diff --git a/src/chameleon.c b/src/chameleon.c
index c6aa22e9d5e9e7c65cbe9839c1f0d5a215d9fcad..0f3a184039269033ed1b3e4817abfb72492c4e71 100644
--- a/src/chameleon.c
+++ b/src/chameleon.c
@@ -1,5 +1,6 @@
#include "main.h"
#ifdef HAVE_CHAMELEON
+#include "chameleon.h"
cham_flttype_t chameleonType;
int (*CHAMELEON_sytrf_Tile)( cham_uplo_t, CHAM_desc_t * ) = NULL; // LLt facto
int (*CHAMELEON_sytrs_Tile)( cham_uplo_t, CHAM_desc_t *, CHAM_desc_t * ) = NULL; // LLt solve
@@ -14,6 +15,7 @@ testFEMBEM_initChameleon( ScalarType stype ) {
/* Set the scalar type and the functions used by CHAMELEON */
switch(stype) {
+#if defined(CHAMELEON_PREC_S)
case (SIMPLE_PRECISION) :
chameleonType = ChamRealFloat;
CHAMELEON_sytrf_Tile = CHAMELEON_spotrf_Tile;
@@ -22,6 +24,8 @@ testFEMBEM_initChameleon( ScalarType stype ) {
CHAMELEON_getrf_nopiv_Tile = CHAMELEON_sgetrf_nopiv_Tile;
CHAMELEON_getrs_nopiv_Tile = CHAMELEON_sgetrs_nopiv_Tile;
break ;
+#endif
+#if defined(CHAMELEON_PREC_D)
case (DOUBLE_PRECISION) :
chameleonType = ChamRealDouble;
CHAMELEON_sytrf_Tile = CHAMELEON_dpotrf_Tile;
@@ -30,6 +34,8 @@ testFEMBEM_initChameleon( ScalarType stype ) {
CHAMELEON_getrf_nopiv_Tile = CHAMELEON_dgetrf_nopiv_Tile;
CHAMELEON_getrs_nopiv_Tile = CHAMELEON_dgetrs_nopiv_Tile;
break ;
+#endif
+#if defined(CHAMELEON_PREC_C)
case (SIMPLE_COMPLEX) :
chameleonType = ChamComplexFloat;
CHAMELEON_sytrf_Tile = CHAMELEON_csytrf_Tile;
@@ -38,6 +44,8 @@ testFEMBEM_initChameleon( ScalarType stype ) {
CHAMELEON_getrf_nopiv_Tile = CHAMELEON_cgetrf_nopiv_Tile;
CHAMELEON_getrs_nopiv_Tile = CHAMELEON_cgetrs_nopiv_Tile;
break ;
+#endif
+#if defined(CHAMELEON_PREC_Z)
case (DOUBLE_COMPLEX) :
chameleonType = ChamComplexDouble;
CHAMELEON_sytrf_Tile = CHAMELEON_zsytrf_Tile;
@@ -46,6 +54,7 @@ testFEMBEM_initChameleon( ScalarType stype ) {
CHAMELEON_getrf_nopiv_Tile = CHAMELEON_zgetrf_nopiv_Tile;
CHAMELEON_getrs_nopiv_Tile = CHAMELEON_zgetrs_nopiv_Tile;
break ;
+#endif
default :
SETERRQ(1, "testHCHAMELEON : unknown scalar type\n") ;
break ;
@@ -60,30 +69,38 @@ int CHAMELEON_gemm_Tile( CHAM_desc_t *descA,
int ierr = 0;
switch(stype) {
+#if defined(CHAMELEON_PREC_S)
case (SIMPLE_PRECISION) :
ierr = CHAMELEON_sgemm_Tile( ChamNoTrans, ChamNoTrans,
(float)1., descA, descX,
(float)0., descY );
CHKERRQ(ierr);
break ;
+#endif
+#if defined(CHAMELEON_PREC_D)
case (DOUBLE_PRECISION) :
ierr = CHAMELEON_dgemm_Tile( ChamNoTrans, ChamNoTrans,
(double)1., descA, descX,
(double)0., descY);
CHKERRQ(ierr);
break ;
+#endif
+#if defined(CHAMELEON_PREC_C)
case (SIMPLE_COMPLEX) :
ierr = CHAMELEON_cgemm_Tile( ChamNoTrans, ChamNoTrans,
(CHAMELEON_Complex32_t)1., descA, descX,
(CHAMELEON_Complex32_t)0., descY);
CHKERRQ(ierr);
break ;
+#endif
+#if defined(CHAMELEON_PREC_Z)
