[petsc-dev] [petsc-maint #87339] Re: ex19 on GPU

Barry Smith bsmith at mcs.anl.gov
Sun Sep 18 11:56:59 CDT 2011



   Ok, the copy up and down are not a problem. 

   Except for VecMAXPY() the vector operations are terrible (like they are not using the GPU, but they must be?) The MatMult() must be GPU because it is pretty good 2779??? 

   Does someone else have access to a similar system and can they run the exact same test to see what numbers they get? Satish, could you on breadboard? Maybe on Magellion :-)


   Barry



VecDot                 2 1.0 1.7049e-03 1.0 1.60e+05 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0    94
VecMDot             2024 1.0 8.6273e+00 1.0 2.54e+09 1.0 0.0e+00 0.0e+00 0.0e+00 50 29  0  0  0  66 29  0  0  0   295
VecNorm             2096 1.0 1.5544e+00 1.0 1.68e+08 1.0 0.0e+00 0.0e+00 0.0e+00  9  2  0  0  0  12  2  0  0  0   108
VecScale            2092 1.0 3.7774e-01 1.0 8.37e+07 1.0 0.0e+00 0.0e+00 0.0e+00  2  1  0  0  0   3  1  0  0  0   222
VecCopy             2072 1.0 3.8258e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  2  0  0  0  0   3  0  0  0  0     0
VecSet                70 1.0 1.3119e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
VecAXPY              108 1.0 4.7407e-02 1.0 8.64e+06 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0   182
VecWAXPY              68 1.0 1.2545e-02 1.0 2.72e+06 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0   217
VecMAXPY            2092 1.0 6.4464e-01 1.0 2.71e+09 1.0 0.0e+00 0.0e+00 0.0e+00  4 31  0  0  0   5 31  0  0  0  4198
VecScatterBegin        5 1.0 1.5609e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
VecReduceArith         2 1.0 3.8650e-03 1.0 1.60e+05 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0    41
VecReduceComm          1 1.0 0.0000e+00 0.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
VecCUSPCopyTo         49 1.0 3.0950e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
VecCUSPCopyFrom       44 1.0 2.0876e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
SNESSolve              1 1.0 1.3044e+01 1.0 8.87e+09 1.0 0.0e+00 0.0e+00 0.0e+00 75100  0  0  0 100100  0  0  0   680
SNESLineSearch         2 1.0 1.1921e-02 1.0 5.49e+06 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0   461
SNESFunctionEval       3 1.0 2.7192e-03 1.0 2.52e+06 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0   927
SNESJacobianEval       2 1.0 2.0424e-01 1.0 3.85e+07 1.0 0.0e+00 0.0e+00 0.0e+00  1  0  0  0  0   2  0  0  0  0   188
KSPGMRESOrthog      2024 1.0 9.2522e+00 1.0 5.09e+09 1.0 0.0e+00 0.0e+00 0.0e+00 53 57  0  0  0  71 57  0  0  0   550
KSPSetup               2 1.0 5.1975e-05 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
KSPSolve               2 1.0 1.2819e+01 1.0 8.83e+09 1.0 0.0e+00 0.0e+00 0.0e+00 74 99  0  0  0  98 99  0  0  0   689
PCSetUp                2 1.0 9.5367e-07 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
PCApply             2024 1.0 3.8054e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  2  0  0  0  0   3  0  0  0  0     0
MatMult             2092 1.0 1.1950e+00 1.0 3.32e+09 1.0 0.0e+00 0.0e+00 0.0e+00  7 37  0  0  0   9 37  0  0  0  2779

On Sep 18, 2011, at 10:29 AM, Shiyuan wrote:

> 
> 
> On Sat, Sep 17, 2011 at 10:48 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>  Run the first one  with -da_vec_type seqcusp and -da_mat_type seqaijcusp
> 
> > VecScatterBegin     2097 1.0 1.0270e+00 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  5  0  0  0  0   7  0  0  0  0     0
> > VecCUSPCopyTo       2140 1.0 2.4991e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  1  0  0  0  0   2  0  0  0  0     0
> > VecCUSPCopyFrom     2135 1.0 1.0437e+00 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  5  0  0  0  0   7  0  0  0  0     0
> 
>   Why is it doing all these vector copy ups and downs? It is run on one process it shouldn't be doing more than a handful total.
> 
>   Barry
> 
> ./ex19 -da_vec_type seqcusp -da_mat_type seqaijcusp -pc_type none -dmmg_nlevels 1 -da_grid_x 100 -da_grid_y 100 -log_summary -mat_no_inode -preload off  -cusp_synchronize -cuda_set_device 0 | tee ex19p2.txt
> 
> Summary of Stages:   ----- Time ------  ----- Flops -----  --- Messages ---  -- Message Lengths --  -- Reductions --
>                         Avg     %Total     Avg     %Total   counts   %Total     Avg         %Total   counts   %Total
>  0:      Main Stage: 4.2393e+00  24.4%  0.0000e+00   0.0%  0.000e+00   0.0%  0.000e+00        0.0%  0.000e+00   0.0%
>  1:           SetUp: 4.9079e-02   0.3%  0.0000e+00   0.0%  0.000e+00   0.0%  0.000e+00        0.0%  0.000e+00   0.0%
>  2:           Solve: 1.3071e+01  75.3%  8.8712e+09 100.0%  0.000e+00   0.0%  0.000e+00        0.0%  0.000e+00   0.0%
> 
> ------------------------------------------------------------------------------------------------------------------------
> 
> VecScatterBegin        5 1.0 1.5609e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
> VecReduceArith         2 1.0 3.8650e-03 1.0 1.60e+05 1.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0    41
> VecReduceComm          1 1.0 0.0000e+00 0.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
> VecCUSPCopyTo         49 1.0 3.0950e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
> VecCUSPCopyFrom       44 1.0 2.0876e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
> 
> The complete log is attached.  Thanks. 
> <ex19p2.txt>




More information about the petsc-dev mailing list