Nvidia CUDA Toolkit: Difference between revisions

From UFRC
Jump to navigation Jump to search
← Older editNewer edit →
Taylor (talk | contribs)
606 edits
No edit summary
Taylor (talk | contribs)
606 edits
No edit summary
Line 130: Line 130:
|}}
|}}


==Example: CUBLAS vs. Optimized BLAS==
==CUDA Examples==
See the [[{{PAGENAME}}_Examples]] page for CUDA development examples.


Note that double-precision linear algebra is a less than ideal application for the GPUs.  Still, it is a functional example of using one of the available CUDA runtime libraries.
===cuda_bm.c===
{{#fileAnchor: cuda_bm.c}}
Download raw source of the [{{#fileLink: cuda_bm.c}} cuda_bm.c]
<source lang=c>
#include <stdio.h>
#include <stdlib.h>
#include <mkl.h>
#include <math.h>
#include <time.h>
#include <cublas.h>
void resuse(char *str);
double timeDiff( struct timespec *t1, struct timespec *t2)
{
  double T1, T2;
  T2 = (double)t2->tv_sec + (double)t2->tv_nsec / 1.0e9;
  T1 = (double)t1->tv_sec - (double)t1->tv_nsec / 1.0e9;
  return(T2 - T1);
}
main()
{
  int      dim = 9100;
  int      i,j,k;
  int      status;
  double    *psa, *psb, *psc;
  double    *sap, *sbp, *scp;
  double    *pda, *pdb, *pdc;
  double    *dap, *dbp, *dcp;
  double alpha  = 1.0;
  double beta    = 0.0;
  double gflops  = 0.0;
  float  deltaT  = 0.0;
  double gflopCnt = 2.0 * dim * dim * dim / 1.0e9;
  struct timespec t1;
  struct timespec t2;
  int      ptime();
  pda = NULL;
  pdb = NULL;
  pdc = NULL;
  psa = (double *) malloc(dim * dim * sizeof(*psa) );
  psb = (double *) malloc(dim * dim * sizeof(*psb) );
  psc = (double *) malloc(dim * dim * sizeof(*psc) );
  printf("Initializing Matrices...");
  clock_gettime(CLOCK_MONOTONIC, &t1);
  sap = psa;
  sbp = psb;
  scp = psc;
  for (i = 0; i < dim; i++)
    for (j = 0; j < dim; j++) {
      *sap++ = 1.0;
      *sbp++ = 1.0;
      *scp++ = 0.0;
    }
  clock_gettime(CLOCK_MONOTONIC, &t2);
  deltaT = timeDiff(&t1, &t2);
  printf("Done.    Elapsed Time = %6.4f secs\n", deltaT);
  fflush(stdout);
  printf("Starting parallel DGEMM...");
  fflush(stdout);
  clock_gettime(CLOCK_MONOTONIC, &t1);
  cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, dim, dim, dim, alpha, psa, dim, psb, dim, beta, psc, dim);
  clock_gettime(CLOCK_MONOTONIC, &t2);
  deltaT = timeDiff(&t1, &t2);
  printf("Done.  Elapsed Time = %6.4f secs\n", deltaT);
  printf("                                ");
  printf("GFlOP Rate  = %8.4f\n", gflopCnt/deltaT);
  if ( (float) dim - psc[0] > 1.0e-5  ||
      (float) dim - psc[dim*dim-1] > 1.0e-5 ) {
      printf("Error: Incorrect Results!\n");
      printf("C[%2d,%2d] = %10.4f\n", 1,1,psc[0]);
      printf("C[%2d,%2d] = %10.4f\n", dim,dim,psc[dim*dim-1]);
  }
  /* Initialize CUDA */
  printf("Initializing CUDA...");
  status = cublasInit();
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! CUBLAS initialization error\n");
      return EXIT_FAILURE;
  }
  printf("Done.\n");
  /* Re-initialize the matrices */
  printf("Re-initializing Matrices...");
  clock_gettime(CLOCK_MONOTONIC, &t1);
  sap = psa;
  sbp = psb;
  scp = psc;
  for (i = 0; i < dim; i++) {
    for (j = 0; j < dim; j++) {
      *sap++ = 1.0;
      *sbp++ = 1.0;
      *scp++ = 0.0;
    }
  }
  clock_gettime(CLOCK_MONOTONIC, &t2);
  deltaT = timeDiff(&t1, &t2);
  printf("Done.  Elapsed Time = %6.4f secs\n", deltaT);
  fflush(stdout);
  /* Allocate device memory for the matrices */
  printf("Starting CUDA DGEMM...");
  clock_gettime(CLOCK_MONOTONIC, &t1);
  status = cublasAlloc(dim*dim, sizeof(*pda), (void**) &pda);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device memory allocation error (A)\n");
      return EXIT_FAILURE;
  }
  status = cublasAlloc(dim*dim, sizeof(*pdb), (void**) &pdb);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device memory allocation error (B)\n");
      return EXIT_FAILURE;
  }
  status = cublasAlloc(dim*dim, sizeof(*pdc), (void**) &pdc);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device memory allocation error (C)\n");
