Changes between Version 83 and Version 84 of DataParallel/BenchmarkStatus

Show
Ignore:
Timestamp:
12/05/10 17:01:06 (2 years ago)
Author:
benl
Comment:

--

Legend:

Unmodified
Added
Removed
Modified

  • DataParallel/BenchmarkStatus

    v83 v84  
    108108  || dph.dotp.vector.seq.N4 ||  68ms || 1 || 1 || || 
    109109  || dph.dotp.vectorised.seq.N4 || 58ms || 1.17 || || A || 
    110   || dph.dotp.vectorised.par.N1 || 55ms || 1.24 || 1.24 || || 
     110  || dph.dotp.vectorised.par.N1 || 55ms || 1.24 || 1.24 || B || 
    111111  || dph.dotp.vectorised.par.N2 || 33ms || 2.06 || 1.03 || || 
    112112  || dph.dotp.vectorised.par.N4 || 25ms || 2.72 || 0.68 || || 
    113113  
    114   A: The sequential vectorised version is faster than with Data.Vector. Why was this? 
     114  A: The core for the vectorised.seq version is equivalent to the vector version. We expect the backend has compiled it differently. Check this again with LLVM. 
     115  B: The vectorised.par version runs faster than vectorised.seq because the latter has a duplicate loop counter in the inner loop. We need a duplicate-loop-counter removal optimisation. 
    115116 
    116117  '''Status''': fine[[br]] 
     118  '''Todo''': Check again with LLVM. 
    117119 
    118120[[br]]