HDF5 Fortran interface. Integer 8 (double precision)

Hdf-forum archives
1

Hi all,

I'm trying to implement a layer to export the data of our application to
XDMF format on top of HDF5. The data
types that we want to handle are integers and reals of both single and
double precision.

I would like to know the actual state of the HDF5 Fortran interface for
handling double precision integers.

In particular, if we observe to the interface of the H5DWrite_f subroutine
below (and its comments) we can see that the
data type of the buffer (*buf*) could be INTEGER, and the data type of the
dimensions is HSIZE_T.

*SUBROUTINE h5dwrite_vl_f(dset_id, mem_type_id, buf, dims, len, hdferr, &
                     mem_space_id, file_space_id, xfer_prp)
  IMPLICIT NONE
  ...*

* INTEGER(HID_T), INTENT(IN) :: mem_type_id ! Memory datatype
identifier TYPE, INTENT(IN), & DIMENSION(dims(1),dims(2)) :: buf
                                            ! Data buffer; may be a scalar
                                            ! or an array
                                            ! TYPE must be one of the following
                                            ! INTEGER
                                            ! REAL
                                            ! CHARACTER
  INTEGER(HSIZE_T), INTENT(IN), DIMENSION(2) :: dims
                                            ! Array to hold corresponding
                                            ! dimension sizes of data
                                            ! buffer buf
                                            ! dim(k) has value of the k-th
                                            ! dimension of buffer buf
                                            ! Values are ignored if buf is
                                            ! a scalar ...*

I've read that the *HSIZE_T* data type depends on the architecture and
it's defined at HDF5 compilation time and I can check
that in my own compilation it's as double precision integer (64 bits).
Is there any case where this value is a single precision
integer (32)? In any case, how can I handle the writing of datasets
bigger than max(HSIZE_T)?

A different questions are about the data type of the raw data. In my
HDF5 compilation it seems that the H5DWrite_f procedure
doesn't compile if the *buf* actual argument is a double precision
integer. There is a native Fortran HDF5 *mem_type_id* for
double precision integers?

In some forum I've also read that the *H5T_NATIVE_INTEGER* could be a
doble precision integer. If this is true, it is posible to
handle single precision and double precision integers in the same
application/software?

Thanks in advance,

Víctor.

2

I would suggest you use the Fortran 2003 interface for h5dwrite_f instead:

SUBROUTINE h5dwrite_f(dset_id, mem_type_id, buf, hdferr, &
                        mem_space_id, file_space_id, xfer_prp)
    INTEGER(HID_T), INTENT(IN) :: dset_id
    INTEGER(HID_T), INTENT(IN) :: mem_type_id
    TYPE(C_PTR) , INTENT(IN) :: buf
    INTEGER , INTENT(OUT) :: hdferr
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: mem_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: file_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: xfer_prp

as you don’t have to worry about passing the dimensions of the array to the API.

You can also use ‘h5kind_to_type' to pass the appropriate type, for example
!
! Find the HDF type corresponding to the specified KIND
!
  h5_kind_type_i = h5kind_to_type(ip,H5_INTEGER_KIND)

Take a look at h5ex_d_rdwr_kind_F03.f90 for an example of how to handle different kinds of integers and reals,

https://www.hdfgroup.org/HDF5/examples/api18-fortran.html

Scot

···

On Nov 10, 2015, at 2:17 AM, victor sv <victorsv@gmail.com> wrote:

Hi all,

I'm trying to implement a layer to export the data of our application to XDMF format on top of HDF5. The data
types that we want to handle are integers and reals of both single and double precision.

I would like to know the actual state of the HDF5 Fortran interface for handling double precision integers.

In particular, if we observe to the interface of the H5DWrite_f subroutine below (and its comments) we can see that the
data type of the buffer (buf) could be INTEGER, and the data type of the dimensions is HSIZE_T.

SUBROUTINE h5dwrite_vl_f(dset_id, mem_type_id, buf, dims
, len, hdferr, &
                     mem_space_id, file_space_id, xfer_prp)
  IMPLICIT NONE
  ...

  INTEGER(HID_T), INTENT(IN) :: mem_type_id ! Memory datatype identifier
  TYPE, INTENT(IN), & DIMENSION(dims(1),dims(2)) :: buf
                                            ! Data buffer; may be a scalar
                                            ! or an array
                                            ! TYPE must be one of the following
                                            ! INTEGER
                                            ! REAL
                                            ! CHARACTER
  INTEGER(HSIZE_T), INTENT(IN), DIMENSION(2) :: dims
                                            ! Array to hold corresponding
                                            ! dimension sizes of data
                                            ! buffer buf
                                            ! dim(k) has value of the k-th
                                            ! dimension of buffer buf
                                            ! Values are ignored if buf is
                                            ! a scalar

   ...

I've read that the HSIZE_T data type depends on the architecture and it's defined at HDF5 compilation time and I can check
that in my own compilation it's as double precision integer (64 bits). Is there any case where this value is a single precision
integer (32)? In any case, how can I handle the writing of datasets bigger than max(HSIZE_T)?

