History

Ye Luo b63b331d32 Add safe-guard in fft_parallel.f90		2020-07-05 17:50:15 -05:00
..
examples	execution order of the ffts in the scalar case (cftt3ds) changed from	2016-07-07 15:44:16 +00:00
FFTXlib.md	Added FORD documentation for test.f90	2016-08-05 09:14:08 +00:00
Makefile	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
README.md	List all source files in README.md	2020-07-02 11:40:27 -05:00
fft_error.f90	QE made ready to use MPI module instead of mpif.h include file. See the three	2016-11-25 14:11:28 +00:00
fft_fwinv.f90	new FFTXlib based on OpenMP	2020-02-13 16:03:41 +00:00
fft_ggen.f90	fft_set_nl and fft_set_nlm unified in a single routine. mill is optional output.	2018-01-24 07:31:29 +01:00
fft_helper_subroutines.f90	See comments in https://gitlab.com/QEF/q-e/merge_requests/721	2019-12-29 20:39:59 +01:00
fft_interfaces.f90	AGAIN on the develop branch ...	2018-02-08 14:23:10 +01:00
fft_interpolate.f90	define a generic interface for fft_interpolate and move the correspondig routine to FFTXlib	2018-01-08 23:02:08 +01:00
fft_parallel.f90	Add safe-guard in fft_parallel.f90	2020-07-05 17:50:15 -05:00
fft_param.f90	adding nl and nlm to fft_type	2017-12-24 00:15:56 +01:00
fft_scalar.ARM_LIB.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_scalar.DFTI.f90	Rename __FFT_USE_OMP_THREADPRIVATE to __FFT_SCALAR_THREAD_SAFE	2020-05-18 18:34:55 -05:00
fft_scalar.ESSL.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_scalar.FFTW.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_scalar.FFTW3.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_scalar.SX6.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_scalar.f90	Add thread-safety guard in FFTXLIB.	2020-07-05 09:17:23 -05:00
fft_smallbox.f90	MAJOR restructuring of the FFTXlib library	2017-08-01 20:31:02 +00:00
fft_smallbox_type.f90	MAJOR restructuring of the FFTXlib library	2017-08-01 20:31:02 +00:00
fft_stick.c	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
fft_support.f90	Minor changes to FFTXlib. The only change (marginally) affecting	2019-04-25 21:06:44 +02:00
fft_types.f90	add OpenMP Tasks	2020-05-19 10:39:57 +01:00
fftw.c	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
fftw.h	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
fftw_dp.c	- adding static in the function declaration as well	2019-05-25 08:54:04 +02:00
fftw_dp.h	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
fftw_interfaces.f90	Missing interfaces for FFTW + misspells, noticed by Anton	2017-08-08 14:17:21 +00:00
fftw_sp.c	- adding static in the function declaration as well	2019-05-25 08:54:04 +02:00
fftw_sp.h	- internal new single precision fft driver	2019-05-25 08:39:53 +02:00
gen_test_params.py	Added option to set max k value used in the definition of g vectors. Added small python script to simplify the definition of command line arguments for fft_test.x.	2018-03-13 18:19:29 +01:00
konst.h	- FFT Modules replaced by FFTXlib	2015-11-21 10:37:48 +00:00
make.depend	Minor fixes	2019-12-02 11:35:54 +01:00
scatter_mod.f90	change howmany and iproc loops position	2020-05-21 15:07:49 +01:00
stick_base.f90	a forgotten deallocation created a serious memory leak when fft is initialized several times.	2017-08-24 15:29:52 +00:00
test.f90	fix an error in taskgroups	2020-03-16 12:28:16 +00:00
test0.f90	new interface to fft calls	2018-01-02 17:45:45 +01:00
tg_gather.f90	missing file in previous commit.	2017-08-02 07:36:06 +00:00

README.md

FFTXlib

Implements real space grid parallelization of FFT and task groups.

Testing and Benchmarking

This library also provides a testing and timing code to asses the performance of your FFT, estimate the scalability and the optimal parameters for your simulation.

To compile the test program, once you have properly configure QE within a parallel environment, go inside the directory FFTXlib and type:

make TEST

Then you can run your FFT tests using command like:

mpirun -np 4 ./fft_test.x -ecutwfc 80 -alat 20  -nbnd 128 -ntg 4

Command line arguments:

-ecutwfc  Plane wave energy cut off
-alat     Lattice parameter (for hard coded lattice structure)
-nbnd     Number of bands (fft cycles)
-ntg      Number of task groups
-av1  x y z    First lattice vector, in atomic units. N.B.: when using -av1, -alat is ignored!
-av2  x y z    Second lattice vector, in atomic units. N.B.: when using -av2, -alat is ignored!
-av3  x y z    Third lattice vector, in atomic units. N.B.: when using -av3, -alat is ignored!
-kmax kx ky kz    Reciprocal lattice vector inside the BZ with maximum norm. Used to calculate max(|G+K|). (2pi/a)^2 units.

A python script to extract the parameters from an output file of pw.x is also available. Example usage:

$ python gen_test_params.py a_pw_output
To analize performances run with:
mpirun -np X ./fft_test.x -ntg Y -ecutwfc 36.7500 -ecutrho 147.0000 -av1 36.6048 0.0 0.0 -av2 -18.3024 31.70067192 0.0 -av3 0.0 0.0 18.3024 -nbnd 400 -gamma .true.

Replace X and Y with appropriate values for your simualtion.

Files

Compile time parameters:

fft_param.f90

Descriptor types:

stick_base.f90
fft_types.f90
fft_smallbox_type.f90

Parallel execution routines:

fft_interfaces.f90 fft_fwinv.f90
  fft_parallel.f90
    scatter_mod.f90
      tg_gather.f90
  fft_interpolate.f90
  fft_smallbox.f90

Low level library wrappers:

fft_scalar.f90
fft_scalar.ARM_LIB.f90
fft_scalar.DFTI.f90
fft_scalar.ESSL.f90
fft_scalar.FFTW.f90 fftw_interfaces.f90
  fft_stick.c fftw.c fftw_dp.c fftw_sp.c fftw_dp.h fftw.h fftw_sp.h konst.h
fft_scalar.FFTW3.f90
fft_scalar.SX6.f90

Misc. helper routines:

fft_ggen.f90
fft_error.f90
fft_helper_subroutines.f90
fft_support.f90

Tests:

test0.f90
test.f90