Re: [eigen] Help needed to run a benchmark on many machines

[ Thread Index | Date Index | More Archives ]

Wow, thanks a bunch!
It will take me a while to digest all this data... thanks again.

2015-02-19 9:37 GMT-05:00 Ilja Honkonen <ilja.j.honkonen@xxxxxxxx>:
I'm attaching a benchmark to this email. It tests all sorts of matrix
product sizes with all sorts of blocking parameters, and outputs both a
complete log and a table that might be what we need to directly provide
Eigen with.

I finally got through some of the queues on, attached are a few results.

If you run it, please record its output into a text file and send it
back to me together with some description of your device (e.g. "nexus
6"). The benchmark already records some hardware info such as
/proc/cpuinfo on linux.

The webpage above has quite a detailed description.

Please also tell me your Eigen changeset, compiler and command line.

I downloaded Eigen with hg about 24 h ago.

  - on intel, please compile with -mavx if available

Compile commands are in the compile.pbs files, I used march=native with gcc and xHost with intel. The eigen.o* files list flags that gcc used when compiling, you can get them with grep enabled eigen.o* | awk '{print $1}' | tr '\n' ' ' && echo. Looks like AVX and other suggestions were included where available. I got a complaint from intel when compiling:
In file included from ../../eigen/Eigen/Core(298),
                 from benchmark-blocking-sizes.cpp(16):
.../../eigen/Eigen/src/Core/arch/AVX/MathFunctions.h(15): warning #47: incompatible redefinition of macro "_mm256_setr_m128" (declared at line 1553 of "/nasa/intel/Compiler/2015.0.090/composer_xe_2015.0.090/compiler/include/immintrin.h")
  #define _mm256_setr_m128(lo, hi) \

PGI 14.2 compiled but the test coredumped immediately.


Mail converted by MHonArc 2.6.19+