A different questions are about the data type of the raw data. In my HDF5 compilation it seems that the H5DWrite_f procedure
doesn't compile if the buf actual argument is a double precision integer. There is a native Fortran HDF5 mem_type_id for
double precision integers?

In some forum I've also read that the H5T_NATIVE_INTEGER could be a doble precision integer. If this is true, it is posible to
handle single precision and double precision integers in the same application/software?

Thanks in advance,
Víctor.

_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@lists.hdfgroup.org
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5

3

Thank you Scott for the quick response.

It seems that the Fortran2003 interface for H5Dwrite solves both questions,
but there are other subroutines like H5Screate_simple_f or
H5Sselect_hyperslab_f using HSIZE_T integers to specify dimensions or
offsets.

I still have the following doubts:

   - HSIZE_T can be an integer of different kinds depending on the platform?
   - What is the easiest way to handle dataspaces with offsets or
   dimensions bigger than HSIZE_T?

Best regards,
Víctor.

···

2015-11-10 15:53 GMT+01:00 Scot Breitenfeld <brtnfld@hdfgroup.org>:

I would suggest you use the Fortran 2003 interface for h5dwrite_f instead:

SUBROUTINE h5dwrite_f(dset_id, mem_type_id, buf, hdferr, &
                        mem_space_id, file_space_id, xfer_prp)
    INTEGER(HID_T), INTENT(IN) :: dset_id
    INTEGER(HID_T), INTENT(IN) :: mem_type_id
    TYPE(C_PTR) , INTENT(IN) :: buf
    INTEGER , INTENT(OUT) :: hdferr
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: mem_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: file_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: xfer_prp

as you don’t have to worry about passing the dimensions of the array to
the API.

You can also use ‘h5kind_to_type' to pass the appropriate type, for example
!
! Find the HDF type corresponding to the specified KIND
!
  h5_kind_type_i = h5kind_to_type(ip,H5_INTEGER_KIND)

Take a look at h5ex_d_rdwr_kind_F03.f90 for an example of how to handle
different kinds of integers and reals,

https://www.hdfgroup.org/HDF5/examples/api18-fortran.html

Scot

> On Nov 10, 2015, at 2:17 AM, victor sv <victorsv@gmail.com> wrote:
>
> Hi all,
>
> I'm trying to implement a layer to export the data of our application to
XDMF format on top of HDF5. The data
> types that we want to handle are integers and reals of both single and
double precision.
>
> I would like to know the actual state of the HDF5 Fortran interface for
handling double precision integers.
>
> In particular, if we observe to the interface of the H5DWrite_f
subroutine below (and its comments) we can see that the
> data type of the buffer (buf) could be INTEGER, and the data type of the
dimensions is HSIZE_T.
>
> SUBROUTINE h5dwrite_vl_f(dset_id, mem_type_id, buf, dims
> , len, hdferr, &
> mem_space_id, file_space_id, xfer_prp)
> IMPLICIT NONE
> ...
>
> INTEGER(HID_T), INTENT(IN) :: mem_type_id ! Memory datatype identifier
> TYPE, INTENT(IN), & DIMENSION(dims(1),dims(2)) :: buf
> ! Data buffer; may be a
scalar
> ! or an array
> ! TYPE must be one of the
following
> ! INTEGER
> ! REAL
> ! CHARACTER
> INTEGER(HSIZE_T), INTENT(IN), DIMENSION(2) :: dims
> ! Array to hold corresponding
> ! dimension sizes of data
> ! buffer buf
> ! dim(k) has value of the
k-th
> ! dimension of buffer buf
> ! Values are ignored if buf
is
> ! a scalar
>
> ...
>
> I've read that the HSIZE_T data type depends on the architecture and
it's defined at HDF5 compilation time and I can check
> that in my own compilation it's as double precision integer (64 bits).
Is there any case where this value is a single precision
> integer (32)? In any case, how can I handle the writing of datasets
bigger than max(HSIZE_T)?
>
> A different questions are about the data type of the raw data. In my
HDF5 compilation it seems that the H5DWrite_f procedure
> doesn't compile if the buf actual argument is a double precision
integer. There is a native Fortran HDF5 mem_type_id for
> double precision integers?
>
> In some forum I've also read that the H5T_NATIVE_INTEGER could be a
doble precision integer. If this is true, it is posible to
> handle single precision and double precision integers in the same
application/software?
>
> Thanks in advance,
> Víctor.
>
>
>
> _______________________________________________
> Hdf-forum is for HDF software users discussion.
> Hdf-forum@lists.hdfgroup.org
> http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
> Twitter: https://twitter.com/hdf5

_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@lists.hdfgroup.org
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5

4

Thank you Scott for the quick response.

It seems that the Fortran2003 interface for H5Dwrite solves both questions, but there are other subroutines like H5Screate_simple_f or H5Sselect_hyperslab_f using HSIZE_T integers to specify dimensions or offsets.

