Skip to content

tavianator/fpfd

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
doc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Please note that FPFD is currently in development, and cannot even remotely
begin to be useful for anyone needing an implementation of IEEE 754-2008 decimal
arithmetic.  For starters, FPFD currently only implements the decimal32 type for
arithmetic, which is not even specified as an arithmetic type by IEEE 754-2008.
This is to make life easier as development begins, and also because C's proposed
decimal floating-point extension treats decimal32 as an arithmetic type.  As
well, only some of the operations are implemented (addition and subtraction for
DPD, and multiplication for DPD).  Further, none of these implementations have
undergone extensive testing, and it is very likely that there are bugs.
Finally, FPFD does not, as yet, implement `cohort' handling correctly (or really
at all) as per IEEE 754-2008.

Additional `configure' options:

--enable-gcc-override: Override some of gcc's operations for its built-in
_Decimal32, ...64, and ...128 types.  Currently this overrides _Decimal32's
addition, subtraction, multiplication, and division operations with FPFD's
implementation (which is incomplete for all but a few).  Note that you must link
any programs which wish to use this feature with the appropriate encoding
library (libfpfd-bid for the BID encoding, which is default on x86 and x86_64,
or libfpfd-dpd for the DPD encoding).  This may be a nicer interface than FPFD's
C API, for examining the library.

Additional `make' targets:

Running `make bench' will run FPFD's benchmarking suite, which produces cycle
counts for FPFD's internal and user-facing routines.  If your site has gnuplot
installed, you may run

$ ./bench/plot.sh bench/dpd/fpfd32_add.dat     # Display the results
$ ./bench/plot.sh bench/dpd/fpfd32_add.dat png # Make a png file of the results

or similar to view the results.  The results are stored in
bench/<encoding>/<routine name>.dat.  FPFD also benchmarks some basic
architecture-related things; these results can be found in bench/arch.  Notably,
bench/arch/<architecture>-overhead.dat is a benchmark of an emptly loop, and
can be used to find the granularity of the benchmarks (usually within +/-0.05
cycles).  Lastly, FPFD can benchmark GCC's built-in arithmetic types, for
comparison, but ONLY if --enable-gcc-override is not specified.  These results
are found in bench/gcc/__<encoding>_<operation><operand size>.dat.  Here,
<operand size> is `sd' for decimal32, `dd' for decimal64, and `td' for
decimal128.  For the BID encoding, gcc uses Intel's IDFPL implementation, and
for the DPD encoding, gcc uses IBM's decNumber implementation.

About

Automatically exported from code.google.com/p/fpfd

Resources

License

LGPL-3.0, GPL-3.0 licenses found

Licenses found

LGPL-3.0
COPYING.LESSER
GPL-3.0
COPYING

Stars

Watchers

Forks

Releases

No releases published

Sponsor this project

 

Packages

No packages published

Languages