#ifndef HALIDE_FFT_H #define HALIDE_FFT_H #include #include #include #include #include "Halide.h" #include "complex.h" // This is an optional extra description for the details of computing an FFT. struct Fft2dDesc { // Gain to apply to the FFT. This is folded into gains already being applied // to the FFT when possible. Halide::Expr gain = 1.0f; // The following option specifies that a particular vector width should be // used when the vector width can change the results of the FFT. // Some parts of the FFT algorithm use the vector width to change the way // floating point operations are ordered and grouped, which causes the results // to vary with respect to the target architecture. Setting this option forces // such stages to use the specified vector width (independent of the actual // architecture's vector width), which eliminates the architecture specific // behavior. int vector_width = 0; // The following option indicates that the FFT should parallelize within a // single FFT. This only makes sense to use on large FFTs, and generally only // if there is no outer loop around FFTs that can be parallelized. bool parallel = false; // This option will schedule the input to the FFT at the innermost location // that makes sense. bool schedule_input = false; // A name to prepend to the name of the Funcs the FFT defines. std::string name = ""; }; // Compute the N0 x N1 2D complex DFT of the first 2 dimensions of a complex // valued function x. The first 2 dimensions of x should be defined on at least // [0, N0) and [0, N1) for dimensions 0, 1, respectively. sign = -1 indicates a // forward FFT, sign = 1 indicates an inverse FFT. There is no normalization of // the FFT in either direction, i.e.: // // X = fft2d_c2c(x, N0, N1, -1); // x = fft2d_c2c(X, N0, N1, 1) / (N0 * N1); ComplexFunc fft2d_c2c(ComplexFunc x, int N0, int N1, int sign, const Halide::Target &target, const Fft2dDesc &desc = Fft2dDesc()); // Compute the N0 x N1 2D complex DFT of the first 2 dimensions of a real valued // function r. The first 2 dimensions of r should be defined on at least [0, N0) // and [0, N1) for dimensions 0, 1, respectively. Note that the transform domain // has dimensions N0 x N1 / 2 + 1 due to the conjugate symmetry of real DFTs. // There is no normalization. ComplexFunc fft2d_r2c(Halide::Func r, int N0, int N1, const Halide::Target &target, const Fft2dDesc &desc = Fft2dDesc()); // Compute the real valued N0 x N1 2D inverse DFT of dimensions 0, 1 of c. Note // that the transform domain has dimensions N0 x N1 / 2 + 1 due to the conjugate // symmetry of real DFTs. There is no normalization. Halide::Func fft2d_c2r(ComplexFunc c, int N0, int N1, const Halide::Target &target, const Fft2dDesc &desc = Fft2dDesc()); #endif