[CP2K:7187] Re: How to use GPU to accelerate a calculations

贾建峰 jjf_... at 163.com
Sat Nov 21 09:19:58 UTC 2015


Hi Ole,
It's a big trick. It work fine!!!.


thanks.


Jianfeng






At 2015-11-20 20:58:30, "Ole Schütt" <o... at schuett.name> wrote:

Hi Jianfeng,

try replacing -D__DBCSR_CUDA  with -D__DBCSR_ACC. That might just do the trick.

-Ole

Am Freitag, 20. November 2015 13:04:25 UTC+1 schrieb jjf... at yahoo.com.cn:
Dear all,
I have compiled my CP2K with:
DFLAGS   = -D__INTEL -D__FFTSG  -D__parallel -D__BLACS -D__SCALAPACK -D__FFTW3 -D__FFTMKL  -D__LIBINT -D__LIBXC2 -D__ACC -D__CUDAPW -D__DBCSR_CUDA  -D__LIBINT_MAX_AM=6  -D__LIBDERIV_MAX_AM1=5

I have revised the generate.py as following:
    triples  = combinations(23)                 # blocked H2O (benchmark)
    triples += combinations(6)                  # idem min basis
    triples += combinations(14,16,29)           # RPA water
    triples += combinations(5, 32, 13, 24, 26)
    triples += combinations(9, 32, 22)
    triples += combinations(32)
    triples += combinations(64)
    triples += combinations(78)
    triples += combinations(16,29,55)
    triples += combinations(13,32,13)
    triples += combinations(26,32,13)
    triples += combinations(13,32,26)
    triples += combinations(26,32,26)
    triples += combinations(13,32,9)
    triples += combinations(9,32,13)
    triples += combinations(26,32,9)
    triples += combinations(9,32,26)

However, I don't find the acceleration from MY GPU (K20m). The DBCSR STATISTICS was as following:

