From b4957baf7539f67123fa3b19e1c9a0f5e0e72380 Mon Sep 17 00:00:00 2001 From: Pratik Nayak Date: Wed, 30 Jan 2019 12:49:04 +0100 Subject: [PATCH] Add documentation to readme.html and ginkgo.html files --- doc/external-libs/ginkgo.html | 96 +++++++++++++++++++++++++++++++++++ doc/readme.html | 16 ++++++ 2 files changed, 112 insertions(+) create mode 100644 doc/external-libs/ginkgo.html diff --git a/doc/external-libs/ginkgo.html b/doc/external-libs/ginkgo.html new file mode 100644 index 0000000000..d918e17f73 --- /dev/null +++ b/doc/external-libs/ginkgo.html @@ -0,0 +1,96 @@ + + + + The deal.II Readme on interfacing to Ginkgo + + + + + + + +

Interfacing deal.II to Ginkgo

+ +

+ Ginkgo is + A numerical linear algebra software package that provides its users with + highly optimized fine grid level linear algebra operations. It currently has + the capability to solve on the CPU with the support of OpenMP and on the GPU + with NVIDIA's CUDA libraries. For some operations it uses NVIDIA's own libraries, CuSparse + and CuBLAS, but some routines also have self implemented CUDA functions.See the Ginkgo documentation + for more details. +

+ +

+ deal.II has wrapper classes to the linear algebra + parts of Ginkgo that provide almost the + same interfaces as the built-in deal.II linear algebra operations. +

+ +

Installing deal.II with Ginkgo

+ +

+ During the CMake configuration, the following flags should be specified: +

-DDEAL_II_WITH_GINKGO=ON, to enable the Ginkgo library. +
-DGINKGO_DIR=, to specify the path where Ginkgo has been installed. +

+ +

Installing Ginkgo

+ +

+ Installing Ginkgo is quite simple. Ginkgo currently uses CMake and hence one can use + the following commands to easily install Ginkgo. For different dependencies and compatible + libraries, please refer to Ginkgo. + The different flags that can be added are: +

-DBUILD_REFERENCE={ON,OFF}: Builds the reference single thread implementation + used to check the more sophisticated implementations of the GPU/CPU. +
-DBUILD_CUDA={ON,OFF}: Builds the GPU implementation, specifically the NVIDIA + CUDA implementations. This needs CUDA to be installed on the machine. +
-DBUILD_OMP={ON,OFF}: Builds the multithreaded OpenMP implementations. +

+ To install Ginkgo, you would need to: +

+	git clone https://github.com/ginkgo-project/ginkgo.git
+  mkdir build; cd build 
+  cmake -DBUILD_REFERENCE=on/off -DBUILD_CUDA=on/off -DBUILD_OMP=on/off -DCMAKE_INSTALL_PREFIX=/path/to/install/ -DCMAKE_BUILD_TYPE=Debug
+	make install
+

+ +

Running the Ginkgo tests

+ +

+ Ginkgo uses gtests to run unit tests. The reference implementations are + non-optimized versions to which the OpenMP and CUDA versions are compared with for + correctness. + + The following command, executed in Ginkgo's build directory, launches its test suite. +

+	make test
+

+ The output should contain several lines of the form: +

+     Start  1: path/to/test
+ 1/13 Test  #1: path/to/test .............................   Passed    0.01 sec
+

+ To run only a specific test for more details, from the build directory, run +

+	./path/to/test
+

+ +

+ +

+ + diff --git a/doc/readme.html b/doc/readme.html index 2294bfbfa6..406981e5f7 100644 --- a/doc/readme.html +++ b/doc/readme.html @@ -423,6 +423,22 @@ about compatibility and configuration can be found here.

+ Ginkgo

+ Ginkgo + is a numerical linear algebra library with highly optimized kernels + for many core architectures with a focus on fine level parallelism. + It allows for easy switching of the executing paradigm (CUDA, OpenMP, .etc) + while providing a multitude of linear solvers and preconditioners + and extension to other linear operators with minimal changes required in the code. + To enable Ginkgo, pass -DGINKGO_DIR=/path/to/ginkgo to CMake when + configuring deal.II. For more detailed instructions, one can refer to + this page. +

Gmsh

-- 2.39.5