case (DOUBLE_COMPLEX) :
ierr = CHAMELEON_zgemm_Tile( ChamNoTrans, ChamNoTrans,
(CHAMELEON_Complex64_t)1., descA, descX,
(CHAMELEON_Complex64_t)0., descY);
CHKERRQ(ierr);
break ;
+#endif
default :
SETERRQ(1, "CHAMELEON_gemm_Tile: unknown scalar type\n") ;
break ;
@@ -93,9 +110,9 @@ int CHAMELEON_gemm_Tile( CHAM_desc_t *descA,
// indices are 0 based, bounds included
static int
-CHAMELEON_build_callback_FEMBEM( const CHAM_desc_t *desc,
- cham_uplo_t uplo, int m, int n,
- CHAM_tile_t *tile, void *op_args )
+CHAMELEON_build_callback_FEMBEM( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *desc, CHAM_tile_t *tile, ... )
{
ASSERTA(tile);
@@ -110,12 +127,15 @@ CHAMELEON_build_callback_FEMBEM( const CHAM_desc_t *desc,
&Mpf_i_zero, &Mpf_i_one, &Mpf_i_one, &Mpf_i_one,
&size_of_buffer, &(tile->ld), tile->mat, op_args );
+ (void)ndata;
return 0;
}
int CHAMELEON_generate_matrix( cham_flttype_t flttype, int NB, int PQ[2],
CHAM_desc_t **descA )
{
+ cham_map_data_t data;
+ cham_map_operator_t op;
int ierr;
int N = nbPts;
cham_uplo_t uplo = symMatSolver ? ChamLower : ChamUpperLower;
@@ -133,8 +153,15 @@ CHAM_desc_t **descA )
CHKERRQ(ierr);
// Assembly driven by chameleon and using the callback CHAMELEON_build_callback_FEMBEM() defined above
- ierr = CHAMELEON_map_Tile( ChamW, uplo, *descA,
- CHAMELEON_build_callback_FEMBEM, myCtx );
+ data.access = ChamW;
+ data.desc = *descA;
+
+ op.name = "BuildFEMBEM";
+ op.cpufunc = CHAMELEON_build_callback_FEMBEM;
+ op.cudafunc = NULL;
+ op.hipfunc = NULL;
+
+ ierr = CHAMELEON_mapv_Tile( uplo, 1, &data, &op, myCtx );
CHKERRQ(ierr);
MpfFree( myCtx );
diff --git a/src/hchameleon.c b/src/hchameleon.c
index 04c6c834d92c166169b5427c5c8e4e301182cf98..0fbc09d1716d0c5c4f9fd1eef15acdb257eabd1c 100644
--- a/src/hchameleon.c
+++ b/src/hchameleon.c
@@ -30,9 +30,9 @@ inline static int chameleon_desc_islocal( const CHAM_desc_t *A, int m, int n )
// indices are 0 based, bounds included
/* Initialize a given tile with an h-matrix, given the row and cluster tree */
static int
-HCHAMELEON_build_callback_FEMBEM( const CHAM_desc_t *desc,
- cham_uplo_t uplo, int m, int n,
- CHAM_tile_t *tile, void *op_args )
+HCHAMELEON_build_callback_FEMBEM( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *desc, CHAM_tile_t *tile, ... )
{
contextTestFEMBEM ctx_fembem;
struct mpf_hmat_create_compression_args_t compression_ctx;
@@ -193,8 +193,19 @@ hmat_interface_t *hi )
*/
#if 1
cham_uplo_t uplo = symMatSolver ? ChamLower : ChamUpperLower;
- ierr = CHAMELEON_map_Tile( ChamW, uplo, descA,
- HCHAMELEON_build_callback_FEMBEM, hdescA );
+ cham_map_data_t data;
+ cham_map_operator_t op;
+
+ data.access = ChamW;
+ data.desc = descA;
+
+ op.name = "BuildFEMBEM-H";
+ op.cpufunc = HCHAMELEON_build_callback_FEMBEM;
