kml_fft(f/h)_plan_guru_split_dft_r2c
Create a plan for the n-dimensional R2C transform of multiple data sequences.
The data sequence of a single FFT does not need to be contiguous and can be strided. The input and output of the split interfaces are stored in the real and imaginary arrays, respectively.
Interface Definition
C interface:
kml_fft_plan kml_fft_plan_guru_split_dft_r2c(int rank, const kml_fft_iodim *dims, int howmany_rank, const kml_fft_iodim *howmany_dims, double *in, double *ro, double *io, unsigned flags);
kml_fftf_plan kml_fftf_plan_guru_split_dft_r2c(int rank, const kml_fftf_iodim *dims, int howmany_rank, const kml_fftf_iodim *howmany_dims, float *in, float *ro, float *io, unsigned flags);
kml_ffth_plan kml_ffth_plan_guru_split_dft_r2c(int rank, const kml_ffth_iodim *dims, int howmany_rank, const kml_ffth_iodim *howmany_dims, __fp16 *in, __fp16 *ro, __fp16 *io, unsigned flags);
Fortran interface:
RES = KML_FFT_PLAN_GURU_SPLIT_DFT_R2C(RANK, DIMS, HOWMANY_RANK, HOWMANY_DIMS, IN, RO, IO, FLAGS);
RES = KML_FFTF_PLAN_GURU_SPLIT_DFT_R2C(RANK, DIMS, HOWMANY_RANK, HOWMANY_DIMS, IN, RO, IO, FLAGS);
RES = KML_FFTH_PLAN_GURU_SPLIT_DFT_R2C(RANK, DIMS, HOWMANY_RANK, HOWMANY_DIMS, IN, RO, IO, FLAGS);
Return Value
The function returns a structure pointer of the kml_fft(f/h)_plan type. This object can be passed to the kml_fft(f/h)_execute function as a parameter to perform FFT on the input and outputs ro and io. This object can also be passed to the kml_fft(f/h)_execute_split_dft_r2c function as a parameter to perform FFT on new input and outputs ro and io.
If this function returns a non-null pointer, the plan has been successfully created. Otherwise, the plan fails to be created.
Parameters
Parameter |
Data Type |
Description |
Input/Output |
|---|---|---|---|
rank |
int |
Dimension of a single FFT sequence. Constraint: 1 ≤ rank ≤ 4 |
Input |
dims |
|
dims is a structure array whose size is rank. dims[i] contains the following members:
Constraint: dims[i].n ≥ 1, for i in 0 to rank - 1. |
Input |
howmany_rank |
int |
The memory allocation between multiple rank-dimension FFTs is described by the howmany_dims array of the howmany_rank dimension. howmany_rank indicates the number of dimensions required by the memory access mode for the start address of each rank-dimension FFT to be calculated. Constraint: 0 ≤ howmany_rank ≤ 4 |
Input |
howmany_dims |
|
howmany_dims is a structure array whose size is howmany_rank. howmany_dims[i] contains the following members:
|
Input |
in |
|
Inputs the data to be transformed. |
Input |
ro |
|
Outputs the real part of the data to be transformed. |
Input |
io |
|
Outputs the imaginary part of the data to be transformed. |
Input |
flags |
unsigned int |
Planning option, which describes the ESTIMATE or PATIENT mode. KML_FFT_ESTIMATE: ESTIMATE mode KML_FFT_PATIENT: PATIENT mode |
Input |
Dependencies
C: "kfft.h"
Examples
C interface:
int rank = 2;
kml_fft_iodim *dims;
dims = (kml_fft_iodim*)kml_fft_malloc(sizeof(kml_fft_iodim) * rank);
dims[0].n = 2;
dims[0].is = 3;
dims[0].os = 2;
