TAF

Summary:
Transformation of Algorithms in Fortran (TAF) is a source-to-source AD-tool for Fortran-95 programs. TAF supports forward and reverse mode of AD and Automatic Sparsity Detection (ASD) for detection of the sparsity structure of Jacobians.

URL: http://fastopt.de/products/taf/taf.shtml

Developers:

• Ralf Giering, FastOpt
  

Mode:	Forward Reverse
Method:	Source transformation
Supported Language:	Fortran 77/90 Fortran2003 Fortran2008 Fortran2018 Fortran77 Fortran95

Reference:


Features:

Fortran-95 Support:

TAF supports the full Fortran-95 standard.

Fortran-2003 Support:

TAF supports allmost all of the Fortran-2003 standard.

Fortran-2008 Support:

TAF supports most of the Fortran-2008 standard.

Fortran-2018 Support:

TAF supports most of the Fortran-2018 standard.

MPI Support:

TAF supports collective communications, blocking, and non-blocking point-to-point communications

OpenMP Support:

TAF supports most of OpenMP-3.1

Analyses:

TAF normalizes the code and applies a control flow analysis. TAF applies
an intraprocedural data dependence and an interprocedural data flow analysis.
Given the independent and dependent variables of the specified top-level
routine, TAF determines all active routines and variables and produces
derivative code only for those.

Performance:

TAF predicts the computational costs of code fragments for an (sub)optimal
decision between recomputation and storing of intermediate results in reverse
mode.

Directives:

TAF accepts several kinds of directives. Using the reverse mode storing
of variables is controlled by directives.


Multi level checkpointing can be implemented by splitting a loop and
inserting directives.

Black box (library) routines are handled using flow information given
by directives.

Awards

Heinz-Billing-Award 1995 for the advancement of scientific computing

Supported Platforms:

• Application Server

Licensing: license

Entries in our publication database that actually use TAF in the numerical experiments:  66

The following diagram shows these entries versus the year of the publication.

10+ #Entries 0	1	1	1	11	18	7	14	5	3	2	1	1	1
	'96	'98	'01	'02	'03	'04	'05	'06	'07	'08	'09	'11	'16
	Year

Related Research Groups:

• FastOpt, Germany