+ op.cudafunc = NULL;
+ op.hipfunc = NULL;
+
+ ierr = CHAMELEON_mapv_Tile( uplo, 1, &data, &op, hdescA );
+ CHKERRQ(ierr);
#else
tile = descA->tiles;
for(j=0; j < descA->lnt; j++) {
@@ -253,7 +264,7 @@ HCHAMELEON_destroy_matrix( HCHAM_desc_t *hdescA )
}
/* Provides a flat matrix (array of values) given an hmatrix structure */
-typedef int (*core_lacpy_fct_t)( cham_uplo_t, int, int, const void *, int, void *, int );
+typedef void (*core_lacpy_fct_t)( cham_uplo_t, int, int, const void *, int, void *, int );
static void
TCORE_uncompress( HCHAM_desc_t *hdescA,
@@ -301,23 +312,31 @@ TCORE_uncompress( HCHAM_desc_t *hdescA,
free(col_ptr);
}
else {
- int (*CORE_lacpy)( cham_uplo_t, int, int,
+ void (*CORE_lacpy)( cham_uplo_t, int, int,
const void *, int, void *, int ) = NULL;
switch( hdescA->super->dtyp ) {
+#if defined(CHAMELEON_PREC_S)
case ChamRealFloat:
CORE_lacpy = (core_lacpy_fct_t)CORE_slacpy;
break ;
+#endif
+#if defined(CHAMELEON_PREC_D)
case ChamRealDouble:
CORE_lacpy = (core_lacpy_fct_t)CORE_dlacpy;
break ;
+#endif
+#if defined(CHAMELEON_PREC_C)
case ChamComplexFloat:
CORE_lacpy = (core_lacpy_fct_t)CORE_clacpy;
break ;
+#endif
+#if defined(CHAMELEON_PREC_Z)
case ChamComplexDouble:
CORE_lacpy = (core_lacpy_fct_t)CORE_zlacpy;
break ;
+#endif
default :
- //SETERRQ(1, "HCHAMELEON_uncompress_matrix: unknown scalar type\n") ;
+ SETERRQ(1, "HCHAMELEON_uncompress_matrix: unknown scalar type\n") ;
break ;
}
CORE_lacpy( ChamUpperLower, tileA->m, tileA->n,
@@ -377,9 +396,9 @@ typedef struct getinfo_args_s {
} getinfo_args_t;
static int
-HCHAMELEON_map_getinfo( const CHAM_desc_t *desc,
- cham_uplo_t uplo, int m, int n,
- CHAM_tile_t *tile, void *op_args )
+HCHAMELEON_map_getinfo( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *desc, CHAM_tile_t *tile, ... )
{
getinfo_args_t *getinfo = (getinfo_args_t*)op_args;
hmat_info_t info;
@@ -390,6 +409,10 @@ HCHAMELEON_map_getinfo( const CHAM_desc_t *desc,
getinfo->uncompress_size += info.uncompressed_size;
pthread_mutex_unlock( &(getinfo->mutex) );
+ (void)ndata;
+ (void)uplo;
+ (void)m;
+ (void)n;
return 0;
}
@@ -404,9 +427,18 @@ hmat_info_t HCHAMELEON_getinfo( HCHAM_desc_t *hdesc )
.mutex = PTHREAD_MUTEX_INITIALIZER,
};
cham_uplo_t uplo = symMatSolver ? ChamLower : ChamUpperLower;
+ cham_map_data_t data;
+ cham_map_operator_t op;
+
+ data.access = ChamR;
+ data.desc = hdesc->super;
+
+ op.name = "HCHAM_getInfo";
+ op.cpufunc = HCHAMELEON_map_getinfo;
+ op.cudafunc = NULL;
+ op.hipfunc = NULL;
- CHAMELEON_map_Tile( ChamR, uplo, hdesc->super,
- HCHAMELEON_map_getinfo, &op_args );
+ CHAMELEON_mapv_Tile( uplo, 1, &data, &op, &op_args );
MPI_Allreduce( &(op_args.compress_size), &ginfo.compressed_size, 1,
MPI_UNSIGNED_LONG, MPI_SUM, MPI_COMM_WORLD );