/*@
MPI_Type_contiguous - Creates a contiguous datatype
Input Parameters:
+ count - replication count (nonnegative integer)
- oldtype - old datatype (handle)
Output Parameter:
. newtype - new datatype (handle)
.N ThreadSafe
.N Fortran
.N Errors
.N MPI_SUCCESS
.N MPI_ERR_TYPE
.N MPI_ERR_COUNT
.N MPI_ERR_EXHAUSTED
@*/
int MPI_Type_contiguous(int count,
MPI_Datatype old_type,
MPI_Datatype *new_type_p)
{
int mpi_errno = MPI_SUCCESS;
MPID_MPI_STATE_DECL(MPID_STATE_MPI_TYPE_CONTIGUOUS);
MPIR_ERRTEST_INITIALIZED_ORDIE();
MPIU_THREAD_CS_ENTER(ALLFUNC,);
MPID_MPI_FUNC_ENTER(MPID_STATE_MPI_TYPE_CONTIGUOUS);
# ifdef HAVE_ERROR_CHECKING
{
MPID_BEGIN_ERROR_CHECKS;
{
MPID_Datatype *datatype_ptr = NULL;
/* MPIR_ERRTEST_XXX macros defined in mpiimpl.h */
MPIR_ERRTEST_COUNT(count, mpi_errno);
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
MPIR_ERRTEST_DATATYPE(old_type, "datatype", mpi_errno);
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
if (HANDLE_GET_KIND(old_type) != HANDLE_KIND_BUILTIN) {
MPID_Datatype_get_ptr(old_type, datatype_ptr);
MPID_Datatype_valid_ptr(datatype_ptr, mpi_errno);
}
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
}
MPID_END_ERROR_CHECKS;
}
# endif /* HAVE_ERROR_CHECKING */
/* ... body of routine ... */
mpi_errno = MPIR_Type_contiguous_impl(count, old_type, new_type_p);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
/* ... end of body of routine ... */
fn_exit:
MPID_MPI_FUNC_EXIT(MPID_STATE_MPI_TYPE_CONTIGUOUS);
MPIU_THREAD_CS_EXIT(ALLFUNC,);
return mpi_errno;
fn_fail:
/* --BEGIN ERROR HANDLING-- */
# ifdef HAVE_ERROR_CHECKING
{
mpi_errno = MPIR_Err_create_code(
mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**mpi_type_contiguous",
"**mpi_type_contiguous %d %D %p", count, old_type, new_type_p);
}
# endif
mpi_errno = MPIR_Err_return_comm( NULL, FCNAME, mpi_errno );
goto fn_exit;
/* --END ERROR HANDLING-- */
}
int MPIR_Type_contiguous_x_impl(MPI_Count count,
MPI_Datatype oldtype,
MPI_Datatype *newtype)
{
/* to make 'count' fit MPI-3 type processing routines (which take integer
* counts), we construct a type consisting of N INT_MAX chunks followed by
* a remainder. e.g for a count of 4000000000 bytes you would end up with
* one 2147483647-byte chunk followed immediately by a 1852516353-byte
* chunk */
MPI_Datatype chunks, remainder;
MPI_Aint lb, extent, disps[2];
int blocklens[2];
MPI_Datatype types[2];
int mpi_errno;
/* truly stupendously large counts will overflow an integer with this math,
* but that is a problem for a few decades from now. Sorry, few decades
* from now! */
MPIR_Assert(count/INT_MAX == (int)(count/INT_MAX));
int c = (int)(count/INT_MAX); /* OK to cast until 'count' is 256 bits */
int r = count%INT_MAX;
mpi_errno = MPIR_Type_vector_impl(c, INT_MAX, INT_MAX, oldtype, &chunks);
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
mpi_errno = MPIR_Type_contiguous_impl(r, oldtype, &remainder);
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
MPIR_Type_get_extent_impl(oldtype, &lb, &extent);
blocklens[0] = 1; blocklens[1] = 1;
disps[0] = 0; disps[1] = c*extent*INT_MAX;
types[0] = chunks; types[1] = remainder;
mpi_errno = MPIR_Type_create_struct_impl(2, blocklens, disps, types, newtype);
MPIR_Type_free_impl(&chunks);
MPIR_Type_free_impl(&remainder);
fn_exit:
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPII_Dataloop_convert_subarray(int ndims,
int *array_of_sizes,
int *array_of_subsizes,
int *array_of_starts,
int order, MPI_Datatype oldtype, MPI_Datatype * newtype)
{
int mpi_errno = MPI_SUCCESS;
MPI_Aint extent, disps[3], size;
int i, blklens[3];
MPI_Datatype tmp1, tmp2, types[3];
MPIR_Datatype_get_extent_macro(oldtype, extent);
if (order == MPI_ORDER_FORTRAN) {
/* dimension 0 changes fastest */
if (ndims == 1) {
mpi_errno = MPIR_Type_contiguous_impl(array_of_subsizes[0], oldtype, &tmp1);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
} else {
