use std::collections::HashMap; use std::sync::Arc; use crate::algorithm::type2and3_butterflies::*; use crate::algorithm::*; use crate::mdct::*; use crate::{ Dct1, Dct5, Dct6And7, Dct8, Dst1, Dst5, Dst6And7, Dst8, TransformType2And3, TransformType4, }; use rustfft::FftPlanner; use crate::DctNum; const DCT2_BUTTERFLIES: [usize; 5] = [2, 3, 4, 8, 16]; /// The DCT planner is used to make new DCT algorithm instances. /// /// RustDCT has several DCT algorithms available for each DCT type; For a given DCT type and problem size, the DctPlanner /// decides which of the available DCT algorithms to use and then initializes them. /// /// ~~~ /// // Perform a DCT Type 4 of size 1234 /// use std::sync::Arc; /// use rustdct::DctPlanner; /// /// let mut planner = DctPlanner::new(); /// let dct4 = planner.plan_dct4(1234); /// /// let mut buffer = vec![0f32; 1234]; /// dct4.process_dct4(&mut buffer); /// /// // The DCT instance returned by the planner is stored behind an `Arc`, so it's cheap to clone /// let dct4_clone = Arc::clone(&dct4); /// ~~~ /// /// If you plan on creating multiple DCT instances, it is recommnded to reuse the same planner for all of them. This /// is because the planner re-uses internal data across DCT instances wherever possible, saving memory and reducing /// setup time. (DCT instances created with one planner will never re-use data and buffers with DCT instances created /// by a different planner) /// /// Each DCT instance owns `Arc`s to its shared internal data, rather than borrowing it from the planner, so it's /// perfectly safe to drop the planner after creating DCT instances. pub struct DctPlanner { fft_planner: FftPlanner, dct1_cache: HashMap>>, dst1_cache: HashMap>>, dct23_cache: HashMap>>, dct4_cache: HashMap>>, dct5_cache: HashMap>>, dst5_cache: HashMap>>, dct6_cache: HashMap>>, dst6_cache: HashMap>>, dct8_cache: HashMap>>, dst8_cache: HashMap>>, mdct_cache: HashMap>>, } impl DctPlanner { pub fn new() -> Self { Self { fft_planner: FftPlanner::new(), dct1_cache: HashMap::new(), dst1_cache: HashMap::new(), dct23_cache: HashMap::new(), dct4_cache: HashMap::new(), dct5_cache: HashMap::new(), dst5_cache: HashMap::new(), dct6_cache: HashMap::new(), dst6_cache: HashMap::new(), dct8_cache: HashMap::new(), dst8_cache: HashMap::new(), mdct_cache: HashMap::new(), } } /// Returns a DCT Type 1 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct1(&mut self, len: usize) -> Arc> { if self.dct1_cache.contains_key(&len) { Arc::clone(self.dct1_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct1(len); self.dct1_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct1(&mut self, len: usize) -> Arc> { //benchmarking shows that below about 10, it's faster to just use the naive DCT1 algorithm if len < 10 { Arc::new(Dct1Naive::new(len)) } else { let fft = self.fft_planner.plan_fft_forward((len - 1) * 2); Arc::new(Dct1ConvertToFft::new(fft)) } } /// Returns a DCT Type 2 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct2(&mut self, len: usize) -> Arc> { if self.dct23_cache.contains_key(&len) { Arc::clone(self.dct23_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct2(len); self.dct23_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct2(&mut self, len: usize) -> Arc> { if DCT2_BUTTERFLIES.contains(&len) { self.plan_dct2_butterfly(len) } else if len.is_power_of_two() && len > 2 { let half_dct = self.plan_dct2(len / 2); let quarter_dct = self.plan_dct2(len / 4); Arc::new(Type2And3SplitRadix::new(half_dct, quarter_dct)) } else { // Benchmarking shows that it's always faster let fft = self.fft_planner.plan_fft_forward(len); Arc::new(Type2And3ConvertToFft::new(fft)) } } fn plan_dct2_butterfly(&mut self, len: usize) -> Arc> { match len { 2 => Arc::new(Type2And3Butterfly2::new()), 3 => Arc::new(Type2And3Butterfly3::new()), 4 => Arc::new(Type2And3Butterfly4::new()), 8 => Arc::new(Type2And3Butterfly8::new()), 16 => Arc::new(Type2And3Butterfly16::new()), _ => panic!("Invalid butterfly size for DCT2: {}", len), } } /// Returns DCT Type 3 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct3(&mut self, len: usize) -> Arc> { self.plan_dct2(len) } /// Returns a DCT Type 4 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct4(&mut self, len: usize) -> Arc> { if self.dct4_cache.contains_key(&len) { Arc::clone(self.dct4_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct4(len); self.dct4_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct4(&mut self, len: usize) -> Arc> { //if we have an even size, we can use the DCT4 Via DCT3 algorithm if len % 2 == 0 { //benchmarking shows that below 6, it's faster to just use the naive DCT4 algorithm if len < 6 { Arc::new(Type4Naive::new(len)) } else { let inner_dct = self.plan_dct3(len / 2); Arc::new(Type4ConvertToType3Even::new(inner_dct)) } } else { //odd size, so we can use the "DCT4 via FFT odd" algorithm //benchmarking shows that below about 7, it's faster to just