dims[1].n = 3;
dims[1].is = 1;
dims[1].os = 1;
int howmany_rank = 1;
kml_fft_iodim *howmany_dims;
howmany_dims = (kml_fft_iodim*)kml_fft_malloc(sizeof(kml_fft_iodim) * howmany_rank);
howmany_dims[0].n = 2;
howmany_dims[0].is = 2 * 3;
howmany_dims[0].os = 2 * 2;
double init[12] = {120, 8, 0, 0, 0, -8, -8, -8, -16, 0, -40, -8};
double *in;
in = (double*)kml_fft_malloc(sizeof(double) * 12);
for (int i = 0; i < 12; i++) {
in[i] = init[i];
}
kml_fft_plan plan;
double *ro;
double *io;
ro = (double*)kml_fft_malloc(sizeof(double) * 8);
io = (double*)kml_fft_malloc(sizeof(double) * 8);
plan = kml_fft_plan_guru_split_dft_r2c(rank, dims, howmany_rank, howmany_dims, in, ro, io, KML_FFT_ESTIMATE);
kml_fft_execute_split_dft_r2c(plan, in, ro, io);
kml_fft_destroy_plan(plan);
kml_fft_free(howmany_dims);
kml_fft_free(dims);
kml_fft_free(in);
kml_fft_free(ro);
kml_fft_free(io);
/*
* ro = {1.200000e+02, 1.200000e+02, 1.360000e+02, 1.120000e+02,
* -8.000000e+01, 2.800000e+01, 1.600000e+01, -2.000000e+01}
*/
/*
* io = {0.000000e+00, -1.385641e+01, 0.000000e+00, 0.000000e+00,
* 0.000000e+00, 2.078461e+01, 0.000000e+00, -3.464102e+01}
*/
Fortran interface:
INTEGER(C_INT) :: RANK = 2
INTEGER(C_INT) :: HOWMANY_RANK = 1
TYPE(KML_FFT_IODIM), POINTER :: DIMS(:), HOWMANY_DIMS(:)
REAL(C_DOUBLE), DIMENSION(12) :: INIT
TYPE(C_DOUBLE), POINTER :: IN(:), RO(:), IO(:)
TYPE(C_PTR) :: PIN, PRO, PIO, PDIMS, PHOWMANY_DIMS
INTEGER(C_SIZE_T) :: SIZE1, SIZE2, SIZE3, SIZE4
SIZE1 = 8 * 12
SIZE2 = 8 * 8
SIZE3 = 12 * RANK
SIZE4 = 12 * HOWMANY_RANK
PDIMS = KML_FFT_MALLOC(SIZE3)
PHOWMANY_DIMS = KML_FFT_MALLOC(SIZE4)
PIN = KML_FFT_MALLOC(SIZE1)
PRO = KML_FFT_MALLOC(SIZE2)
PRI = KML_FFT_MALLOC(SIZE2)
CALL C_F_POINTER(PIN, IN, SHAPE=[12])
CALL C_F_POINTER(PRO, RO, SHAPE=[8])
CALL C_F_POINTER(PIO, IO, SHAPE=[8])
CALL C_F_POINTER(PDIMS, DIMS, SHAPE=[RANK])
CALL C_F_POINTER(PHOWMANY_DIMS, HOWMANY_DIMS, SHAPE=[HOWMANY_RANK])
DIMS(0)%N = 2
DIMS(0)%IS = 3
DIMS(0)%OS = 2
DIMS(1)%N = 3
DIMS(1)%IS = 1
DIMS(1)%OS = 1
HOWMANY_DIMS(0)%N = 2
HOWMANY_DIMS(0)%IS = 2 * 3
HOWMANY_DIMS(0)%OS = 2 * 3
DATA INIT/120, 8, 0, 0, 0, -8, -8, -8, -16, 0, -40, -8/
INTEGER :: I
DO WHILE(I <= 12)
IN(I) = INIT(I)
I = I + 1
END DO
TYPE(C_PTR) :: PLAN
PLAN = KML_FFT_PLAN_GURU_SPLIT_DFT_R2C(RANK, DIMS, HOWMANY_RANK, HOWMANY_DIMS, IN, RO, IO, KML_FFT_ESTIMATE)
CALL KML_FFT_EXECUTE_SPLIT_DFT_R2C(PLAN, IN, RO, IO)
CALL KML_FFT_DESTROY_PLAN(PLAN)
CALL KML_FFT_FREE(PHOWMANY_DIMS)
CALL KML_FFT_FREE(PDIMS)
CALL KML_FFT_FREE(PIN)
CALL KML_FFT_FREE(PRO)
CALL KML_FFT_FREE(PIO)
!
! RO = /1.200000E+02, 1.200000E+02, 1.360000E+02, 1.120000E+02,
! -8.000000E+01, 2.800000E+01, 1.600000E+01, -2.000000E+01/
!
!
! IO = /0.000000E+00, -1.385641E+01, 0.000000E+00, 0.000000E+00,
! 0.000000E+00, 2.078461E+01, 0.000000E+00, -3.464102E+01/
!
The plan returned by kml_fft(f)_plan_guru_split_dft_r2c can use kml_fft(f)_execute to calculate the plan input or kml_fft(f)_execute_split_dft_r2c to calculate a different input. The input of a plan does not need to be initialized, but the input must be valid (cannot be NULL).