parallel_example

This is a small parallel programming experience for testing what the author has learned in the parallel program designing.

There are mainly three topics which includes the calculation of pi、the test of SPECOMP2012 and the calculation of matrix.

0x1.For the calculation of pi

we try to get the specific value as accurate as possible. Besides, it's also our goal to speed up the program with specific machine includeing AVX-float、AVX-double、SSE-float、SSE-double、serial and openmp versions. Set the compling argvs as follows:

gcc -mavx2 -msse2 -fopenmp -O3 pi.c -o result

Get the normal experience according to the results:

AVX-float>AVX-double≈SSE-float>SSE-double>serial

Details can be reached at topic-pi.docx

0x2.For the test of SPECOMP2012

As to the second, just test the SPECOMP2012 with different threads or setting the environment at dynamic ,static or guided. Details can be reached at topic-SPECOMP2012.docx

0x3.For the calculation of matrix

The last, may be the most familiar to us.The versions includes tranpose、blocked、openmp+tranpose and MPI. Details can be reached at topic-matrix.docx

gcc -mavx2 -msse2 -O3 -fopenmp matrix.c -o test

Assuming the matrix $A * B = C$, all are square matrices, $A=1_{nn}, B=2_{nn}$, $n=1024$, the result about time consumption is as follows:

Time: ms	1	2	3	4	average
origin	6602757.0	6284082.0	6060442.0	6298664.0	6311486.3
tranpose	914533.0	873473.0	931000.0	1010096.0	932275.5
blocked	6349623.0	5452277.0	6361220.0	6153948.0	6079267.0
openmp+tranpose	724606.5	796924.2	995741.2	809043.0	831578.7

Acknowledgement

If anything wrong you find, welcome to correct me.