ENH: Vectorizing umath module using AVX-512 (open sourced from Intel Short Vector Math Library, SVML)

.circleci/config.yml

@@ -19,6 +19,12 @@ jobs:
           command: |
             if [[ -v CI_PULL_REQUEST ]] ; then git pull --ff-only origin "refs/pull/${CI_PULL_REQUEST//*pull\//}/merge" ; fi
 
+      - run:
+          name: update submodules
+          command: |
+            git submodule init
+            git submodule update
+
       - run:
           name: create virtual environment, install dependencies
           command: |

.gitattributes

@@ -11,6 +11,7 @@ numpy/linalg/lapack_lite/f2c.c linguist-vendored
 numpy/linalg/lapack_lite/f2c.h linguist-vendored
 tools/npy_tempita/* linguist-vendored
 numpy/core/include/numpy/libdivide/* linguist-vendored
+numpy/core/src/umath/svml/* linguist-vendored
 
 # Mark some files as generated
 numpy/linalg/lapack_lite/f2c_*.c linguist-generated

.gitmodules

@@ -1,3 +1,6 @@
 [submodule "doc/source/_static/scipy-mathjax"]
 	path = doc/source/_static/scipy-mathjax
 	url = https://github.com/scipy/scipy-mathjax.git
+[submodule "numpy/core/src/umath/svml"]
+	path = numpy/core/src/umath/svml
+	url = https://github.com/numpy/SVML.git

azure-pipelines.yml

@@ -23,6 +23,9 @@ stages:
     pool:
       vmImage: 'ubuntu-20.04'
     steps:
+    - script: |
+        git submodule update --init
+      displayName: 'Fetch submodules'
     - script: |
             if ! `gcc 2>/dev/null`; then
                 sudo apt install gcc
@@ -71,6 +74,9 @@ stages:
     pool:
       vmImage: 'ubuntu-20.04'
     steps:
+    - script: |
+        git submodule update --init
+      displayName: 'Fetch submodules'
     - script: |
             docker run -v $(pwd):/numpy -e CFLAGS="-msse2 -std=c99 -UNDEBUG" \
             -e F77=gfortran-5 -e F90=gfortran-5 quay.io/pypa/manylinux2014_i686 \
@@ -258,6 +264,9 @@ stages:
     pool:
       vmImage: 'ubuntu-20.04'
     steps:
+    - script: |
+        git submodule update --init
+      displayName: 'Fetch submodules'
     - script: |
             # create and activate conda environment
             conda env create -f environment.yml

doc/release/upcoming_changes/19478.performance.rst

@@ -0,0 +1,11 @@
+Vectorize umath module using AVX-512
+-------------------------------------
+
+By leveraging Intel Short Vector Math Library (SVML), 18 umath functions
+(``exp2``, ``log2``, ``log10``, ``expm1``, ``log1p``, ``cbrt``, ``sin``,
+``cos``, ``tan``, ``arcsin``, ``arccos``, ``arctan``, ``sinh``, ``cosh``,
+``tanh``, ``arcsinh``, ``arccosh``, ``arctanh``) are vectorized using AVX-512
+instruction set for both single and double precision implementations.  This
+change is currently enabled only for Linux users and on processors with
+AVX-512 instruction set.  It provides an average speed up of 32x and 14x for
+single and double precision functions respectively.