pxdct 0.3.2

Discrete cosine transform
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349

/*
 * // Copyright (c) Radzivon Bartoshyk 1/2026. All rights reserved.
 * //
 * // Redistribution and use in source and binary forms, with or without modification,
 * // are permitted provided that the following conditions are met:
 * //
 * // 1.  Redistributions of source code must retain the above copyright notice, this
 * // list of conditions and the following disclaimer.
 * //
 * // 2.  Redistributions in binary form must reproduce the above copyright notice,
 * // this list of conditions and the following disclaimer in the documentation
 * // and/or other materials provided with the distribution.
 * //
 * // 3.  Neither the name of the copyright holder nor the names of its
 * // contributors may be used to endorse or promote products derived from
 * // this software without specific prior written permission.
 * //
 * // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
 * // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
use crate::bidirectional::BidirectionalStore;
use crate::dct4::Dct4MixedRadix11Sample;
use crate::mla::fmla;
use crate::neon::dct4::mixed_radix3f::dct4_radix_n_rotation_twiddles_neon;
use crate::neon::util::{NeonStoreF, boring_neon_mixed_radix};
use crate::util::{DctSample, try_vec, validate_scratch};
use crate::{PxdctError, PxdctExecutor};
use num_traits::One;
use std::sync::Arc;

pub(crate) struct NeonDct4MixedRadix11f {
    inner_dct4: Arc<dyn PxdctExecutor<f32> + Send + Sync>,
    inner_dct_scratch_size: usize,
    rotation_twiddles: Vec<NeonStoreF>,
    execution_length: usize,
    q_modules: usize,
    s: usize,
}

impl NeonDct4MixedRadix11f {
    pub(crate) fn new(
        len: usize,
        dct4: Arc<dyn PxdctExecutor<f32> + Send + Sync>,
    ) -> Result<Self, PxdctError> {
        assert_eq!(
            dct4.length(),
            len / 11,
            "DCT-IV Mixed-Radix-11 length DCTs must be one eleventh of DCT-IV"
        );

        let inner_dct4_scratch_size = dct4.scratch_size();

        Ok(Self {
            inner_dct4: dct4,
            inner_dct_scratch_size: inner_dct4_scratch_size,
            execution_length: len,
            rotation_twiddles: dct4_radix_n_rotation_twiddles_neon(11, len / 11, len),
            q_modules: len / 11,
            s: 2 * len / 11,
        })
    }
}

impl NeonDct4MixedRadix11f {
    #[inline(always)]
    fn exec_block<S: BidirectionalStore<f32>, const N: usize>(
        &self,
        data: &mut S,
        a_buffer: &[f32],
        s_buffer: &[f32],
        c_buffer: &[f32],
        uk: usize,
        k: usize,
    ) {
        let c_v0 = NeonStoreF::load_n::<N>(unsafe { c_buffer.get_unchecked(k..) });
        let s_v0 =
            NeonStoreF::load_n::<N>(unsafe { s_buffer.get_unchecked(self.q_modules - N - k..) })
                .reverse_n::<N>();
        let a_v0 = NeonStoreF::load_n::<N>(unsafe { a_buffer.get_unchecked(k..) });

        let c_v1 = NeonStoreF::load_n::<N>(unsafe { c_buffer.get_unchecked(self.q_modules + k..) });
        let s_v1 = NeonStoreF::load_n::<N>(unsafe {
            s_buffer.get_unchecked(self.q_modules * 2 - N - k..)
        })
        .reverse_n::<N>();

        let c_v2 =
            NeonStoreF::load_n::<N>(unsafe { c_buffer.get_unchecked(self.q_modules * 2 + k..) });
        let s_v2 = NeonStoreF::load_n::<N>(unsafe {
            s_buffer.get_unchecked(self.q_modules * 3 - N - k..)
        })
        .reverse_n::<N>();

        let c_v3 =
            NeonStoreF::load_n::<N>(unsafe { c_buffer.get_unchecked(self.q_modules * 3 + k..) });
        let s_v3 = NeonStoreF::load_n::<N>(unsafe {
            s_buffer.get_unchecked(self.q_modules * 4 - N - k..)
        })
        .reverse_n::<N>();

        let c_v4 =
            NeonStoreF::load_n::<N>(unsafe { c_buffer.get_unchecked(self.q_modules * 4 + k..) });
        let s_v4 = NeonStoreF::load_n::<N>(unsafe {
            s_buffer.get_unchecked(self.q_modules * 5 - N - k..)
        })
        .reverse_n::<N>();