mpi_errno = MPIR_Type_vector_impl(array_of_subsizes[1],
array_of_subsizes[0],
array_of_sizes[0], oldtype, &tmp1);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
size = (MPI_Aint) (array_of_sizes[0]) * extent;
for (i = 2; i < ndims; i++) {
size *= (MPI_Aint) (array_of_sizes[i - 1]);
mpi_errno = MPIR_Type_hvector_impl(array_of_subsizes[i], 1, size, tmp1, &tmp2);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
MPIR_Type_free_impl(&tmp1);
tmp1 = tmp2;
}
}
/* add displacement and UB */
disps[1] = (MPI_Aint) (array_of_starts[0]);
size = 1;
for (i = 1; i < ndims; i++) {
size *= (MPI_Aint) (array_of_sizes[i - 1]);
disps[1] += size * (MPI_Aint) (array_of_starts[i]);
}
/* rest done below for both Fortran and C order */
}
else { /* order == MPI_ORDER_C */
/* dimension ndims-1 changes fastest */
if (ndims == 1) {
mpi_errno = MPIR_Type_contiguous_impl(array_of_subsizes[0], oldtype, &tmp1);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
} else {
mpi_errno = MPIR_Type_vector_impl(array_of_subsizes[ndims - 2],
array_of_subsizes[ndims - 1],
array_of_sizes[ndims - 1], oldtype, &tmp1);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
size = (MPI_Aint) (array_of_sizes[ndims - 1]) * extent;
for (i = ndims - 3; i >= 0; i--) {
size *= (MPI_Aint) (array_of_sizes[i + 1]);
mpi_errno = MPIR_Type_hvector_impl(array_of_subsizes[i], 1, size, tmp1, &tmp2);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
MPIR_Type_free_impl(&tmp1);
tmp1 = tmp2;
}
}
/* add displacement and UB */
disps[1] = (MPI_Aint) (array_of_starts[ndims - 1]);
size = 1;
for (i = ndims - 2; i >= 0; i--) {
size *= (MPI_Aint) (array_of_sizes[i + 1]);
disps[1] += size * (MPI_Aint) (array_of_starts[i]);
}
}
disps[1] *= extent;
disps[2] = extent;
for (i = 0; i < ndims; i++)
disps[2] *= (MPI_Aint) (array_of_sizes[i]);
disps[0] = 0;
blklens[0] = blklens[1] = blklens[2] = 1;
types[0] = MPI_LB;
types[1] = tmp1;
types[2] = MPI_UB;
mpi_errno = MPIR_Type_struct_impl(3, blklens, disps, types, newtype);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
MPIR_Type_free_impl(&tmp1);
fn_exit:
return mpi_errno;
fn_fail:
goto fn_exit;
}
static int MPII_Type_block(int *array_of_gsizes, int dim, int ndims, int nprocs,
int rank, int darg, int order, MPI_Aint orig_extent,
MPI_Datatype type_old, MPI_Datatype * type_new, MPI_Aint * st_offset)
{
/* nprocs = no. of processes in dimension dim of grid
rank = coordinate of this process in dimension dim */
int mpi_errno = MPI_SUCCESS;
int blksize, global_size, mysize, i, j;
MPI_Aint stride;
global_size = array_of_gsizes[dim];
if (darg == MPI_DISTRIBUTE_DFLT_DARG)
blksize = (global_size + nprocs - 1) / nprocs;
else {
blksize = darg;
MPIR_ERR_CHKINTERNAL(blksize <= 0, mpi_errno, "blksize must be > 0");
MPIR_ERR_CHKINTERNAL(blksize * nprocs < global_size, mpi_errno,
"blksize * nprocs must be >= global size");
}
j = global_size - blksize * rank;
mysize = (blksize < j) ? blksize : j;
if (mysize < 0)
mysize = 0;
stride = orig_extent;
if (order == MPI_ORDER_FORTRAN) {
if (dim == 0) {
mpi_errno = MPIR_Type_contiguous_impl(mysize, type_old, type_new);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
} else {
for (i = 0; i < dim; i++)
stride *= (MPI_Aint) (array_of_gsizes[i]);
mpi_errno = MPIR_Type_hvector_impl(mysize, 1, stride, type_old, type_new);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
}
} else {
if (dim == ndims - 1) {
mpi_errno = MPIR_Type_contiguous_impl(mysize, type_old, type_new);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
} else {
for (i = ndims - 1; i > dim; i--)
stride *= (MPI_Aint) (array_of_gsizes[i]);
mpi_errno = MPIR_Type_hvector_impl(mysize, 1, stride, type_old, type_new);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
}
}
*st_offset = blksize * rank;
/* in terms of no. of elements of type oldtype in this dimension */
if (mysize == 0)
*st_offset = 0;
fn_exit:
return mpi_errno;
fn_fail:
goto fn_exit;
}
