OVERVIEW
========
VHCC is an efficient Sparse Matrix Vector Multiply (SpMV) implementation on Xeon Phi. It is particularly optimized for scale-free sparse matrices.

PREREQUISITES
=============
Intel Compiler version >= 13.1

HOW TO
======
- Type "make" to compile
- Copy the compiled "phi_spmv" binary to Xeon Phi
- Copy lib/libiomp5.so to Xeon Phi
- On Xeon Phi, export LD_LIBRARY_PATH with path to libiomp5.so
- Execute ./phi_spmv -f matrix.mtx -t 240 -r 512 -c 8192 -p 4

OPTIONS
=======
-f: matrix market format
-t: number of threads (suggested: 240)
-r: row blocking factor (suggested: 512)
-c: column blocking factor (suggested: 8192)
-p: number of panels (see below)

SUGGESTED PANEL NUMBERS
=======================
Suggested numbers of panels for our tested matrices
Stanford		5
IMDB			5
wiki-Talk		1
web-Google		2
connectus		40
NotreDame_actors	1
citationCiteseer	5
soc-sign-epinions	1
human_gene2		6
mouse_gene		15