- Notifications
You must be signed in to change notification settings - Fork17
Fortran type-free variable and type-free dictionary
License
zerothi/fdict
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
A variable and dictionary in pure fortran for retaining any data-typeand a fast hash-table dictionary.
This module consists of two separate modules which co-exist formaintenance and usage reasons.
First, the variable module which is a type-free variable that can containany variable type, and any dimension as well.
Second, the dictionary module which contains a hash-table of variablesthat can containany data-type allowed by the variable module.
Installing fdict requires a download of the libraryhosted atgithub atfdict@git.
Installation can be done via 2 different back-ends. 1)smeka build system, or 2)CMake build.
Extract and create ansetup.make
file for compilation, a minimalsetup.make
file can look like this
FC=gfortranFFLAGS = -g
PREFIX =
Typemake
and a library calledlibfdict.a
is created.Subsequently the installation may be performed by:
make PREFIX=/path/to/fdict install
which installs the required files (modules and libraries) to the folder.It will also install pkg-config files for auto-detection.
CMake procedure can be done via the normal procedure:
cmake -S. -Bbuild-fdictcmake --build build-fdict
fdict should also be able to be used in a sub-project. If problemsoccur, feel free to open up an issue.
To use the dictionary you need to add include statements for themodules as well as linking to the program.
To link fdict to your program the following can be used in aMakefile
FDICT_PATH = /path/to/fdict/parentFDICT_LIBS = -L$(FDICT_PATH) -lfdictFDICT_INC = -I$(FDICT_PATH)
Alternatively, one can use pkg-config for obtaining the include flags andlibraries.
For parent programs that usesfdict
there are 2 ways of knowing whichfdict
version one is using:
- A simple header file (like C-preprocessor statements), this informationis found in
fdict.inc
- A
fypp
compatible include file which contains library version andwhich data types are included in the built library, see the filefdict.fypp
The filefdict.inc
may be included in projects which exposes the followingdefinitions:
_FDICT_MAJOR_ 0_FDICT_MINOR_ 9_FDICT_PATCH_ 0_FDICT_VERSION_ 0.9.0
This is mainly meant as a feature usable when the fdict interface ande.g. modules change names.
Alternatively thefdict.fypp
inclusion file exposes variables such as:
- the library version numbers (as above)
- which data-types are enabled
- the number of ranks for each kind
Thefdict.fypp
file is handy when you are already relying onfypp
whereas the regularfdict.inc
header files are easy to use in standardfortran source compilation.
Typically not needed: allows for customization of different interfaces.
By default the number of dimensions allowed by the library is 5, i.e.there is no interface created forreal a(:,:,:,:,:,:)
, etc. However,to accomodate arbitrary dimensions you must define constants in yoursetup.make
file.
There are several fine-tuning options that allows creating more or fewerinterfaces. As the number of dimensions increases, so does the librarysize. If only a specific maximum range of ranks are required, it mightbe beneficial to reduce maximum ranks to the used range.
Currently thefdict
library supports the types listed in the below table:
Type | Precision format (GNU) | C-type | Default |
---|---|---|---|
type(variable_t) | --- | yes | |
character(len=1) | char | yes | |
integer | selected_int_kind(2) | byte | no |
integer | selected_int_kind(4) | short | no |
integer | selected_int_kind(9) | int | yes |
integer | selected_int_kind(18) | long | yes |
real | selected_real_kind(6) | float | yes |
real | selected_real_kind(15) | double | yes |
real | selected_real_kind(18) | ext. double | no |
real | selected_real_kind(30) | quad | no |
complex | selected_real_kind(6) | float complex | yes |
complex | selected_real_kind(15) | double complex | yes |
complex | selected_real_kind(18) | ext. double complex | no |
complex | selected_real_kind(30) | quad complex | no |
logical | selected_int_kind(2) | byte | no |
logical | selected_int_kind(4) | short | no |
logical | selected_int_kind(9) | int | yes |
logical | selected_int_kind(18) | long | no |
type(c_ptr) | void * | no | |
type(c_funptr) | (procedure)void * | no |
In theDefault
column one can see which data-types are enabled by default. The mostcommonly used data-types are enabled.
To enable the non-default data types you can do so with (Makefile scheme):
FYPPFLAGS += -DWITH_INT8=1 # for int kind(2)FYPPFLAGS += -DWITH_INT16=1 # for int kind(4)# Note that not all compilers support extended precisions# If you experience compiler errors, this is likely the cause.FYPPFLAGS += -DWITH_REAL80=1 # for real and complex kind(18)FYPPFLAGS += -DWITH_REAL128=1 # for real and complex kind(30)FYPPFLAGS += -DWITH_LOG8=1 # for logical kind(2)FYPPFLAGS += -DWITH_LOG16=1 # for logical kind(4)FYPPFLAGS += -DWITH_LOG64=1 # for logical kind(18)FYPPFLAGS += -DWITH_ISO_C=1 # for enabling c_ptr and c_funptr
Forcmake
the variables are all prefixed withFDICT_
, e.g.-DFDICT_FYPPFLAGS
,to ensure there are no clashes with parent programs.
By defaultfdict
generates the kind specifications from theselected_*_kind
routines,however, if one wishes to use theiso_fortran_env
module simply addFYPPFLAGS += -DWITH_ISO_ENV
.
