From 96c8941f6e6fdbad793bfeaa4f75be48a871566a Mon Sep 17 00:00:00 2001
From: yamadafuyuka <yamada.fuyuka@jp.fuijitsu.com>
Date: Mon, 17 Apr 2023 18:50:23 +0900
Subject: [PATCH] modify README.md

---
 README.md | 70 +++++++++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 60 insertions(+), 10 deletions(-)

diff --git a/README.md b/README.md
index 3322c7c..db5bdc7 100644
--- a/README.md
+++ b/README.md
@@ -1,18 +1,68 @@
-# `mkl_umath`
+`mkl_umath for SVE`
+====
 
-`mkl_umath._ufuncs` exposes [Intel(R) Math Kernel Library](https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/onemkl.html)
-powered version of loops used in the patched version of [NumPy](https://numpy.org), that used to be included in
-[Intel(R) Distribution for Python*](https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/distribution-for-python.html).
+This repository provides Python module `mkl_umath for SVE` to speed up
+transcendental-function calculation of NumPy on CPUs supporting the
+AArch64 SVE instruction set.
+`mkl_umath for SVE` is based on [mkl_umath](https://github.com/IntelPython/mkl_umath).
 
-Patches were factored out per community feedback ([NEP-36](https://numpy.org/neps/nep-0036-fair-play.html)).
+## Supported environment
 
-## Building
+- Linux running on AArch64 CPU which supports SVE instructions.
 
-Intel(R) C compiler and Intel(R) Math Kernel Library are required to build `mkl_umath` from source:
+## Requirement
+- Linux
+- cmake
+- Cython
+- GCC (10.2.x or later is mandatory, because ACLE (Arm C Language Extension) support is needed.)
+- ninja-build
+- NumPy (1.24.x or later is recommended.)
 
+## Build and Install
+### Build SLEEF
+
+Because [SLEEF (SIMD Library for for Evaluating Elementary Functions)](https://sleef.org)
+is used as the vectorized implementation of transcendental functions,
+SLEEF must be built beforehand.
+
+```sh
+git clone https://github.com/shibatch/sleef sleef
+mkdir build && cd build
+cmake -DCMAKE_INSTALL_PREFIX=directory_path_where_you_want_to_install_sleef -DENFORCE_SVE=TRUE -DBUILD_SHARED_LIBS=FALSE -G "Ninja" ../sleef
+ninja
+ninja test (This step can be skipped.)
+ninja install
+```
+
+### Build mkl_umath for SVE
 ```sh
-# ensure that MKL is installed, icc is activated
-export MKLROOT=$CONDA_PREFIX
-python setup.py config_cc --compiler=intelem build_ext --inplace
+git clone git@github.com:fujitsu/mkl_umath.git
+cd mkl_umath
+CC=gcc CXX=g++ SLEEF_PATH=directory_path_where_you_installed_sleef python setup.py build install
+```
+
+## Usage
+```python
+import numpy as np
+a = np.double(np.random.random_sample(1024))
+np.sin(a) # NumPy's transcendental function
+
+import mkl_umath as um
+um._patch.use_in_numpy()
+np.sin(a) # mkl_umath for SVE
 ```
 
+## LICENCE
+
+`mkl_umath for SVE` is licensed under
+[BSD 3-Clause "New" or "Revised" License](LICENSE).
+Refer to the ["LICENSE"](LICENSE) file for the full license text and copyright notice.
+
+This distribution includes and/or uses third party software governed by separate license terms.
+
+- BSD 3-Clause "New" or "Revised" License
+  - [NumPy](https://github.com/numpy/numpy)
+  - [mkl_umath](https://github.com/IntelPython/mkl_umath)
+- Boost Software License 1.0
+  - [SLEEF](https://github.com/shibatch/sleef)
+