Attached is my FORTRAN subroutine for writing compound datatypes in HDF5 format and its output dumped in ascii file. The dataset "ep_data" is written without problem. However, my "den_data" dataset, which I mimicked as a simple compound type in the h5write subroutine, actually has variable length arrays and 1-member scalar fields. That is to say:
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want
to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you
explain what in h5ex_t_vlen_F03.f90 you want to avoid doing?
I am sorry that I could not make myself clear. What I meant to say is that I aim to form a compound datatype which comprises VL (variable length) and fixed-length array fields/datatypes. In the example for creating VL datatype, using a user-defined datatype which includes a pointer is suggested (Example 27 on page 227 in the HDF user manual). That is what I want to avoid but it seems it is the only way. What I want to achieve is summarized in Figure 18 on page 238 in the HDF user manual; a compound datatype built up of numerous VL and atomic datatypes.
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
Please let me put it this way. I have a user-defined type in FORTRAN called "ep_data" of dimension 4, i.e. ep_data(4). It has fields named biomass, pob, qob, ... etc. and so on. Each field holds a scalar value. For instance, if you print "ep_data(1)%biomass", it will give you, let's say, "4". In other words, fields of "ep_data" are not arrays. I succeeded to write this "ep_data" to HDF5 by using compound datatype. So no problem up till here. However, I have another user-defined type called "ms_data" in FORTRAN of size 7, i.e. , "ms_data(7)" and just like "ep_data", it also has fields named biomass, mortality, qob, ... etc. and so on. Differently from "ep_data" type, "ms_data" fields are arrays of different size. For instance, if you print "ms_data(1)%biomass(:)"|, you will get an array of real numbers. And if you print "ms_data(1)%mortality(:)", you also get an array of real numbers of different length that of "biomass" field. So, how can I store "ms_data" in HDF5 format?
For your reference, my FORTRAN implementation of storing "ep_data" is attached.
I am sorry that I could not make myself clear. What I meant to say is that I aim to form a compound datatype which comprises VL (variable length) and fixed-length array fields/datatypes. In the example for creating VL datatype, using a user-defined datatype which includes a pointer is suggested (Example 27 on page 227 in the HDF user manual). That is what I want to avoid but it seems it is the only way. What I want to achieve is summarized in Figure 18 on page 238 in the HDF user manual; a compound datatype built up of numerous VL and atomic datatypes.
Ekin
On 03/14/2012 07:16 AM, brtnfld@hdfgroup.org wrote:
Hi,
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
I missed to add a very important detail to my previous e-mail. As the sizes of ms_data(1)%biomass( and ms_data(1)%mortality( are not equal; also the sizes of ms_data(1)%biomass( and ms_data(2)%biomass( are not equal.
Regards,
Ekin
路路路
On 03/14/2012 01:49 PM, Ekin Akoglu wrote:
Hi,
Please let me put it this way. I have a user-defined type in FORTRAN called "ep_data" of dimension 4, i.e. ep_data(4). It has fields named biomass, pob, qob, ... etc. and so on. Each field holds a scalar value. For instance, if you print "ep_data(1)%biomass", it will give you, let's say, "4". In other words, fields of "ep_data" are not arrays. I succeeded to write this "ep_data" to HDF5 by using compound datatype. So no problem up till here. However, I have another user-defined type called "ms_data" in FORTRAN of size 7, i.e. , "ms_data(7)" and just like "ep_data", it also has fields named biomass, mortality, qob, ... etc. and so on. Differently from "ep_data" type, "ms_data" fields are arrays of different size. For instance, if you print "ms_data(1)%biomass(:)"|, you will get an array of real numbers. And if you print "ms_data(1)%mortality(:)", you also get an array of real numbers of different length that of "biomass" field. So, how can I store "ms_data" in HDF5 format?
For your reference, my FORTRAN implementation of storing "ep_data" is attached.
Regards,
Ekin
On 03/14/2012 10:00 AM, Ekin Akoglu wrote:
Hello,
I am sorry that I could not make myself clear. What I meant to say is that I aim to form a compound datatype which comprises VL (variable length) and fixed-length array fields/datatypes. In the example for creating VL datatype, using a user-defined datatype which includes a pointer is suggested (Example 27 on page 227 in the HDF user manual). That is what I want to avoid but it seems it is the only way. What I want to achieve is summarized in Figure 18 on page 238 in the HDF user manual; a compound datatype built up of numerous VL and atomic datatypes.
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
I found out to use a complex compound type circumvent my problem. However, I could not figure it out what is wrong with my implementation attached to this mail. It creates the complex compound structure but does not write the data into it. The h5dump is:
There are zeros where should be data instead. I would be happy if you spare some time to help me on this. Thank you in advance for your time and patience.
I missed to add a very important detail to my previous e-mail. As the sizes of ms_data(1)%biomass( and ms_data(1)%mortality( are not equal; also the sizes of ms_data(1)%biomass( and ms_data(2)%biomass( are not equal.
