Skip to main content

oxifft_codegen/
lib.rs

1//! `OxiFFT` Codelet Generator
2//!
3//! This proc-macro crate generates optimized FFT codelets at compile time.
4//! It replaces FFTW's OCaml-based genfft with Rust procedural macros.
5//!
6//! # Overview
7//!
8//! Codelets are highly optimized kernels for small FFT sizes (2-64).
9//! They are generated at compile time with:
10//! - Common subexpression elimination
11//! - Strength reduction
12//! - Optimal instruction ordering
13//! - SIMD-aware code patterns
14//!
15//! # Usage
16//!
17//! ```ignore
18//! use oxifft_codegen::gen_dft_codelet;
19//!
20//! // Generate size-8 DFT codelet
21//! gen_dft_codelet!(8);
22//! ```
23
24extern crate proc_macro;
25
26use proc_macro::TokenStream;
27
28/// Generate a non-twiddle (base case) DFT codelet.
29///
30/// # Arguments
31/// * `size` - The FFT size (must be 2, 4, 8, 16, 32, or 64)
32///
33/// # Example
34/// ```ignore
35/// gen_notw_codelet!(8);
36/// ```
37#[proc_macro]
38pub fn gen_notw_codelet(input: TokenStream) -> TokenStream {
39    let input2: proc_macro2::TokenStream = input.into();
40    oxifft_codegen_impl::gen_notw::generate(input2)
41        .unwrap_or_else(|e| e.to_compile_error())
42        .into()
43}
44
45/// Generate a twiddle-factor DFT codelet.
46#[proc_macro]
47pub fn gen_twiddle_codelet(input: TokenStream) -> TokenStream {
48    let input2: proc_macro2::TokenStream = input.into();
49    oxifft_codegen_impl::gen_twiddle::generate(input2)
50        .unwrap_or_else(|e| e.to_compile_error())
51        .into()
52}
53
54/// Generate a split-radix twiddle codelet.
55///
56/// The split-radix FFT decomposes an N-point DFT into one N/2-point DFT
57/// (even-indexed elements) and two N/4-point DFTs (odd-indexed elements)
58/// with twiddle factors `W_N^k` and `W_N^{3k`}, reducing the total multiply count.
59///
60/// # Usage
61/// ```ignore
62/// // Generate generic runtime-parameterized split-radix twiddle codelet
63/// gen_split_radix_twiddle_codelet!();
64///
65/// // Generate specialized unrolled version for N=8
66/// gen_split_radix_twiddle_codelet!(8);
67///
68/// // Generate specialized unrolled version for N=16
69/// gen_split_radix_twiddle_codelet!(16);
70/// ```
71#[proc_macro]
72pub fn gen_split_radix_twiddle_codelet(input: TokenStream) -> TokenStream {
73    let input2: proc_macro2::TokenStream = input.into();
74    oxifft_codegen_impl::gen_twiddle::generate_split_radix(input2)
75        .unwrap_or_else(|e| e.to_compile_error())
76        .into()
77}
78
79/// Generate a SIMD-optimized codelet.
80#[proc_macro]
81pub fn gen_simd_codelet(input: TokenStream) -> TokenStream {
82    let input2: proc_macro2::TokenStream = input.into();
83    oxifft_codegen_impl::gen_simd::generate(input2)
84        .unwrap_or_else(|e| e.to_compile_error())
85        .into()
86}
87
88/// Convenience macro to generate all codelets for a size.
89#[proc_macro]
90pub fn gen_dft_codelet(input: TokenStream) -> TokenStream {
91    let input2: proc_macro2::TokenStream = input.into();
92    oxifft_codegen_impl::gen_notw::generate(input2)
93        .unwrap_or_else(|e| e.to_compile_error())
94        .into()
95}
96
97/// Generate a real-to-half-complex (R2HC) or half-complex-to-real (HC2R) codelet.
98///
99/// The generated function has the same signature and produces numerically equivalent
100/// results to the hand-written codelets in `oxifft/src/rdft/codelets/mod.rs`.
101///
102/// # Usage
103/// ```ignore
104/// use oxifft_codegen::gen_rdft_codelet;
105///
106/// // Generates `pub fn r2hc_4_gen<T: crate::kernel::Float>(x: &[T], y: &mut [Complex<T>])`
107/// gen_rdft_codelet!(size = 4, kind = R2hc);
108///
109/// // Generates `pub fn hc2r_4_gen<T: crate::kernel::Float>(y: &[Complex<T>], x: &mut [T])`
110/// gen_rdft_codelet!(size = 4, kind = Hc2r);
111/// ```
112///
113/// Supported sizes: 2, 4, 8.
114#[proc_macro]
115pub fn gen_rdft_codelet(input: TokenStream) -> TokenStream {
116    let input2: proc_macro2::TokenStream = input.into();
117    oxifft_codegen_impl::gen_rdft::generate(input2)
118        .unwrap_or_else(|e| e.to_compile_error())
119        .into()
120}