Update PaRSEC

src/cuda/lapack_cuda_stage_in.c

@@ -33,14 +33,14 @@ dplasma_cuda_lapack_stage_in(parsec_gpu_task_t *gtask,
         if(flow_mask & (1U << i)){
             copy_in = task->data[i].data_in;
             copy_out = task->data[i].data_out;
-            ddc = (dplasma_data_collection_t*)gtask->flow_dc[i];
+            ddc = (dplasma_data_collection_t*)gtask->flow_info[i].flow_dc;
             assert(ddc != NULL);
             elem_sz = parsec_datadist_getsizeoftype(ddc->dc_original->mtype);
             in_elem_dev = (parsec_device_gpu_module_t*)parsec_mca_device_get( copy_in->device_index);
             if( (in_elem_dev->super.type == PARSEC_DEV_CUDA) || (ddc->dc_original->storage != PARSEC_MATRIX_LAPACK)){
                 ret = (cudaError_t)cudaMemcpyAsync( copy_out->device_private,
                                                     copy_in->device_private,
-                                                    gtask->flow_nb_elts[i],
+                                                    gtask->flow_info[i].flow_span,
                                                     (in_elem_dev->super.type != PARSEC_DEV_CUDA)?
                                                             cudaMemcpyHostToDevice : cudaMemcpyDeviceToDevice,
                                                     cuda_stream->cuda_stream);
@@ -109,15 +109,15 @@ dplasma_cuda_lapack_stage_out(parsec_gpu_task_t *gtask,
         if(flow_mask & (1U << i)){
             copy_in = task->data[i].data_out;
             copy_out = copy_in->original->device_copies[0];
-            ddc = (dplasma_data_collection_t*)gtask->flow_dc[i];
+            ddc = (dplasma_data_collection_t*)gtask->flow_info[i].flow_dc;
             assert(ddc != NULL);
             elem_sz = parsec_datadist_getsizeoftype(ddc->dc_original->mtype);
             out_elem_dev = (parsec_device_gpu_module_t*)parsec_mca_device_get( copy_out->device_index);
 
             if( (out_elem_dev->super.type == PARSEC_DEV_CUDA) || (ddc->dc_original->storage != PARSEC_MATRIX_LAPACK)){
                 ret = (cudaError_t)cudaMemcpyAsync( copy_out->device_private,
                                                     copy_in->device_private,
-                                                    gtask->flow_nb_elts[i],
+                                                    gtask->flow_info[i].flow_span,
                                                     out_elem_dev->super.type != PARSEC_DEV_CUDA ?
                                                             cudaMemcpyDeviceToHost : cudaMemcpyDeviceToDevice,
                                                     cuda_stream->cuda_stream);

src/dplasmaaux.c

@@ -98,7 +98,7 @@ dplasma_aux_getGEMMLookahead( parsec_tiled_matrix_t *A )
      * look ahead based on the global information to get the same one on all
      * nodes.
      */
-    int nbunits = vpmap_get_nb_total_threads() * A->super.nodes;
+    int nbunits = parsec_vpmap_get_nb_total_threads() * A->super.nodes;
     double alpha =  3. * (double)nbunits / ( A->mt * A->nt );
 
     if ( A->super.nodes == 1 ) {

src/utils/dplasma_lapack_adtt.c

@@ -288,7 +288,7 @@ static parsec_data_t* data_of(parsec_data_collection_t *desc, ...)
                 "data_of CP %p [old type %p] loc %d -> dtt %p target_shape %d layout %d",
                 cp, cp->dtt, loc, adt->opaque_dtt, info.shape, info.layout);
             dt = parsec_data_create_with_type( dt->dc,
-                                               dt->key, cp->device_private, dt->nb_elts,
+                                               dt->key, cp->device_private, dt->span,
                                                adt->opaque_dtt);
         }
     }

src/zgebrd_ge2gb_wrapper.c

@@ -310,7 +310,7 @@ dplasma_zgebrd_ge2gb_New( int ib,
     dplasma_qrtree_t *qrtre0, *qrtree, *lqtree;
     int P, Q, cores;
 
-    cores = dplasma_imax( vpmap_get_nb_total_threads(), 1 );
+    cores = dplasma_imax( parsec_vpmap_get_nb_total_threads(), 1 );
     qrtree = malloc( sizeof(dplasma_qrtree_t) );
     lqtree = malloc( sizeof(dplasma_qrtree_t) );
 

src/zgetrf_1d.jdf

@@ -57,7 +57,7 @@ descIPIV    [type = "parsec_tiled_matrix_t*" hidden = on default = "((dplasma_da
 
 INFO      [type = "int*"]
 