        let twiddle0_re = unsafe { *self.rotation_twiddles.get_unchecked(uk) };
        let twiddle0_im = unsafe { *self.rotation_twiddles.get_unchecked(uk + 1) };
        let twiddle1_re = unsafe { *self.rotation_twiddles.get_unchecked(uk + 2) };
        let twiddle1_im = unsafe { *self.rotation_twiddles.get_unchecked(uk + 3) };
        let twiddle2_re = unsafe { *self.rotation_twiddles.get_unchecked(uk + 4) };
        let twiddle2_im = unsafe { *self.rotation_twiddles.get_unchecked(uk + 5) };
        let twiddle3_re = unsafe { *self.rotation_twiddles.get_unchecked(uk + 6) };
        let twiddle3_im = unsafe { *self.rotation_twiddles.get_unchecked(uk + 7) };
        let twiddle4_re = unsafe { *self.rotation_twiddles.get_unchecked(uk + 8) };
        let twiddle4_im = unsafe { *self.rotation_twiddles.get_unchecked(uk + 9) };

        let iq0 = fmla(c_v0, twiddle0_re, s_v0 * twiddle0_im);
        let siq0 = fmla(c_v0, twiddle0_im, -s_v0 * twiddle0_re);
        let mut u0 = iq0;
        let mut u1 = u0;
        let mut v0 = siq0;

        u1 *= f32::D4_R11_ROT_TWIDDLE_2;
        v0 *= f32::D4_R11_ROT_TWIDDLE_3;

        let iq1 = fmla(c_v1, twiddle1_re, s_v1 * twiddle1_im);
        let siq1 = fmla(c_v1, twiddle1_im, -s_v1 * twiddle1_re);

        u1 = NeonStoreF::mul_f32_add(iq1, f32::D4_R11_ROT_TWIDDLE_1, u1);
        v0 = NeonStoreF::mul_f32_add(siq1, f32::D4_R11_ROT_TWIDDLE_0, v0);

        let iq2 = fmla(c_v2, twiddle2_re, s_v2 * twiddle2_im);
        let siq2 = fmla(c_v2, twiddle2_im, -s_v2 * twiddle2_re);

        u1 = NeonStoreF::mul_f32_add(iq2, f32::D4_R11_ROT_TWIDDLE_8, u1);
        v0 = NeonStoreF::mul_f32_add(siq2, f32::D4_R11_ROT_TWIDDLE_9, v0);

        let iq3 = fmla(c_v3, twiddle3_re, s_v3 * twiddle3_im);
        let siq3 = fmla(c_v3, twiddle3_im, -s_v3 * twiddle3_re);

        u1 = NeonStoreF::mul_f32_add(iq3, -f32::D4_R11_ROT_TWIDDLE_7, u1);
        v0 = NeonStoreF::mul_f32_add(siq3, f32::D4_R11_ROT_TWIDDLE_6, v0);

        let iq4 = fmla(c_v4, twiddle4_re, s_v4 * twiddle4_im);
        let siq4 = fmla(c_v4, twiddle4_im, -s_v4 * twiddle4_re);

        u1 = NeonStoreF::mul_f32_add(iq4, -f32::D4_R11_ROT_TWIDDLE_4, u1);
        v0 = NeonStoreF::mul_f32_add(siq4, f32::D4_R11_ROT_TWIDDLE_5, v0);

        u0 += iq1 + iq2 + iq3 + iq4 + a_v0;
        u1 = u1 - a_v0;

        let uc0 = u1 - v0;
        let uc1 = u1 + v0;

        u0.write_n::<N>(data.slice_from_mut(k..));
        uc1.write_n::<N>(data.slice_from_mut(self.s + k..));
        uc0.reverse_n::<N>()
            .write_n::<N>(data.slice_from_mut(self.s - N - k..));