      return EXIT_FAILURE;
  }
  /* Initialize the device matrices with the host matrices */
  status = cublasSetVector(dim*dim, sizeof(*psa), psa, 1, pda, 1);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device access error (write A)\n");
      return EXIT_FAILURE;
  }
  status = cublasSetVector(dim*dim, sizeof(*pdb), psb, 1, pdb, 1);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device access error (write B)\n");
      return EXIT_FAILURE;
  }
  status = cublasSetVector(dim*dim, sizeof(*psc), psc, 1, pdc, 1);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device access error (write C)\n");
      return EXIT_FAILURE;
  }
  /* Clear last error */
  cublasGetError();
  /* Performs operation using cublas */
  cublasDgemm('n', 'n', dim, dim, dim, alpha, pda, dim, pdb, dim, beta, pdc, dim);
  status = cublasGetError();
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! kernel execution error.\n");
      return EXIT_FAILURE;
  }
  /* Read the result back */
  status = cublasGetVector(dim*dim, sizeof(*psc), pdc, 1, psc, 1);
  if (status != CUBLAS_STATUS_SUCCESS) {
      fprintf (stderr, "!!!! device access error (read C)\n");
      return EXIT_FAILURE;
  }
  clock_gettime(CLOCK_MONOTONIC, &t2);
  deltaT = timeDiff(&t1, &t2);
  printf("Done.      Elapsed Time = %6.4f secs\n", deltaT);
  printf("                                  ");
  printf("GFlOP Rate  = %8.4f\n", gflopCnt/deltaT);
  if ( (float) dim - psc[0] > 1.0e-5  ||
      (float) dim - psc[dim*dim-1] > 1.0e-5 ) {
      printf("Error: Incorrect Results!\n");
      printf("C[%2d,%2d] = %10.4f\n", 1,1,psc[0]);
      printf("C[%2d,%2d] = %10.4f\n", dim,dim,psc[dim*dim-1]);
  }
}
</source>
===resuse.c===
{{#fileAnchor: resuse.c}}
Download raw source of the [{{#fileLink: resuse.c}} resuse.c]
<source lang=c>
/*---------------------------------------------------------------------------*
*                      resuse.c                                            *
*                                                                          *
* resuse_(desc)                                                              *
* char *desc;                                                              *
*---------------------------------------------------------------------------*
* This routine makes use of the getrusage system call to collect and        *
* display resource utilization from one point in a program to another.      *
* Thus, it works somewhat like a stopwatch in the since that the first call *
* initializes, or starts, the resource collection over the interval.  The  *
* second call stops the collection and displays the results accumulated    *
* since the previous call.  This means, of course, that two calls are      *
* required for each section of code to be monitored.                        *
*
* The underscore at the end of the routine name is there so that the routine*
* may be called as an integer valued FORTRAN function name RESUSE(), under  *
* both the SunOS and Ultrix f77 compilers.  AIX inter-language calling      *
* conventions are different so the routine must be referenced as RESUSE_()  *
* under AIX (RISC/6000) FORTRAN (xlf).
*
* From a FORTRAN routine:
*
* INT = RESUSE("Some String")
*
*    fortran code to be timed
*
* INT = RESUSE("Some String")
*
* Just ignore the returned value.
*---------------------------------------------------------------------------*/
#include <sys/time.h>
#include <sys/resource.h>
static int first_call = 1;
static struct rusage initial;
void resuse(str)
char *str;
{
  int    pgminor, pgmajor, nswap, nvcsw, nivcsw;
  int    inblock, oublock;
  struct  rusage final;
  float  usr, sys, secs;
  getrusage(RUSAGE_SELF, &final);
  if ( ! first_call )
  {
    secs = final.ru_utime.tv_sec + final.ru_utime.tv_usec / 1000000.0;
    usr  = secs;
    secs = initial.ru_utime.tv_sec + initial.ru_utime.tv_usec / 1000000.0;