I still have the following doubts:

  * HSIZE_T can be an integer of different kinds depending on the platform?

Correct, the size of hsize_t can be dependent on the platform.

  * What is the easiest way to handle dataspaces with offsets or dimensions bigger than HSIZE_T?

Keep in mind that the size of hsize_t is the same size as that found in C, that is by design. But I’m assuming that the issue you are referring to is that hsize_t in C is an unsigned integer and Fortran does not have unsigned integers. So, for example, if C’s hsize_t is of type long (8 bytes) the largest number will be 18,446,744,073,709,551,615. But in fortran, for an 8 byte integer, the largest number will be 9,223,372,036,854,775,807 because it is signed. So it is possible that if an HDF5 file was written in C and the number exceeds 9,223,372,036,854,775,807 (for example) then the fortran code will not be able to read the file.

Scot

Best regards,
Víctor.

2015-11-10 15:53 GMT+01:00 Scot Breitenfeld <brtnfld@hdfgroup.org<mailto:brtnfld@hdfgroup.org>>:
I would suggest you use the Fortran 2003 interface for h5dwrite_f instead:

SUBROUTINE h5dwrite_f(dset_id, mem_type_id, buf, hdferr, &
                        mem_space_id, file_space_id, xfer_prp)
    INTEGER(HID_T), INTENT(IN) :: dset_id
    INTEGER(HID_T), INTENT(IN) :: mem_type_id
    TYPE(C_PTR) , INTENT(IN) :: buf
    INTEGER , INTENT(OUT) :: hdferr
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: mem_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: file_space_id
    INTEGER(HID_T), INTENT(IN) , OPTIONAL :: xfer_prp

as you don’t have to worry about passing the dimensions of the array to the API.

You can also use ‘h5kind_to_type' to pass the appropriate type, for example
!
! Find the HDF type corresponding to the specified KIND
!
  h5_kind_type_i = h5kind_to_type(ip,H5_INTEGER_KIND)

Take a look at h5ex_d_rdwr_kind_F03.f90 for an example of how to handle different kinds of integers and reals,

https://www.hdfgroup.org/HDF5/examples/api18-fortran.html

Scot

···

On Nov 10, 2015, at 9:45 AM, victor sv <victorsv@gmail.com<mailto:victorsv@gmail.com>> wrote:

On Nov 10, 2015, at 2:17 AM, victor sv <victorsv@gmail.com<mailto:victorsv@gmail.com>> wrote:

Hi all,

I'm trying to implement a layer to export the data of our application to XDMF format on top of HDF5. The data
types that we want to handle are integers and reals of both single and double precision.

I would like to know the actual state of the HDF5 Fortran interface for handling double precision integers.

In particular, if we observe to the interface of the H5DWrite_f subroutine below (and its comments) we can see that the
data type of the buffer (buf) could be INTEGER, and the data type of the dimensions is HSIZE_T.

SUBROUTINE h5dwrite_vl_f(dset_id, mem_type_id, buf, dims
, len, hdferr, &
                     mem_space_id, file_space_id, xfer_prp)
  IMPLICIT NONE
  ...

  INTEGER(HID_T), INTENT(IN) :: mem_type_id ! Memory datatype identifier
  TYPE, INTENT(IN), & DIMENSION(dims(1),dims(2)) :: buf
                                            ! Data buffer; may be a scalar
                                            ! or an array
                                            ! TYPE must be one of the following
                                            ! INTEGER
                                            ! REAL
                                            ! CHARACTER
  INTEGER(HSIZE_T), INTENT(IN), DIMENSION(2) :: dims
                                            ! Array to hold corresponding
                                            ! dimension sizes of data
                                            ! buffer buf
                                            ! dim(k) has value of the k-th
                                            ! dimension of buffer buf
                                            ! Values are ignored if buf is
                                            ! a scalar

   ...

I've read that the HSIZE_T data type depends on the architecture and it's defined at HDF5 compilation time and I can check
that in my own compilation it's as double precision integer (64 bits). Is there any case where this value is a single precision
integer (32)? In any case, how can I handle the writing of datasets bigger than max(HSIZE_T)?

A different questions are about the data type of the raw data. In my HDF5 compilation it seems that the H5DWrite_f procedure
doesn't compile if the buf actual argument is a double precision integer. There is a native Fortran HDF5 mem_type_id for
double precision integers?

In some forum I've also read that the H5T_NATIVE_INTEGER could be a doble precision integer. If this is true, it is posible to
handle single precision and double precision integers in the same application/software?

Thanks in advance,
Víctor.

_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@lists.hdfgroup.org<mailto:Hdf-forum@lists.hdfgroup.org>
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5

_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@lists.hdfgroup.org<mailto:Hdf-forum@lists.hdfgroup.org>
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5

_______________________________________________
Hdf-forum is for HDF software users discussion.
Hdf-forum@lists.hdfgroup.org<mailto:Hdf-forum@lists.hdfgroup.org>
http://lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
Twitter: https://twitter.com/hdf5