Commit b711782a authored by D'azevedo, Ed's avatar D'azevedo, Ed
Browse files

use dmrg_is_managed()

parent 96aaf214

src/apply_Htarget_sparse.c

+26 −11
Original line number Diff line number Diff line
@@ -50,6 +50,10 @@ void apply_Htarget_sparse( 


 FpType *X_ = &(Xin_[0]);

 FpType *Y_ = &(Yout_[0]);

#ifdef USE_MAGMA

  int need_allocate_X = !dmrg_is_managed( Xin_ );

  int need_allocate_Y = !dmrg_is_managed( Yout_ );

#endif



/*

 ------------------
@@ -78,23 +82,29 @@ void apply_Htarget_sparse( 
    right_patch_size(ipatch) = R2 - R1 + 1;

    };



 long Xsize = 0;

#ifdef USE_MAGMA

 long xy_size = 0;

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

    Xsize += left_patch_size(ipatch) * right_patch_size(ipatch);

    xy_size += left_patch_size(ipatch) * right_patch_size(ipatch);

    };

 long xy_size_dim = ialign * ICEIL( xy_size, ialign );



#ifdef MAGMA

 /*

  * -----------------------

  * allocate unified memory

  * -----------------------

  */

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

  Y_ = (FpType *) dmrg_malloc( sizeof(FpType) * Xsize );

  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 );

    };



  if (need_allocate_Y) {

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

    assert( Y_ != NULL );

    };



  memcpy( &(X_[0]), &(Xin_[0]), nbytes );

  

#endif
@@ -480,9 +490,14 @@ void apply_Htarget_sparse( 
 * free unified memory

 * -------------------

 */

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

  

 if (need_allocate_X) {

    dmrg_free( X_ ); X_ = NULL;

    };

 if (need_allocate_Y) {

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

   dmrg_free( Y_ ); Y_ = NULL;

   };

#endif



}

src/dmrg_malloc.c

+16 −0
Original line number Diff line number Diff line
@@ -6,6 +6,22 @@ 
#include "cuda_runtime.h"

#endif



int dmrg_is_managed( const void *ptr )

{

  const int false = (0 == 1);

  int is_managed = false;

 

#ifdef USE_MAGMA

  struct cudaPointerAttributes attribute;

  cudaError_t ierr = cudaPointerGetAttributes( &attribute, ptr );

  assert( ierr == cudaSuccess );



  is_managed = attribute.isManaged;

#endif

  return( is_managed );

}

  



void *dmrg_malloc(const size_t alloc_size )

{

void *ptr = NULL;

src/dmrg_vbatch.h

+2 −0
Original line number Diff line number Diff line
@@ -46,6 +46,8 @@ extern 
void dmrg_init();





extern

int dmrg_is_managed( const void *ptr );



extern

void *dmrg_malloc( const size_t alloc_size );