    usr  = usr - secs;
    secs = final.ru_stime.tv_sec + final.ru_stime.tv_usec / 1000000.0;
    sys  = secs;
    secs = initial.ru_stime.tv_sec + initial.ru_stime.tv_usec / 1000000.0;
    sys  = sys - secs;
    pgminor = final.ru_minflt  - initial.ru_minflt;
    pgmajor = final.ru_majflt  - initial.ru_majflt;
    nswap  = final.ru_nswap  - initial.ru_nswap;
    inblock = final.ru_inblock - initial.ru_inblock;
    oublock = final.ru_oublock - initial.ru_oublock;
    nvcsw  = final.ru_nvcsw  - initial.ru_nvcsw ;
    nivcsw  = final.ru_nivcsw  - initial.ru_nivcsw ;
    printf("=============================================================\n");
    printf("%s: Resource Usage Data...\n", str);
    printf("-------------------------------------------------------------\n");
    printf("User Time  (secs)          : %10.3f\n", usr);
    printf("System Time (secs)          : %10.3f\n", sys);
    printf("Total Time  (secs)          : %10.3f\n", usr + sys);
    printf("Minor Page Faults          : %10d\n", pgminor);
    printf("Major Page Faults          : %10d\n", pgmajor);
    printf("Swap Count                  : %10d\n", nswap);
    printf("Voluntary Context Switches  : %10d\n", nvcsw);
    printf("Involuntary Context Switches: %10d\n", nivcsw);
    printf("Block Input Operations      : %10d\n", inblock);
    printf("Block Output Operations    : %10d\n", oublock);
    printf("=============================================================\n");
  }
  else
  {
    printf("=============================================================\n");
    printf("%s: Collecting Resource Usage Data\n", str);
    printf("=============================================================\n");
  }
  first_call = !first_call;
  initial = final;
  return;
}
</source>
===Makefile for Building cuda_bm===
{{#fileAnchor: Makefile}}
Download raw source of the [{{#fileLink: Makefile}} Makefile]
<source lang=make>
#
# Makefile
#
CC          = icc
CFLAGS      = -O3 -openmp
MKL_INCS    = -I$(HPC_MKL_INC)
MKL_LIBS    = -L$(HPC_MKL_LIB) -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
MKL_MP_LIBS = -L$(HPC_MKL_LIB) -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core
CUDA_INCS  = -I$(HPC_CUDA_INC)
CUDA_LIBS  = -L$(HPC_CUDA_LIB) -lcublas
LIBS        = $(MKL_MP_LIBS) $(CUDA_LIBS) -lpthread -lrt
INCS        = $(MKL_INCS) $(CUDA_INCS)
bm: bm.o ptime.o resuse.o
$(CC) $(CFLAGS)  -o bm bm.o ptime.o resuse.o $(LIBS)
cuda_bm: cuda_bm.c resuse.o
$(CC) $(CFLAGS) $(INCS) -o cuda_bm cuda_bm.c resuse.o $(LIBS)
cuda_sp_bm: cuda_sp_bm.c resuse.o
$(CC) $(CFLAGS) $(INCS) -o cuda_sp_bm cuda_sp_bm.c resuse.o $(LIBS)
clean:
rm -f *.o core.* bm cuda_bm
</source>
=== Compiling and Linking ===
Download the above three files into a single directory.  Then type:
<pre>
[taylor@c11a-s15 Bench]$ module load intel/2013
[taylor@c11a-s15 Bench]$ module load cuda/5.5
[taylor@c11a-s15 Bench]$ make clean
rm -f *.o core.* bm cuda_bm
[taylor@c11a-s15 Bench]$ make cuda_bm
icc -O3 -openmp -I/opt/intel/composer_xe_2013_sp1.0.080/mkl/include  -c -o resuse.o resuse.c
icc -O3 -openmp -I/opt/intel/composer_xe_2013_sp1.0.080/mkl/include -I/opt/cuda/5.5/include -I/opt/intel/composer_xe_2013_sp1.0.080/mkl/include -o cuda_bm cuda_bm.c resuse.o -L/opt/cuda/5.5/lib64 -lcublas -L/opt/intel/composer_xe_2013_sp1.0.080/mkl/lib/intel64 -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -lpthread -lrt
</pre>
=== Output ===
# Host: Single, Hex-Core Intel 5675 @ 3.0 GHz
# GPU: Kepler K20c
<pre>
$ export OMP_NUM_THREADS=4
$ ./cuda_bm
Initializing Matrices...Done.    Elapsed Time = 2.3768 secs
Starting parallel DGEMM...Done.  Elapsed Time = 33.2302 secs
                                GFlOP Rate  =  45.3546
Initializing CUDA...Done.
Re-initializing Matrices...Done.  Elapsed Time = 2.1733 secs
Starting CUDA DGEMM...Done.      Elapsed Time = 2.3522 secs
                                  GFlOP Rate  = 640.7458
</pre>
|}}
|}}


