diff --git a/CMakeLists.txt b/CMakeLists.txt
index 5ecad49b276fd783d159dfc0a022c8a7f6f47ad6..de8df69df660c388518ce08b652504db6f4d024f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -101,9 +101,6 @@ set( RP_CHAMELEON_DICTIONNARY ${CMAKE_SOURCE_DIR}/cmake_modules/local_subs.py )
set( RP_CHAMELEON_PRECISIONS "s;d;c;z" )
include(RulesPrecisions)
-option(CHAMELEON_USE_MIGRATE
- "This options enables the data migration in QR algorithms" ON)
-
# Options to select the runtime
# -----------------------------
@@ -175,6 +172,14 @@ if (CHAMELEON_ENABLE_MPI AND NOT CHAMELEON_USE_MPI)
message("-- ${BoldGreen}CHAMELEON_USE_MPI is set to OFF, turn it ON to use MPI (unsupported by Quark)${ColourReset}")
endif()
+cmake_dependent_option(CHAMELEON_USE_MIGRATE
+ "This options enables the data migration in QR algorithms" ON
+ "CHAMELEON_USE_MPI" OFF)
+
+cmake_dependent_option(CHAMELEON_USE_MPI_DATATYPES
+ "Enable MPI datatypes wheneve supported by the runtime" ON
+ "CHAMELEON_USE_MPI" OFF)
+
# Enable Cuda kernels if StarPU (only if StarPU or PaRSEC runtime is enabled)
# TODO: Default should be changed to ON/OFF when it will be ok
cmake_dependent_option(CHAMELEON_USE_CUDA
@@ -658,7 +663,11 @@ if( CHAMELEON_SCHED_STARPU )
endif()
check_function_exists(starpu_mpi_interface_datatype_register HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER)
if ( HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER )
- message("-- ${Blue}Add definition HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER${ColourReset}")
+ message("-- ${Blue}Add definition HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER${ColourReset}")
+ else()
+ if( CHAMELEON_USE_MPI_DATATYPES )
+ set( CHAMELEON_USE_MPI_DATATYPES OFF )
+ endif()
endif()
check_function_exists(starpu_interface_copy2d HAVE_STARPU_INTERFACE_COPY2D)
if ( HAVE_STARPU_INTERFACE_COPY2D )
diff --git a/include/chameleon/config.h.in b/include/chameleon/config.h.in
index 688df8866668276dbda279cc97b070c0ae130454..10eb9136608e47c242bb7e4ad553b2b7531ca7e5 100644
--- a/include/chameleon/config.h.in
+++ b/include/chameleon/config.h.in
@@ -38,6 +38,10 @@
#if !defined(CHAMELEON_USE_MPI) && defined(CHAMELEON_USE_MIGRATE)
#undef CHAMELEON_USE_MIGRATE
#endif
+#cmakedefine CHAMELEON_USE_MPI_DATATYPES
+#if !defined(CHAMELEON_USE_MPI) && defined(CHAMELEON_USE_MPI_DATATYPES)
+#undef CHAMELEON_USE_MPI_DATATYPES
+#endif
/* GPU Support */
#cmakedefine CHAMELEON_USE_CUDA
diff --git a/runtime/starpu/include/cham_tile_interface.h b/runtime/starpu/include/cham_tile_interface.h
index 1a7f887aae261c7c23f043f163ffbe16c7557239..0e04100a515861b4b68988a81755a3bcb21c6324 100644
--- a/runtime/starpu/include/cham_tile_interface.h
+++ b/runtime/starpu/include/cham_tile_interface.h
@@ -34,7 +34,7 @@ struct starpu_cham_tile_interface_s
cham_flttype_t flttype; /**< Type of the elements of the matrix */
size_t allocsize; /**< size actually currently allocated */
size_t tilesize; /**< size of the elements of the matrix */
- CHAM_tile_t tile; /**< Internal tile structure used to store
+ CHAM_tile_t tile; /**< Internal tile structure used to store
information on non memory home_node */
};
diff --git a/runtime/starpu/include/chameleon_starpu.h.in b/runtime/starpu/include/chameleon_starpu.h.in
index f5a4fc1e6819f07549341639eb0c08324d01a799..9b8dc1372ee5bfdaeb0022d0de0dbb75e5460525 100644
--- a/runtime/starpu/include/chameleon_starpu.h.in
+++ b/runtime/starpu/include/chameleon_starpu.h.in
@@ -40,6 +40,10 @@
#cmakedefine HAVE_STARPU_MPI_WAIT_FOR_ALL
#cmakedefine HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER
+#if !defined(HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER) && defined(CHAMELEON_USE_MPI_DATATYPES)
+#error "This version of StarPU does not support MPI datatypes (Please compile with -DCHAMELEON_USE_MPI_DATATYPES=OFF)"
+#endif
+
#if defined(CHAMELEON_USE_MPI)
