[sage-support] OpenBlas multithreaded or not ?

[Thierry Dumont]

Hi,

I am doing some performances measurements  with Sage/scipy.
It's matrix x matrix multiplication (not a very new benchmark....).

It appears that, on the same computer Sage version 8.0 uses Openblas
with multithreading, but not Sage version 8.1.beta3:



-Example (4 core, broadwell, 3.4ghz. Peak performance 217 Gflops !):
size of matrices: 640

  -> Sage version 8.0 :
  Gflops=  138.359946456  Portion peak perf =  0.6358  (top says %cpu=400)

  -> Sage version 8.1.beta3:
  Gflops=  48.5945679306  Portion peak perf =  0.2233 (%cpu=100)

Both of Sage versions where compiled from source.

My questions:
Will OpenBlas be fixed to 1 thread in the next versions of sage?

t.d.

-- 
You received this message because you are subscribed to the Google Groups 
"sage-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-support+unsubscr...@googlegroups.com.
To post to this group, send email to sage-support@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-support.
For more options, visit https://groups.google.com/d/optout.

# Broadwell architecture, 4 core, 3.4 Ghz
# Each cpu as 2 vector ports (AVX), each one doing 4 vector operations
#                    y=a*x + b (that is: 8 flops)  by clock cycle.
ncore = 4
freq  = 3.4*10^9
vector_ports = 2
perf  = 8*vector_ports*ncore*freq
print "Peak perf, Gflops:",perf
for n in [10*2^i for i in [1..10]]:
    c= random_matrix(RDF,n)
    d= random_matrix(RDF,n)
    t= timeit("c*d",seconds=True)
    flops= 2*n^3/t
    print n,t,"Gflops= ",flops/10^9," Portion peak perf = ",flops/perf
#

<<attachment: tdumont.vcf>>