Problem with multithreading in numpy with OpenBLAS

[gavru...@gmail.com]

Hi All,

I have installed numpy with OpenBLAS according this instruction

after making the OpenBlas the output was:

==================================================

OpenBLAS build complete. (BLAS CBLAS LAPACK LAPACKE)

  OS               ... Linux             

  Architecture     ... x86_64               

  BINARY           ... 64bit                 

  C compiler       ... GCC  (command line : gcc)

  Fortran compiler ... GFORTRAN  (command line : gfortran)

  Library Name     ... libopenblas_nehalemp-r0.2.9.rc1.a (Multi threaded; Max num-threads is 2)

To install the library, you can run "make PREFIX=/path/to/your/installation install".

==================================================

I have checked that everything is ok by running this test

But when I try to change the number of cores OpenBLAS utilizes via OPENBLAS_NUM_THREADS nothing happens(I see from htop that numpy still uses only one cpu). The same for GOTO_NUM_THREADSЪ and OMP_NUM_THREADS.

How can I resolve this problem?

[Zhang Xianyi]

Hi,

What's your CPU? Did you use a virtual machine?

Xianyi

2014-01-30 <gavru...@gmail.com>:

--
You received this message because you are subscribed to the Google Groups "OpenBLAS-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openblas-user...@googlegroups.com.
To post to this group, send email to openbla...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[gavru...@gmail.com]

Yes it is a virtual machine.

processor       : 0

vendor_id       : GenuineIntel

cpu family      : 6

model           : 26

model name      : Intel(R) Xeon(R) CPU           X5550  @ 2.67GHz

stepping        : 5

microcode       : 0x11

cpu MHz         : 2666.760

cache size      : 8192 KB

physical id     : 0

siblings        : 2

core id         : 0

cpu cores       : 1

apicid          : 0

initial apicid  : 4˙

fpu             : yes

fpu_exception   : yes

cpuid level     : 11

wp              : yes

flags           : fpu de tsc msr pae cx8 sep cmov pat clflush mmx fxsr sse sse2 ss ht syscall nx lm constant_tsc rep_good nopl nonstop_tsc pni ssse3 cx16 sse4_1 sse4_2 popcnt hypervisor lahf_lm dtherm

bogomips        : 5333.52

clflush size    : 64

cache_alignment : 64

address sizes   : 40 bits physical, 48 bits virtual

power management:

processor       : 1

vendor_id       : GenuineIntel

cpu family      : 6

model           : 26

model name      : Intel(R) Xeon(R) CPU           X5550  @ 2.67GHz

stepping        : 5

microcode       : 0x11

cpu MHz         : 2666.760

cache size      : 8192 KB

physical id     : 0

siblings        : 2

core id         : 0

cpu cores       : 1

apicid          : 0

initial apicid  : 4

fpu             : yes

fpu_exception   : yes

cpuid level     : 11

wp              : yes

flags           : fpu de tsc msr pae cx8 sep cmov pat clflush mmx fxsr sse sse2 ss ht syscall nx lm constant_tsc rep_good nopl nonstop_tsc pni ssse3 cx16 sse4_1 sse4_2 popcnt hypervisor lahf_lm dtherm

bogomips        : 5333.52

clflush size    : 64

cache_alignment : 64

address sizes   : 40 bits physical, 48 bits virtual

power management:

Пʼятниця, 31 січня 2014 р. 10:20:24 UTC+2 користувач Zhang Xianyi написав:

[gavru...@gmail.com]

By virtual I mean that it is amazon ec2 instance.

Неділя, 2 лютого 2014 р. 15:04:09 UTC+2 користувач gavru...@gmail.com написав:

[Zhang Xianyi]

Hi,

According to your /proc/cpuinfo, actually, it is one CPU core with hyper-threading. Thus, you can see 2 cores.

By default, if there are 2 hyper-threading logical cores, OpenBLAS will use only one core.

You can compile OpenBLAS with NO_AFFINITY=1 to disable this feature.

Xianyi

2014-02-03 <gavru...@gmail.com>:

[gavru...@gmail.com]

Thank you very much. Compilation with NO_AFFINITY=1 resolves my problem. 

So my questions are:

Will affinity increase performance in the case of real dual-core?

And what parameters will be optimal in the case of both multi-core and hyper-threaded system?

Понеділок, 3 лютого 2014 р. 16:53:02 UTC+2 користувач Zhang Xianyi написав:

[Carter Schonwald]

there are rules of thumb, but they can only help guide. When in doubt, benchmark! :)