        let mut u2 = iq0;
        let mut v2 = siq0;
        u2 *= f32::D4_R11_ROT_TWIDDLE_4;
        v2 *= f32::D4_R11_ROT_TWIDDLE_5;
        u2 = NeonStoreF::mul_f32_add(iq1, -f32::D4_R11_ROT_TWIDDLE_8, u2);
        v2 = NeonStoreF::mul_f32_add(siq1, f32::D4_R11_ROT_TWIDDLE_9, v2);
        u2 = NeonStoreF::mul_f32_add(iq2, -f32::D4_R11_ROT_TWIDDLE_2, u2);
        v2 = NeonStoreF::mul_f32_add(siq2, f32::D4_R11_ROT_TWIDDLE_3, v2);
        u2 = NeonStoreF::mul_f32_add(iq3, -f32::D4_R11_ROT_TWIDDLE_1, u2);
        v2 = NeonStoreF::mul_f32_add(siq3, -f32::D4_R11_ROT_TWIDDLE_0, v2);
        u2 = NeonStoreF::mul_f32_add(iq4, f32::D4_R11_ROT_TWIDDLE_7, u2);
        v2 = NeonStoreF::mul_f32_add(siq4, -f32::D4_R11_ROT_TWIDDLE_6, v2);
        u2 += a_v0;
        let uc2 = u2 - v2;
        let uc3 = u2 + v2;

        uc2.reverse_n::<N>()
            .write_n::<N>(data.slice_from_mut(2 * self.s - N - k..));
        uc3.write_n::<N>(data.slice_from_mut(2 * self.s + k..));

        let mut u3 = iq0;
        let mut v3 = siq0;
        u3 *= f32::D4_R11_ROT_TWIDDLE_1;
        v3 *= f32::D4_R11_ROT_TWIDDLE_0;
        u3 = NeonStoreF::mul_f32_add(iq1, -f32::D4_R11_ROT_TWIDDLE_4, u3);
        v3 = NeonStoreF::mul_f32_add(siq1, f32::D4_R11_ROT_TWIDDLE_5, v3);
        u3 = NeonStoreF::mul_f32_add(iq2, -f32::D4_R11_ROT_TWIDDLE_7, u3);
        v3 = NeonStoreF::mul_f32_add(siq2, -f32::D4_R11_ROT_TWIDDLE_6, v3);
        u3 = NeonStoreF::mul_f32_add(iq3, f32::D4_R11_ROT_TWIDDLE_2, u3);
        v3 = NeonStoreF::mul_f32_add(siq3, -f32::D4_R11_ROT_TWIDDLE_3, v3);
        u3 = NeonStoreF::mul_f32_add(iq4, f32::D4_R11_ROT_TWIDDLE_8, u3);
        v3 = NeonStoreF::mul_f32_add(siq4, f32::D4_R11_ROT_TWIDDLE_9, v3);
        u3 = u3 - a_v0;
        let uc4 = u3 - v3;
        let uc5 = u3 + v3;

        uc4.reverse_n::<N>()
            .write_n::<N>(data.slice_from_mut(3 * self.s - N - k..));
        uc5.write_n::<N>(data.slice_from_mut(3 * self.s + k..));

        let mut u4 = iq0;
        let mut v4 = siq0;
        u4 *= f32::D4_R11_ROT_TWIDDLE_7;
        v4 *= f32::D4_R11_ROT_TWIDDLE_6;
        u4 = NeonStoreF::mul_f32_add(iq1, -f32::D4_R11_ROT_TWIDDLE_2, u4);
        v4 = NeonStoreF::mul_f32_add(siq1, -f32::D4_R11_ROT_TWIDDLE_3, v4);
        u4 = NeonStoreF::mul_f32_add(iq2, f32::D4_R11_ROT_TWIDDLE_4, u4);
        v4 = NeonStoreF::mul_f32_add(siq2, -f32::D4_R11_ROT_TWIDDLE_5, v4);
        u4 = NeonStoreF::mul_f32_add(iq3, -f32::D4_R11_ROT_TWIDDLE_8, u4);
        v4 = NeonStoreF::mul_f32_add(siq3, f32::D4_R11_ROT_TWIDDLE_9, v4);
        u4 = NeonStoreF::mul_f32_add(iq4, -f32::D4_R11_ROT_TWIDDLE_1, u4);
        v4 = NeonStoreF::mul_f32_add(siq4, -f32::D4_R11_ROT_TWIDDLE_0, v4);
        u4 += a_v0;
        let uc6 = u4 - v4;
        let uc7 = u4 + v4;

        uc6.reverse_n::<N>()
            .write_n::<N>(data.slice_from_mut(4 * self.s - N - k..));
        uc7.write_n::<N>(data.slice_from_mut(4 * self.s + k..));

