Commit 4852f828 authored by D'azevedo, Ed's avatar D'azevedo, Ed
Browse files

add estimate_work()

parent c62b2870

src/apply_Htarget_sparse.c

+8 −2
Original line number Diff line number Diff line
@@ -103,7 +103,10 @@ void apply_Htarget_sparse( 
  if (need_allocate_X) {

    X_ = (FpType *) dmrg_malloc( sizeof(FpType) * xy_size_dim );

    assert( X_ != NULL );

    memcpy( &(X_[0]), &(Xin_[0]), sizeof(FpType) * xy_size );

    void *dest = (void *) &(X_[0]);

    void *src = (void *) &(Xin_[0]);

    size_t count = sizeof(FpType) * xy_size;

    dmrg_memcpy( dest, src, count );

    };



  if (need_allocate_Y) {
@@ -501,7 +504,10 @@ void apply_Htarget_sparse( 
    dmrg_free( X_ ); X_ = NULL;

    };

 if (need_allocate_Y) {

   memcpy( &(Yout_[0]), &(Y_[0]), sizeof(FpType) * xy_size );

   void *dest = &(Yout_[0]);

   void *src = &(Y_[0]);

   size_t count = sizeof(FpType) * xy_size;

   dmrg_memcpy( dest, src, count );

   dmrg_free( Y_ ); Y_ = NULL;

   };

#endif

src/dmrg_malloc.c

+12 −0
Original line number Diff line number Diff line
@@ -20,6 +20,18 @@ int dmrg_is_managed( const void *ptr ) 
  return( is_managed );

}

  

void dmrg_memcpy(void *dest, const void *src, size_t count)

{

#ifdef USE_MAGMA

  cudaError_t istat = cudaMemcpy( dest, src, count, cudaMemcpyDefault );

  if (istat != cudaSuccess) {

    fprintf(stderr,"dmrg_memcpy: %s\n", cudaGetErrorString(istat));

    };

  assert( istat == cudaSuccess );

#else

  memcpy( dest, src, count );

#endif

}



void *dmrg_malloc(const size_t alloc_size )

{

src/dmrg_vbatch.h

+2 −0
Original line number Diff line number Diff line
@@ -45,6 +45,8 @@ extern "C" { 
extern

void dmrg_init();



extern

void dmrg_memcpy(void *dest, const void *src, size_t n);



extern

int dmrg_is_managed( const void *ptr );

src/test_vbatch.c

+38 −44
Original line number Diff line number Diff line
@@ -10,7 +10,6 @@ 
#include <omp.h>

#endif



#include "analysis.h"
@@ -150,41 +149,6 @@ int main(int argc, char *argv[]) 
     xy_size += (nrowX * ncolX );

     };

 }

/*

 -------------------

 estimate total work

 -------------------

 */

 double total_flops_method1 = 0.0;

 double total_flops_method2 = 0.0;

 double total_flops = 0.0;

 {

 int ipatch = 0;

 int jpatch = 0;



 for(jpatch=1; jpatch <= npatches; jpatch++) {

 for(ipatch=1; ipatch <= npatches; ipatch++) {

    double flops_total = 0.0;

    double flops_method1 = 0.0;

    double flops_method2 = 0.0;

    int nrowA = left_patch_size(ipatch);

    int ncolA = left_patch_size(jpatch);

    int nrowB = right_patch_size(ipatch);

    int ncolB = right_patch_size(jpatch);



    cal_kron_flops( nrowA, nrowB, ncolA, ncolB, 

            &flops_total, &flops_method1,   &flops_method2);



    total_flops += flops_total;

    total_flops_method1 += flops_method1;

    total_flops_method2 += flops_method2;

    };

    };

  };

  

 total_flops *= noperator;

 total_flops_method1 *= noperator;

 total_flops_method2 *= noperator;
@@ -194,15 +158,7 @@ int main(int argc, char *argv[]) 
         keep_left_states,    keep_right_states, noperator, (long int) xy_size );



 printf("npatches=%d\n", npatches );

 {

 double total_gflops = total_flops/(1000.0*1000.0*1000.0);

 double total_gflops_method1 = total_flops_method1 / (1000.0*1000.0*1000.0);

 double total_gflops_method2 = total_flops_method2 / (1000.0*1000.0*1000.0);



 printf("total_gflops=%lf, total_gflops_method1=%lf, total_gflops_method2=%lf\n",

         total_gflops,     total_gflops_method1,     total_gflops_method2 );



 };





 if (idebug >= 1) {
@@ -271,6 +227,14 @@ int main(int argc, char *argv[]) 


   }

 else {

    nC_ = (int *) malloc( sizeof(int) * npatches * npatches );

    assert( nC_ != NULL );



    int i = 0;

    for(i=0; i < npatches*npatches; i++) {

      nC_[i] = noperator;

      };

    

 

    setup_vbatch( noperator, npatches, 

                  &(left_patch_size_[0]),
@@ -284,6 +248,36 @@ int main(int argc, char *argv[]) 
                  Amatrix, ld_Amatrix,  Bmatrix, ld_Bmatrix );

    };





 {

 double total_gflops = 0;

 double gmemA = 0;

 double gmemB = 0;

 double gmemBX = 0;

 

 estimate_work( npatches, 

                left_patch_size_, 

                right_patch_size_,

                nC_, 

                &total_gflops, 

                &gmemA, 

                &gmemB, 

                &gmemBX 

              );



 double giga = 1000.0*1000.0*1000.0;

 double gmemA_gbytes = gmemA * sizeof(double)/giga;

 double gmemB_gbytes = gmemB * sizeof(double)/giga;

 double gmemBX_gbytes = gmemBX * sizeof(double)/giga;



 printf("total_gflops=%lf \n", total_gflops );

 printf("memory for Amat=%lf GBytes\n", gmemA_gbytes );

 printf("memory for Bmat=%lf GBytes\n", gmemB_gbytes );

 printf("memory for BXmat=%lf GBytes\n", gmemBX_gbytes );



 }





#define Abatch(i,j)  Abatch_[indx2f(i,j,ld_Abatch)]

#define Bbatch(i,j)  Bbatch_[indx2f(i,j,ld_Bbatch)]

src/test_vbatch.h

+2 −0
Original line number Diff line number Diff line
@@ -5,6 +5,7 @@ 
#include <math.h>

#include <assert.h>



#include "analysis.h"

#include "dmrg_types.h"

#include "dmrg_vbatch.h"
@@ -12,5 +13,6 @@ 
#include "setup_sparse_batch.h"

#include "setup_matrix.h"



#include "estimate_work.h"



#endif