-nbmaxthrd [type = "int" hidden=on default="( dplasma_imax( 1, dplasma_imin( vpmap_get_nb_threads_in_vp(0) - 1, 48 ) ) )" ] /* 48 is the actual limit of the kernel */
+nbmaxthrd [type = "int" hidden=on default="( dplasma_imax( 1, dplasma_imin( parsec_vpmap_get_vp_threads(0) - 1, 48 ) ) )" ] /* 48 is the actual limit of the kernel */
 KT        [type = "int" hidden=on default="( dplasma_imin( descA->mt, descA->nt )-1 )" ]
 
 getrfdata [type = "CORE_zgetrf_data_t *" hidden = on default = "NULL"]

src/zgetrf_1d_wrapper.c

@@ -83,7 +83,7 @@ dplasma_zgetrf_1d_New( parsec_tiled_matrix_t *A,
                        int *INFO )
 {
     parsec_zgetrf_1d_taskpool_t *parsec_getrf_1d;
-    int nbthreads = dplasma_imax( 1, vpmap_get_nb_threads_in_vp(0) - 1 );
+    int nbthreads = dplasma_imax( 1, parsec_vpmap_get_vp_threads(0) - 1 );
     dplasma_data_collection_t * ddc_A = dplasma_wrap_data_collection((parsec_tiled_matrix_t*)A);
     dplasma_data_collection_t * ddc_IPIV = dplasma_wrap_data_collection((parsec_tiled_matrix_t*)IPIV);
 

src/zgetrf_qrf.jdf

@@ -52,7 +52,7 @@ INFO     [type = "int*"]
 param_p   [type = int default="((parsec_matrix_block_cyclic_t*)descA)->grid.rows" hidden=on ]
 param_q   [type = int default="((parsec_matrix_block_cyclic_t*)descA)->grid.cols" hidden=on ]
 minMNT    [type = int default="dplasma_imin( descA->mt-1, descA->nt-1 )"    hidden=on ]
-nbmaxthrd [type = "int" default="( dplasma_imax( 1, dplasma_imin( vpmap_get_nb_threads_in_vp(0) - 1, 48 ) ) )" hidden=on] /* 48 is the actual limit of the kernel */
+nbmaxthrd [type = "int" default="( dplasma_imax( 1, dplasma_imin( parsec_vpmap_get_vp_threads(0) - 1, 48 ) ) )" hidden=on] /* 48 is the actual limit of the kernel */
 
 getrfdata [type = "CORE_zgetrf_data_t *" hidden = on default = "NULL"]
 

src/zgetrf_qrf_wrapper.c

@@ -166,7 +166,7 @@ dplasma_zgetrf_qrf_New( dplasma_qrtree_t *qrtree,
     int ib = TS->mb;
     size_t sizeW = 1;
     size_t sizeReduceVec = 1;
-    int nbthreads = dplasma_imax( 1, vpmap_get_nb_threads_in_vp(0) - 1 );
+    int nbthreads = dplasma_imax( 1, parsec_vpmap_get_vp_threads(0) - 1 );
 
     /*
      * Compute W size according to criteria used.

tests/common.h

@@ -22,6 +22,7 @@
 #include "parsec/profiling.h"
 #include "parsec/parsec_internal.h"
 #include "parsec/utils/debug.h"
+#include "parsec/interfaces/dtd/insert_function.h"
 #include "dplasma.h"
 #include "dplasma/types.h"
 
@@ -31,6 +32,32 @@
 #include "common_timing.h"
 #include "flops.h"
 
+/* PaRSEC 4.1 removed parsec_dtd_create/destroy_arena_datatype helpers. */
+static inline parsec_arena_datatype_t *
+dplasma_dtd_create_arena_datatype_compat(parsec_context_t *ctx, int *id)
+{
+    parsec_arena_datatype_t *adt = parsec_arena_datatype_new();
+    if( NULL == adt ) {
+        return NULL;
+    }
+    if( PARSEC_SUCCESS != parsec_dtd_attach_arena_datatype(ctx, adt, id) ) {
+        parsec_arena_datatype_release(&adt);
+        return NULL;
+    }
+    return adt;
+}
+
+static inline int
+dplasma_dtd_destroy_arena_datatype_compat(parsec_context_t *ctx, int id)
+{
+    return parsec_dtd_free_arena_datatype(ctx, id);
+}
+
+#define parsec_dtd_create_arena_datatype(ctx, id) \
+    dplasma_dtd_create_arena_datatype_compat((ctx), (id))
+#define parsec_dtd_destroy_arena_datatype(ctx, id) \
+    dplasma_dtd_destroy_arena_datatype_compat((ctx), (id))
+
 /* these are globals in common.c */
 extern char *PARSEC_SCHED_NAME[];
 extern int unix_timestamp;