DD-AVX_v1

DD-AVX [beta] README (BSD License) written by Toshiaki Hishinuma (hishinuma@slis.tsukuba.ac.jp)

DD-AVX: Library of high-precision operations accelerated by AVX. This library has Double-Double (DD) precision operations accelerated by AVX and AVX2. This library needs to merge "Lis" for DD iterative solver. Official support "Lis" version is "lis-1.4.58". After merge "lis" and "DD-AVX", the interface is same as "lis".

What is this library

What is "DD-AVX ver.1"?

This library has Double-Double (DD) precision operations accelerated by AVX and AVX2.

This library includes...

C-code of DD precision operation kernels using AVX and AVX2 (vector operations and SpMV) [1] [2]. Matrix storage format "BSR" (Block CSR) for DD precision SpMV. block size 4x1 is suitable for AVX [3]. Auto selection of OpenMP loop Scheduling for DD precision SpMV [4].

Installing

Merging DD-AVX and lis

1. download "lis-1.4.58" from web site.
2. put "DD-AVX" and "lis-1.4.58" on same dir.
3. sh ./merge.sh [lis dir] [DD-AVX dir] [output dir]

Build

please type "Configure" and "make." example of configure and make is Configure.sh

1. configure

[command]$ ./configure [options]
	options
	 	CC=[compiler]
		CFLAGS=[compile options]
			[--enable-mpi]
			[--enable-omp]
			[--enable-sse2]
			[--enable-avx]
			[--enable-avx2]

2. make

$ make

Sample code (output_dir/test)

1. dont use MPI: test/test1 [options]
2. use MPI: mpirun -np [proc] test/test1 [options]

compile your code

please type following command.

$gcc a.c -L [output_dir]/src/.libs -I [output]/include