#include <starpu_mpi.h>
#else
diff --git a/runtime/starpu/interface/cham_tile_interface.c b/runtime/starpu/interface/cham_tile_interface.c
index eaab054d4c17844f19152095558cb4ad51dcc7d6..ea335c6404d8d52ec38a477c73c11c7909bc1c90 100644
--- a/runtime/starpu/interface/cham_tile_interface.c
+++ b/runtime/starpu/interface/cham_tile_interface.c
@@ -319,30 +319,47 @@ cti_unpack_data( starpu_data_handle_t handle, unsigned node, void *ptr, size_t c
starpu_cham_tile_interface_t *cham_tile_interface = (starpu_cham_tile_interface_t *)
starpu_data_get_interface_on_node(handle, node);
- CHAM_tile_t dsttile;
char *tmp = ptr;
- /* Extract the size of the information t unpack */
- memcpy( &(cham_tile_interface->allocsize), tmp, sizeof(size_t) );
- tmp += sizeof(size_t);
+#if defined(CHAMELEON_USE_MPI_DATATYPES)
+ /*
+ * We may end up here if an early reception occured before the handle of the
+ * received data has been registered. Thus, datatype was not existant and we
+ * need to unpack the data ourselves
+ */
+ STARPU_ASSERT( count == cham_tile_interface->allocsize );
+ STARPU_ASSERT( cham_tile_interface->tile.format & CHAMELEON_TILE_FULLRANK );
+#else
+ {
+ CHAM_tile_t dsttile;
- /* Extract the tile metadata of the remote tile */
- memcpy( &dsttile, tmp, sizeof(CHAM_tile_t) );
- tmp += sizeof(CHAM_tile_t);
+ /* Extract the size of the information to unpack */
+ memcpy( &(cham_tile_interface->allocsize), tmp, sizeof(size_t) );
+ tmp += sizeof(size_t);
- cham_tile_interface->tile.format = dsttile.format;
- cham_tile_interface->tile.ld = cham_tile_interface->tile.m;
- STARPU_ASSERT( cham_tile_interface->tile.m == dsttile.m );
- STARPU_ASSERT( cham_tile_interface->tile.n == dsttile.n );
- STARPU_ASSERT( count == cham_tile_interface->allocsize + sizeof(size_t) + sizeof(CHAM_tile_t) );
+ /* Extract the tile metadata of the remote tile */
+ memcpy( &dsttile, tmp, sizeof(CHAM_tile_t) );
+ tmp += sizeof(CHAM_tile_t);
+ /*
+ * Update with the local information. Data is packed now, and do not
+ * need leading dimension anymore
+ */
+ cham_tile_interface->tile.format = dsttile.format;
+ cham_tile_interface->tile.ld = cham_tile_interface->tile.m;
+
+ STARPU_ASSERT( cham_tile_interface->tile.m == dsttile.m );
+ STARPU_ASSERT( cham_tile_interface->tile.n == dsttile.n );
+ STARPU_ASSERT( count == cham_tile_interface->allocsize + sizeof(size_t) + sizeof(CHAM_tile_t) );
+ }
+#endif
/* Unpack the real data */
if ( cham_tile_interface->tile.format & CHAMELEON_TILE_FULLRANK ) {
cti_unpack_data_fullrank( cham_tile_interface, tmp );
}
else {
- STARPU_ASSERT_MSG( 1, "Unsupported format for pack." );
+ STARPU_ASSERT_MSG( 1, "Unsupported format for unpack." );
}
/* Free the received information */
@@ -481,7 +498,7 @@ cti_handle_get_allocsize( starpu_data_handle_t handle )
return cham_tile_interface->allocsize;
}
-#if defined(HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER)
+#if defined(CHAMELEON_USE_MPI_DATATYPES)
int
cti_allocate_datatype( starpu_data_handle_t handle,
MPI_Datatype *datatype )
@@ -518,7 +535,7 @@ starpu_cham_tile_interface_init()
if ( starpu_interface_cham_tile_ops.interfaceid == STARPU_UNKNOWN_INTERFACE_ID )
{
starpu_interface_cham_tile_ops.interfaceid = starpu_data_interface_get_next_id();
-#if defined(HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER)
+#if defined(CHAMELEON_USE_MPI_DATATYPES)
starpu_mpi_interface_datatype_register( starpu_interface_cham_tile_ops.interfaceid,
cti_allocate_datatype,
cti_free_datatype );
@@ -531,7 +548,7 @@ starpu_cham_tile_interface_fini()
{
if ( starpu_interface_cham_tile_ops.interfaceid != STARPU_UNKNOWN_INTERFACE_ID )
{
-#if defined(HAVE_STARPU_MPI_INTERFACE_DATATYPE_REGISTER)
+#if defined(CHAMELEON_USE_MPI_DATATYPES)
starpu_mpi_interface_datatype_unregister( starpu_interface_cham_tile_ops.interfaceid );
#endif
}