To control the maximum ranks in the interfaces one can add these:
# type(c_ptr), type(c_funptr) and character(len=1)# are data types that are not affected by the MAXRANK variable# globally define the maximum ranks of all but the above listedFYPPFLAGS += -DMAXRANK=n# integer(*) types maximum rankFYPPFLAGS += -DMAXRANK_INT=n# real(*) types maximum rankFYPPFLAGS += -DMAXRANK_REAL=n# complex(*) types maximum rankFYPPFLAGS += -DMAXRANK_CMPLX=n# logical(*) types maximum rankFYPPFLAGS += -DMAXRANK_LOG=n# type(c_ptr), type(c_funptr) types maximum rankFYPPFLAGS += -DMAXRANK_ISO_C=n
Using this module one gains access to a generic type variable whichcan containany data format.
It is used like this:
use variableinteger :: a(3type(variable_t) :: va = 2call assign(v,a)a = 3call assign(a,v)
Also the variable contains an abbreviation for assigning pointers tonot copy data, but retain data locality:
integer, target :: a(3)type(variable_t) :: va = 2call associate(v,a)a = 3! Now v contains a = 3
To delete a variable do:
use variabletype(variable_t) :: vcall delete(v)
However, when the variable is using pointers, instead the user can do
use variabletype(variable_t) :: v! preferredcall nullify(v)! orcall delete(v,dealloc=.false.)
which merely destroys the variable object and thus retains the datawhere it is. As with any other pointer arithmetic it is up to the programmerto ensure there is no memory leaks.
In some cases one does not know which data-type is being stored in a variable.Here it may be beneficial to lookup the type of data:
use variableinteger, target :: a(3)type(variable_t) :: va(:) = 2call associate(v,a)if ( which(v) == which(a) ) then ! signal integer of 1D (i0 for scalar) call assign(a, v)end if! Another possibility is to *try* to get the valuelogical :: successinteger, target :: i1(3)real, target :: r1(3)call assign(r1, v, success=success)if ( .not. success ) then call assign(i1, v, success=success)end if... etc ...
However, it may be better to explicitly check the type usingwhich
.For consistency and API changes, it is encouraged to usewhich(<type>)
toensure that the data-types are as expected. I.e.which([real(real64) ::])
is the preferred way of forcing a data-type contained in a variable.
Usingtype(variable_t)
it becomes easy to create dictionaries in fortran.
Using this module we implement a dictionary which can containany dataformat using akey:val
based formalism. The underlying data structure is alinked list sorted according to hash-values of the keys. Hence searchingfor specific elements in the dictionary isextremely fast. Searching adictionary with 100 keys 300000 times takes less than 0.04 seconds ona Haswell laptop.Concatenating dictionaries is also very fast.
Creating a dictionary is almost as easy as the Python equivalent:
use dictionarytype(dictionary_t) :: dictdict = ('KEY'.kv.1)
To extend a dictionary one uses the concatenating format:
dict = dict // ('Hello'.kv.'world') // ('No'.kv.'world')
Again as is used by thetype(variable_t)
one can with benefit use.kvp.
to createthe dictionary value by pointers instead of copying the content.Hence doing:
real :: r(4)dict = dict // ('reals'.kvp.r)r = 4
will change the value in the dictionary.Note that one can easily create memory leaks with dictionaries:
use dictionarytype(dictionary_t) :: dictdict = ('KEY'.kv.1)dict = dict // ('KEY'.kv.2)dict = ('KEY'.kv.3)
The 1st assignement is valid since the dictionary is empty.The 2nd assignment concatenates and does not produce any memory leaks.In that case the old keyKEY
is deleted and the new value2
is inserted.The 3rd assignment produces a memory leak since the pointer to the originaldictionary gets lost. Be sure to callcall delete(dict)
prior to singleassignments.
There are various ways to access the data in a dictionary.
Accessing specific keys may be exercised using
use dictionary type(dictionary_t) :: dict type(variable_t) :: var integer :: i real :: r logical :: success dict = ('KEY'.kv.1) call assign(r, dict, 'KEY', success=success) if ( .not. success ) call assign(i, dict, 'KEY', success=success) call assign(var, dict, 'KEY')
Since values in dictionaries are stored using
variable_t
we have tofollow the limitations of that implementation. Therefore it may be betterto always use a temporaryvariable_t
to retrieve the values stored. Thiswill remove a redundant lookup in the dictionary.Users may find the
.key.
and.value.
operators which only acts on the firstelement of the dictionary (which may be a surprise). This is only useful for loopingdictionaries.use dictionary type(dictionary_t) :: dict, dict_first type(variable_t) :: var character(DICTIONARY_KEY_LENGTH) :: key integer :: i real :: r logical :: success dict = ('KEY'.kv.1) dict = dict // ('KEY1'.kv.3) ! start looping dict_first = .first. dict do while ( .not. (.empty. dict_first) ) ! now .key. and .value. could be used: key = .key. dict_first call assign(var, dict_first) ! Get next dictionary entry dict_first = .next. dict_first end while
Note that the dictionary can also containany data type.
However, if it needs to do custom data-types the programmer needs toextend the code by supplying a few custom routines.
Intrinsically the dictionary can contain dictionaries by this:
use dictionarytype(dictionary_t) :: d1, d2d1 = ('hello'.kv.'world')d2 = ('hello'.kv.'world')d1 = d1 // ('dict'.kvp.d2)
But it will be up to the user toknow the key for data types other thanintegers, reals, complex numbers, characters andc_*
extension types.
Note that the dictionary contained is passed by reference, and thusif you deleted2
, you will have a dangling pointer ind1
.
I would advice any users to contribute as much feedback and/or PRs to furthermaintain and expand this library.
Please do not hesitate to contribute!
If you find any bugs please form abug report/issue.
If you have a fix please consider adding apull request.
The fdict license isMPL-2.0, see the LICENSE file.
A big thanks goes to Alberto Garcia for contributing ideas and givingme bug reports. Without him the interface would have been much morecomplex!
About
Fortran type-free variable and type-free dictionary