Revision as of 12:45, 11 October 2013

Description

cuda website  
CUDA™ is a parallel computing platform and programming model invented by NVIDIA. It enables dramatic increases in computing performance by harnessing the power of the graphics processing unit (GPU). With millions of CUDA-enabled GPUs sold to date, software developers, scientists and researchers are finding broad-ranging uses for GPU computing with CUDA.

Required Modules

cuda

System Variables

  • HPC_{{#uppercase:cuda}}_DIR
  • HPC_{{#uppercase:cuda}}_BIN
  • HPC_{{#uppercase:cuda}}_INC
  • HPC_{{#uppercase:cuda}}_LIB

Available GPUs

Research Computing has a significant investment in GPU-enabled servers. Each supports from two to eight Nvidia GPUs that range from the earlier S1070s (Tesla T10) to the more recent Kepler K20s (see table below).

GPU Quantity Host Quantity Host Architecture Host Memory Host Interconnect Host Attributes Notes
S1070 8 4 Intel E5462 16 GB DDR IB tesla,t10
M2070 8 4 Intel E5675 24 GB QDR IB fermi,m2070
M2070 8 1 Intel E5620 24 GB GigE fermi,m2070
M2090 10 5 Intel E5620 64 GB FDR IB fermi,m2090
M2090 16 8 AMD Opteron 6220 32 GB QDR IB fermi,m2090
K20c 2 1 Intel E5675 24 GB QDR IB k20 Reserved

Usage Policy

Interactive Use

If you need interactive access to a gpu for development and testing you may do so by requesting an interactive session through the batch system.

In order to gain interactive access to a GPU server you should run similar to the one that follows. 

qsub -I -l nodes=1:gpus=1:tesla,walltime=01:00:00 -q gpu

To gain access to one of the Fermi-class GPUs, you can make a similar request but specify the "fermi" attribute in your resource request as below. 

qsub -I -l nodes=1:gpus=1:fermi,walltime=01:00:00 -q gpu

If a gpu is available, you will get a prompt on a gpu-enabled host within a minute or two. Otherwise, you will have to wait or try another time. If you choose to wait, you will be connected when a gpu is available. The default walltime limit for the gpu queue is 10 minutes. You should request the amount of time you need but be sure to log out and end your session when you are finished so that the GPU will be available to others.

If you need two GPUs in a single host, you would run the following command instead. 

qsub -I -l nodes=1:gpus=2,walltime=01:00:00 -q gpu

If you need two gpus in two separate hosts, you would request 

qsub -I -l nodes=2:gpus=1,walltime=01:00:00 -q gpu

Batch Jobs

The process is much the same for batch jobs. To access a host with an M2090, you can add the following to your submission script. 

#PBS -q gpu
#PBS -l nodes=1:gpus=1:m2090
#PBS -l walltime=1:00:00

To access a server with an M2070 GPU, you can add the following to your submission script. 

#PBS -q gpu
#PBS -l nodes=1:gpus=1:m2070
#PBS -l walltime=1:00:00

Exclusive Mode

The GPUs are configured to run in exclusive mode. This means that the gpu driver will only allow one process at a time to access the GPU. If GPU 0 is in use and your application tries to use it, it will simply block. If your application does not call cudaSetDevice(), the CUDA runtime should assign it to a free GPU. Since everyone will be accessing the GPUs through the batch system, there should be no over-subscription of the GPUs.

Environment

For CUDA development please load the "cuda" module. Doing so will ensure that your environment is set up correctly for the use of the CUDA compiler, header files, and libraries. 

$ module spider cuda
Rebuilding cache, please wait ... (not written to file) done

    Description:
      NVIDIA CUDA Toolkit

     Versions:
        cuda/4.2
        cuda/5.5

$ module load cuda/5.5

$ which nvcc
/opt/cuda/5.5/bin/nvcc

$ printenv | grep CUDA
HPC_CUDA_LIB=/opt/cuda/5.5/lib64
HPC_CUDA_DIR=/opt/cuda/5.5
HPC_CUDA_BIN=/opt/cuda/5.5/bin
HPC_CUDA_INC=/opt/cuda/5.5/include


PBS Script Examples

See the Nvidia CUDA Toolkit_PBS page for cuda PBS script examples.

CUDA Examples

See the Nvidia CUDA Toolkit_Examples page for CUDA development examples.



Retrieved from "https://help.rc.ufl.edu/mediawiki/index.php?title=Nvidia_CUDA_Toolkit&oldid=11775"