 COUNTER                                      CPU                  ACC      ACC%
 number of processed stacks                388804                    0       0.0
 matmuls inhomo. stacks                  12328621                    0       0.0
 matmuls total                          242005257                    0       0.0
 flops   9 x   21 x    9               1073834496                    0       0.0
 flops 224 x  224 x  224               1753350144                    0       0.0
 flops 224 x  224 x  245               1917726720                    0       0.0
 flops 224 x  245 x  224               1917726720                    0       0.0
 flops 245 x  224 x  224               1917726720                    0       0.0
 flops 224 x  245 x  245               2097513600                    0       0.0
 flops 245 x  224 x  245               2097513600                    0       0.0
 flops 245 x  245 x  224               2097513600                    0       0.0
 flops 245 x  245 x  245               2294155500                    0       0.0
 flops   9 x    9 x  213               2556480528                    0       0.0
 flops  26 x   21 x    9               3102188544                    0       0.0
 flops   9 x   21 x   26               3102188544                    0       0.0
 flops 224 x  224 x  256               4007657472                    0       0.0
 flops 224 x  256 x  224               4007657472                    0       0.0
 flops 256 x  224 x  224               4007657472                    0       0.0
 flops 224 x  245 x  256               4383375360                    0       0.0
 flops 224 x  256 x  245               4383375360                    0       0.0
 flops 245 x  224 x  256               4383375360                    0       0.0
 flops 245 x  256 x  224               4383375360                    0       0.0
 flops 256 x  224 x  245               4383375360                    0       0.0
 flops 256 x  245 x  224               4383375360                    0       0.0
 flops   9 x   21 x   13               4737435066                    0       0.0
 flops  13 x   21 x    9               4737435066                    0       0.0
 flops 245 x  245 x  256               4794316800                    0       0.0
 flops 245 x  256 x  245               4794316800                    0       0.0
 flops 256 x  245 x  245               4794316800                    0       0.0
 flops  26 x    9 x  213               7222105800                    0       0.0
 flops   9 x   26 x  213               7247226168                    0       0.0
 flops  26 x   21 x   26               8961878016                    0       0.0
 flops 256 x  256 x  224               9160359936                    0       0.0
 flops 224 x  256 x  256               9160359936                    0       0.0
 flops 256 x  224 x  256               9160359936                    0       0.0
 flops   9 x    9 x  256               9217732608                    0       0.0
 flops 929 x  224 x  224               9882062848                    0       0.0
 flops 256 x  256 x  245              10019143680                    0       0.0
 flops 245 x  256 x  256              10019143680                    0       0.0
 flops 256 x  245 x  256              10019143680                    0       0.0
 flops 929 x  224 x  245              10808506240                    0       0.0
 flops 929 x  245 x  224              10808506240                    0       0.0
 flops   9 x   13 x  213              11030981598                    0       0.0
 flops  13 x    9 x  213              11065522104                    0       0.0
 flops 929 x  245 x  245              11821803700                    0       0.0
 flops 224 x  224 x  929              11933057024                    0       0.0
 flops 224 x  245 x  929              13051781120                    0       0.0
 flops 245 x  224 x  929              13051781120                    0       0.0
 flops  26 x   21 x   13              13997099844                    0       0.0
 flops  13 x   21 x   26              13997099844                    0       0.0
 flops 245 x  245 x  929              14275385600                    0       0.0
 flops  13 x   21 x   13              14415243024                    0       0.0
 flops 256 x  256 x  256              20937965568                    0       0.0
 flops  26 x   26 x  213              21335565888                    0       0.0
 flops 929 x  224 x  256              22587572224                    0       0.0
 flops 929 x  256 x  224              22587572224                    0       0.0
 flops 929 x  245 x  256              24705157120                    0       0.0
 flops 929 x  256 x  245              24705157120                    0       0.0
 flops  26 x    9 x  256              26040268800                    0       0.0
 flops   9 x   26 x  256              26130843648                    0       0.0
 flops 224 x  256 x  929              27275558912                    0       0.0
 flops 256 x  224 x  929              27275558912                    0       0.0
 flops 924 x  224 x  224              29486628864                    0       0.0
 flops 245 x  256 x  929              29832642560                    0       0.0
 flops 256 x  245 x  929              29832642560                    0       0.0
 flops 924 x  224 x  245              32251000320                    0       0.0
 flops 924 x  245 x  224              32251000320                    0       0.0
 flops  13 x   26 x  213              32611122180                    0       0.0
 flops  26 x   13 x  213              32674620888                    0       0.0
 flops  13 x   13 x  213              33962737536                    0       0.0
 flops 924 x  245 x  245              35274531600                    0       0.0
 flops 224 x  224 x  924              35606495232                    0       0.0
 flops 224 x  245 x  924              38944604160                    0       0.0
 flops 245 x  224 x  924              38944604160                    0       0.0
 flops   9 x   13 x  256              39773680128                    0       0.0
 flops  13 x    9 x  256              39898220544                    0       0.0
 flops 245 x  245 x  924              42595660800                    0       0.0
 flops 929 x  224 x  929              47943653632                    0       0.0
 flops   9 x   32 x    9              49089576960                    0       0.0
 flops 929 x  256 x  256              51628736512                    0       0.0
 flops 929 x  245 x  929              52438371160                    0       0.0
 flops 256 x  256 x  929              62344134656                    0       0.0
 flops 924 x  256 x  224              67398008832                    0       0.0
 flops 924 x  224 x  256              67398008832                    0       0.0
 flops 924 x  256 x  245              73716572160                    0       0.0
 flops 924 x  245 x  256              73716572160                    0       0.0
 flops  26 x   26 x  256              76928237568                    0       0.0
 flops 224 x  256 x  924              81386274816                    0       0.0
 flops 256 x  224 x  924              81386274816                    0       0.0
 flops 245 x  256 x  924              89016238080                    0       0.0
 flops 256 x  245 x  924              89016238080                    0       0.0
 flops 929 x  256 x  929             109585494016                    0       0.0
 flops  13 x   26 x  256             117583764480                    0       0.0
 flops  26 x   13 x  256             117812717568                    0       0.0
 flops  13 x   13 x  256             122457194496                    0       0.0
 flops  26 x   32 x    9             141814333440                    0       0.0
 flops   9 x   32 x   26             141814333440                    0       0.0
 flops 924 x  224 x  929             143056843776                    0       0.0
 flops 929 x  224 x  924             143056843776                    0       0.0
 flops 924 x  256 x  256             154052591616                    0       0.0
 flops 924 x  245 x  929             156468422880                    0       0.0
 flops 929 x  245 x  924             156468422880                    0       0.0
 flops 256 x  256 x  924             186025771008                    0       0.0
 flops   9 x   32 x   13             216568460160                    0       0.0
 flops  13 x   32 x    9             216568460160                    0       0.0
 flops 924 x  256 x  929             326987071488                    0       0.0
 flops 929 x  256 x  924             326987071488                    0       0.0
 flops  26 x   32 x   26             409685852160                    0       0.0
 flops 924 x  224 x  924             426860679168                    0       0.0
 flops 924 x  245 x  924             466878867840                    0       0.0
 flops  26 x   32 x   13             639867421440                    0       0.0
 flops  13 x   32 x   26             639867421440                    0       0.0
 flops  13 x   32 x   13             658982538240                    0       0.0
 flops 924 x  256 x  924             975681552384                    0       0.0
 flops total                        9705794368534                    0       0.0
 marketing flops                   10521420653440

MY input file:
&GLOBAL
  PRINT_LEVEL  LOW
  PROJECT_NAME 1
  RUN_TYPE  ENERGY
&END GLOBAL
&MOTION
  &GEO_OPT
    OPTIMIZER  BFGS
    STEP_START_VAL  1
  &END GEO_OPT
  &END MOTION
&FORCE_EVAL
  METHOD  QS
  STRESS_TENSOR  ANALYTICAL
  &DFT
    &SCF
     MAX_SCF  50
     EPS_SCF   1.0E-7
    &OT  T
     MINIMIZER  DIIS
     ENERGY_GAP  0.002
     ALGORITHM   IRAC
     PRECONDITIONER  FULL_ALL
    &END OT
    &OUTER_SCF
     EPS_SCF 1.0E-7
     MAX_SCF 40
     STEP_SIZE 0.1
     EXTRAPOLATION_ORDER 4
    &END OUTER_SCF
   &END SCF
    &QS
      METHOD  GPW
    &END QS
    &MGRID
      CUTOFF   400
    &END MGRID
    &XC
         &XC_FUNCTIONAL  NO_SHORTCUT
        &PBE  T
        &END PBE
      &END XC_FUNCTIONAL
    &END XC
    &POISSON
     periodic xyz
     poisson_solver periodic
    &END POISSON     
  &END DFT
  &SUBSYS
    &CELL
A 16.707 0.00 0.00
B 0.00 15.580 0.00
C 0.00 0.00 30.
   PERIODIC  XYZ
   MULTIPLE_UNIT_CELL  1 1 1
 &END CELL
 &COORD
   &END COORD

anyone can help?


Jianfeng Jia






 

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com.
To post to this group, send email to cp... at googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20151121/5440d121/attachment.htm>


More information about the CP2K-user mailing list