use the naive DCT4 algorithm if len < 7 { Arc::new(Type4Naive::new(len)) } else { let fft = self.fft_planner.plan_fft_forward(len); Arc::new(Type4ConvertToFftOdd::new(fft)) } } } /// Returns a DCT Type 5 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct5(&mut self, len: usize) -> Arc> { if self.dct5_cache.contains_key(&len) { Arc::clone(self.dct5_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct5(len); self.dct5_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct5(&mut self, len: usize) -> Arc> { Arc::new(Dct5Naive::new(len)) } /// Returns a DCT Type 6 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct6(&mut self, len: usize) -> Arc> { if self.dct6_cache.contains_key(&len) { Arc::clone(self.dct6_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct6(len); self.dct6_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct6(&mut self, len: usize) -> Arc> { Arc::new(Dct6And7Naive::new(len)) } /// Returns DCT Type 7 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct7(&mut self, len: usize) -> Arc> { self.plan_dct6(len) } /// Returns a DCT Type 8 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dct8(&mut self, len: usize) -> Arc> { if self.dct8_cache.contains_key(&len) { Arc::clone(self.dct8_cache.get(&len).unwrap()) } else { let result = self.plan_new_dct8(len); self.dct8_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dct8(&mut self, len: usize) -> Arc> { Arc::new(Dct8Naive::new(len)) } /// Returns a DST Type 1 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst1(&mut self, len: usize) -> Arc> { if self.dst1_cache.contains_key(&len) { Arc::clone(self.dst1_cache.get(&len).unwrap()) } else { let result = self.plan_new_dst1(len); self.dst1_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dst1(&mut self, len: usize) -> Arc> { //benchmarking shows that below about 25, it's faster to just use the naive DCT1 algorithm if len < 25 { Arc::new(Dst1Naive::new(len)) } else { let fft = self.fft_planner.plan_fft_forward((len + 1) * 2); Arc::new(Dst1ConvertToFft::new(fft)) } } /// Returns DST Type 2 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst2(&mut self, len: usize) -> Arc> { self.plan_dct2(len) } /// Returns DST Type 3 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst3(&mut self, len: usize) -> Arc> { self.plan_dct2(len) } /// Returns DST Type 4 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst4(&mut self, len: usize) -> Arc> { self.plan_dct4(len) } /// Returns a DST Type 5 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst5(&mut self, len: usize) -> Arc> { if self.dst5_cache.contains_key(&len) { Arc::clone(self.dst5_cache.get(&len).unwrap()) } else { let result = self.plan_new_dst5(len); self.dst5_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dst5(&mut self, len: usize) -> Arc> { Arc::new(Dst5Naive::new(len)) } /// Returns a DST Type 6 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst6(&mut self, len: usize) -> Arc> { if self.dst6_cache.contains_key(&len) { Arc::clone(self.dst6_cache.get(&len).unwrap()) } else { let result = self.plan_new_dst6(len); self.dst6_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dst6(&mut self, len: usize) -> Arc> { if len < 45 { Arc::new(Dst6And7Naive::new(len)) } else { let fft = self.fft_planner.plan_fft_forward(len * 2 + 1); Arc::new(Dst6And7ConvertToFft::new(fft)) } } /// Returns DST Type 7 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst7(&mut self, len: usize) -> Arc> { self.plan_dst6(len) } /// Returns a DST Type 8 instance which processes signals of size `len`. /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_dst8(&mut self, len: usize) -> Arc> { if self.dst8_cache.contains_key(&len) { Arc::clone(self.dst8_cache.get(&len).unwrap()) } else { let result = self.plan_new_dst8(len); self.dst8_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_dst8(&mut self, len: usize) -> Arc> { Arc::new(Dst8Naive::new(len)) } /// Returns a MDCT instance which processes inputs of size ` len * 2` and produces outputs of size `len`. /// /// `window_fn` is a function that takes a `size` and returns a `Vec` containing `size` window values. /// See the [`window_fn`](mdct/window_fn/index.html) module for provided window functions. /// /// If this is called multiple times, it will attempt to re-use internal data between instances pub fn plan_mdct(&mut self, len: usize, window_fn: F) -> Arc> where F: (FnOnce(usize) -> Vec), { if self.mdct_cache.contains_key(&len) { Arc::clone(self.mdct_cache.get(&len).unwrap()) } else { let result = self.plan_new_mdct(len, window_fn); self.mdct_cache.insert(len, Arc::clone(&result)); result } } fn plan_new_mdct(&mut self, len: usize, window_fn: F) -> Arc> where F: (FnOnce(usize) -> Vec), { //benchmarking shows that using the inner dct4 algorithm is always faster than computing the naive algorithm let inner_dct4 = self.plan_dct4(len); Arc::new(MdctViaDct4::new(inner_dct4, window_fn)) } }