Intel® Many Integrated Core Architecture (Intel MIC Architecture)

MPI startup fail


I am running a MPI program which involves 5 ranks communicating with each other. The program runs fine when i run it on Xeon platform but when i put one of the process rank on MIC the program crashes giving me a segmentation fault

mpi version: intel Mpi,

This is how i am setting up the libraries and compiler settings for mic

export LD_LIBRARY_PATH=/opt/intel/composerxe/lib/mic:$LD_LIBRARY_PATH

export I_MPI_MIC=1

scp /opt/intel/impi/ mic0:/bin
scp /opt/intel/impi/* mic0:/lib64/.

Offload error

I am trying to offload a part of code and using 3 arrays of dimensions (14000*3600)

but this is giving some error related to system calls if I reduce the size of arrays to 1400*360 the code works fine 

HOST--ERROR:myoiOSSetPageAccess: mprotect failed!

MPSS version confusion 3.1.x vs 2.1/gold_update_3_5...

Maybe it's just me, but the MPSS versioning seems potentially confusing.  I was recently updating my Compiler/VTUNE stuff from the "registration center" and saw that there's a new MPSS gold_update3_5 that just came out.  Reading through the readme-en.txt I start seeing references to MPSS 2.1.6720-23 and scratching my head.  Looking through my existing 3.1.2 readme it looks like 3.1.2 actually had newer stuff.  So I start poking around. Going through the site I get to the MPSS page which differences the 2.1 vs.

/usr and /usr/lib64 directories are world writable


I noticed that /usr and /usr/lib64 directories are world writable (mode 777), I'm using MPSS 3.1.2/4.

I commented out following line from /opt/intel/mic/ofed/ofed.filelist and after reboot permissions are how they should be.

# MPI fix
#slink /usr/lib64/ 777 0 0

And symbolic link is there.

cast __m512 to __m512d

Hey all,


simple question:


How does the cast operation _mm512_castps_pd work?

A __m512 data type holds 16 floats i.e. 16 elements. Contrary to that a __m512d data type can only hold 8 elements -- so what happens if I use the following instructions

__m512   a_ = _mm512_set1_ps( 2.0 );
__m512d b_ = _mm512_castps_pd( a_ );


Is it possible to load data from memory with _mm512_load_ps and then do a "cast operation" from float to double precision into two __m512d registers.



micflash keeps hanging


I've been struggling with MPSS installation for some time now.

I started out in Ubuntu cross-compiling, but then gave up for obvious reasons. (didn't seem worth the trouble)

I installed CentOS 6.4 minimally, and started building necessary stack from ground-up, finally installing MPSS 3.1.4.

After I finished I mistakenly issued command for C0 stepping (micflash without -smcbootloader option)

I knew it was a mistake because /sys/class/mic/mic0/stepping has 'B1' in it.

It obviously failed, and I found a similar case on Developer Zone 

订阅 Intel® Many Integrated Core Architecture (Intel MIC Architecture)