LAPACK Archives

[Lapack] Some question about Lapack Benchmark

Hi Cathy,
The TIMING programs needed to be updated, they will be back very soon into 
LAPACK.
To report performance, you should use FLOPS (number of floating point operation 
per second) , and certainly GFLOPS for your machine.
This is how performance are generally reported, see 
http://top500.org/list/2011/11/100 for example
All the operation counts and timing is done by the timing routine so that you 
just have to plot your data.
If you want to know the flop (number of floating point operation) of each 
routine, you can have a look at the SOPLA routine that was in LAPACK 3.0.
Also do not forget that most of the LAPACK performance will come from your 
optimized BLAS library.
Hope it helps
Sincerely,
Julie
On Feb 23, 2012, at 7:24 PM, nlf@Domain.Removed wrote:


Dear Sir/Madam,
     I am a graduate student of University of Science and Technology of China.
     In my study, I need to test the performance of Lapack 3.2.1. Timing 
programs provided with LAPACK 3.0 could be abtained, but LAPACK 3.2.1 does 
not provide timing programs. I wonder why don't you continute to provide the 
timing programs. 
     There are some difference between 3.0 and 3.2.1.I am trying to copy the 
corresponding LAPACK 3.0 routines and modify them. I found the performance 
data is reported using two statistics:run-time in seconds and speed in 
MFLOPS.Can I only use the run-time to measure measure performance? If speed 
is necessary, how to count flops correctly?
     By the way, is there any software or tool to test the performace of 
LAPACK?
     I am looking forward to your reply.
     Best wishes!
                                                                              
        Cathy Ning

--
  ???
  ???
  ??????????

_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/lapack


<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or