temporarily, pytest currently aborts # when that is used with python3.6-dbg @@ -32,7 +32,7 @@ envlist = [testenv] deps= -Ur{toxinidir}/test_requirements.txt changedir={envdir} -commands={envpython} {toxinidir}/runtests.py --mode=full {posargs:} +commands={envpython} -b {toxinidir}/runtests.py --mode=full {posargs:} [testenv:py37-not-relaxed-strides] basepython=python3.7 |