Differences between revisions 1 and 6 (spanning 5 versions)
Revision 1 as of 2008-06-20 20:26:39
Size: 1665
Comment:
Revision 6 as of 2008-11-14 13:42:07
Size: 2352
Editor: anonymous
Comment: converted to 1.6 markup
Deletions are marked like this. Additions are marked like this.
Line 30: Line 30:
Line 42: Line 41:
 * RAM consumption for elimination seems lower than Magma, since we don't use any temporaries due to the lack of asymptotically fast elimination. (after you substract the static Sage RAM).
  * Magma: Total time: 340.579 seconds, Total memory usage: 1934.02MB (for 64000^2^ / 8 / 1024.0^2^ = 488.281MB)
 * newest benchmarks:
{{{#!python
sage: A = random_matrix(GF(2),6.4*10^4,6.4*10^4)
sage: time A.echelonize()
CPU times: user 357.87 s, sys: 1.26 s, total: 359.12 s
Wall time: 362.16
}}}
Line 43: Line 51:
{{{
> A:=RandomMatrix(GF(2),64*10^3, 64*10^3);
> time E:=EchelonForm(A);
Time: 336.350
}}}
Line 51: Line 64:

= Benchmark**ing =
 * found out that the mention of "mark**ing" is not allowed on this wiki

Flu

  • almost defeated mine

M4RI

Pronounciation

  • It is pronounced "MARI" now.

PLUQ Factorisation of Dense GF(2) Matrices

  • learned a lot from Clement

  • still work in progress, some initial code is written
  • nothing to be shown yet, but will keep working
  • work-in-progress, alpha, not working code will be released in a few days with the standard M4RI library

M4RI Improvements

  • autodetection of L1 and L2 cache
  • switch over to Strassen-Winograd Multiplication by default in Sage
  • learned a potential further improvement to multiplication from Bill Hart (needs to be implemented)
  • Performance on Core2Duo improved:

   1 sage: A = random_matrix(GF(2),3.2*10^4,3.2*10^4)
   2 sage: B = random_matrix(GF(2),3.2*10^4,3.2*10^4)
   3 sage: time C = A._multiply_strassen(B,cutoff=4092) #Old
   4 CPU times: user 51.32 s, sys: 0.14 s, total: 51.46 s
   5 Wall time: 51.86
   6 
   7 sage: time C = A._multiply_strassen(B,cutoff=8192) #New
   8 CPU times: user 44.93 s, sys: 0.15 s, total: 45.08 s
   9 Wall time: 45.32

   1 sage: A = random_matrix(GF(2),3.2*10^4,3.2*10^4)
   2 sage: time A.echelonize() #Old
   3 CPU times: user 53.67 s, sys: 0.05 s, total: 53.71 s
   4 Wall time: 53.99
   5 
   6 sage: A = random_matrix(GF(2),3.2*10^4,3.2*10^4)
   7 sage: time A.echelonize() #New
   8 CPU times: user 44.25 s, sys: 0.03 s, total: 44.29 s
   9 Wall time: 44.50
  • RAM consumption for elimination seems lower than Magma, since we don't use any temporaries due to the lack of asymptotically fast elimination. (after you substract the static Sage RAM).
    • Magma: Total time: 340.579 seconds, Total memory usage: 1934.02MB (for 640002 / 8 / 1024.02 = 488.281MB)

  • newest benchmarks:

   1 sage: A = random_matrix(GF(2),6.4*10^4,6.4*10^4)
   2 sage: time A.echelonize()
   3 CPU times: user 357.87 s, sys: 1.26 s, total: 359.12 s
   4 Wall time: 362.16

> A:=RandomMatrix(GF(2),64*10^3, 64*10^3);
> time E:=EchelonForm(A);
Time: 336.350

Parallel M4RI

  • Tried to implement parallel elimination and failed
  • If it worked however it would enable in the only parallel Gröbner basis engine (PolyBoRi) in commutative polynomial rings I'm aware of.

Review Process

  • Editor Meetings
  • Reviews

Benchmark**ing

  • found out that the mention of "mark**ing" is not allowed on this wiki

dev1/status/malb (last edited 2008-11-14 13:42:07 by anonymous)