Regards,
Ekin
On 03/14/2012 01:49 PM, Ekin Akoglu wrote:
Hi,
Please let me put it this way. I have a user-defined type in FORTRAN called "ep_data" of dimension 4, i.e. ep_data(4). It has fields named biomass, pob, qob, ... etc. and so on. Each field holds a scalar value. For instance, if you print "ep_data(1)%biomass", it will give you, let's say, "4". In other words, fields of "ep_data" are not arrays. I succeeded to write this "ep_data" to HDF5 by using compound datatype. So no problem up till here. However, I have another user-defined type called "ms_data" in FORTRAN of size 7, i.e. , "ms_data(7)" and just like "ep_data", it also has fields named biomass, mortality, qob, ... etc. and so on. Differently from "ep_data" type, "ms_data" fields are arrays of different size. For instance, if you print "ms_data(1)%biomass(:)"|, you will get an array of real numbers. And if you print "ms_data(1)%mortality(:)", you also get an array of real numbers of different length that of "biomass" field. So, how can I store "ms_data" in HDF5 format?
For your reference, my FORTRAN implementation of storing "ep_data" is attached.
Regards,
Ekin
On 03/14/2012 10:00 AM, Ekin Akoglu wrote:
Hello,
I am sorry that I could not make myself clear. What I meant to say is that I aim to form a compound datatype which comprises VL (variable length) and fixed-length array fields/datatypes. In the example for creating VL datatype, using a user-defined datatype which includes a pointer is suggested (Example 27 on page 227 in the HDF user manual). That is what I want to avoid but it seems it is the only way. What I want to achieve is summarized in Figure 18 on page 238 in the HDF user manual; a compound datatype built up of numerous VL and atomic datatypes.
Ekin
On 03/14/2012 07:16 AM, brtnfld@hdfgroup.org wrote:
Hi,
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
I managed to solve the problem. Having fields of irregular array sizes under one compound type is possible by inserting each arrays into the compound type separately. For your reference, I attach my sample code and h5dump output. Thank you all for your valuable suggestions.
I found out to use a complex compound type circumvent my problem. However, I could not figure it out what is wrong with my implementation attached to this mail. It creates the complex compound structure but does not write the data into it. The h5dump is:
There are zeros where should be data instead. I would be happy if you spare some time to help me on this. Thank you in advance for your time and patience.
Regards,
Ekin
On 03/14/2012 02:18 PM, Ekin Akoglu wrote:
Hi,
I missed to add a very important detail to my previous e-mail. As the sizes of ms_data(1)%biomass( and ms_data(1)%mortality( are not equal; also the sizes of ms_data(1)%biomass( and ms_data(2)%biomass( are not equal.
Regards,
Ekin
On 03/14/2012 01:49 PM, Ekin Akoglu wrote:
Hi,
Please let me put it this way. I have a user-defined type in FORTRAN called "ep_data" of dimension 4, i.e. ep_data(4). It has fields named biomass, pob, qob, ... etc. and so on. Each field holds a scalar value. For instance, if you print "ep_data(1)%biomass", it will give you, let's say, "4". In other words, fields of "ep_data" are not arrays. I succeeded to write this "ep_data" to HDF5 by using compound datatype. So no problem up till here. However, I have another user-defined type called "ms_data" in FORTRAN of size 7, i.e. , "ms_data(7)" and just like "ep_data", it also has fields named biomass, mortality, qob, ... etc. and so on. Differently from "ep_data" type, "ms_data" fields are arrays of different size. For instance, if you print "ms_data(1)%biomass(:)"|, you will get an array of real numbers. And if you print "ms_data(1)%mortality(:)", you also get an array of real numbers of different length that of "biomass" field. So, how can I store "ms_data" in HDF5 format?
For your reference, my FORTRAN implementation of storing "ep_data" is attached.
Regards,
Ekin
On 03/14/2012 10:00 AM, Ekin Akoglu wrote:
Hello,
I am sorry that I could not make myself clear. What I meant to say is that I aim to form a compound datatype which comprises VL (variable length) and fixed-length array fields/datatypes. In the example for creating VL datatype, using a user-defined datatype which includes a pointer is suggested (Example 27 on page 227 in the HDF user manual). That is what I want to avoid but it seems it is the only way. What I want to achieve is summarized in Figure 18 on page 238 in the HDF user manual; a compound datatype built up of numerous VL and atomic datatypes.
Ekin
On 03/14/2012 07:16 AM, brtnfld@hdfgroup.org wrote:
Hi,
suggestions about storing this data as a compound type with variable
length fields without using pointers as explained in
h5ex_t_vlen_F03.f90 [1] example script on HDF GROUP web site. Thank
you in advance for your concerns.
I don't understand what you mean by "not using pointers", do you mean you don't want to use C_LOC and/or F2003 and still do the same thing as h5ex_t_vlen_F03.f90? Can you explain further what in h5ex_t_vlen_F03.f90 you want to avoid doing?
and ms_data(1)%mortality(
and ms_data(1)%mortality(