        let mut u5 = iq0;
        let mut v5 = siq0;
        u5 *= f32::D4_R11_ROT_TWIDDLE_8;
        v5 *= f32::D4_R11_ROT_TWIDDLE_9;
        u5 = NeonStoreF::mul_f32_add(iq1, -f32::D4_R11_ROT_TWIDDLE_7, u5);
        v5 = NeonStoreF::mul_f32_add(siq1, -f32::D4_R11_ROT_TWIDDLE_6, v5);
        u5 = NeonStoreF::mul_f32_add(iq2, f32::D4_R11_ROT_TWIDDLE_1, u5);
        v5 = NeonStoreF::mul_f32_add(siq2, f32::D4_R11_ROT_TWIDDLE_0, v5);
        u5 = NeonStoreF::mul_f32_add(iq3, -f32::D4_R11_ROT_TWIDDLE_4, u5);
        v5 = NeonStoreF::mul_f32_add(siq3, -f32::D4_R11_ROT_TWIDDLE_5, v5);
        u5 = NeonStoreF::mul_f32_add(iq4, f32::D4_R11_ROT_TWIDDLE_2, u5);
        v5 = NeonStoreF::mul_f32_add(siq4, f32::D4_R11_ROT_TWIDDLE_3, v5);
        u5 = u5 - a_v0;
        let uc8 = u5 - v5;
        let uc9 = u5 + v5;

        uc8.reverse_n::<N>()
            .write_n::<N>(data.slice_from_mut(5 * self.s - N - k..));
        uc9.write_n::<N>(data.slice_from_mut(5 * self.s + k..));
    }

    #[inline(always)]
    fn execute_with_store<S: BidirectionalStore<f32>>(
        &self,
        data: &mut S,
        scratch: &mut [f32],
    ) -> Result<(), PxdctError> {
        let (scratch, inner_scratch) = scratch.split_at_mut(self.execution_length);
        let q_modules = self.execution_length / 11;
        let (a_buffer, c_s_buffer) = scratch.split_at_mut(q_modules);
        let (c_buffer, s_buffer) = c_s_buffer.split_at_mut(q_modules * 5);

        // Step 1: Decompose input into A (center), C (even-symmetric), S (odd-symmetric) buffers
        for (n, dst) in a_buffer.iter_mut().enumerate() {
            *dst = data[n * 11 + 5];
        }

        // Extract and combine symmetric pairs with sign alternation for S buffer
        for (m, (c_buffer, s_buffer)) in c_buffer
            .chunks_exact_mut(q_modules)
            .zip(s_buffer.chunks_exact_mut(q_modules))
            .enumerate()
        {
            let mut sign = f32::one();
            for (n, (c_dst, s_dst)) in c_buffer.iter_mut().zip(s_buffer.iter_mut()).enumerate() {
                let u0 = data[11 * n + m];
                let u1 = data[11 * n + 11 - m - 1];

                *c_dst = u0 + u1;
                *s_dst = (u0 - u1).mulsign(sign);

                sign = -sign;
            }
        }

        // Step 2: Apply DCT-IV to all buffers (A, C₀, C₁, S₀, S₁)
        self.inner_dct4
            .execute_with_scratch(scratch, inner_scratch)?;

        let (a_buffer, c_s_buffer) = scratch.split_at_mut(q_modules);
        let (c_buffer, s_buffer) = c_s_buffer.split_at_mut(q_modules * 5);

        // Step 4: Handle k≥0 cases with rotation twiddles
        let mut k = 0usize;
        let mut uk = 0usize;
        while k + 4 <= q_modules {
            self.exec_block::<S, 4>(data, a_buffer, s_buffer, c_buffer, uk, k);

            k += 4;
            uk += 10;
        }

        let rem = q_modules - k;
        if rem == 3 {
            self.exec_block::<S, 3>(data, a_buffer, s_buffer, c_buffer, uk, k);
        } else if rem == 2 {
            self.exec_block::<S, 2>(data, a_buffer, s_buffer, c_buffer, uk, k);
        } else if rem == 1 {
            self.exec_block::<S, 1>(data, a_buffer, s_buffer, c_buffer, uk, k);
        }
        Ok(())
    }
}

boring_neon_mixed_radix!(NeonDct4MixedRadix11f, f32);

#[cfg(test)]
mod tests {
    use super::*;
    use crate::dct4::Dct4Butterfly3;
    use crate::tests::naive_dct4_f32;
    use rand::RngExt;

    #[test]
    fn test_split_dct4() {
        let mut input = vec![0.; 33];
        for z in input.iter_mut() {
            *z = rand::rng().random_range(1.0..2.0);
        }
        let reference_input = input.clone();
        let reference_input = naive_dct4_f32(&reference_input);
        let bf =
            NeonDct4MixedRadix11f::new(input.len(), Arc::new(Dct4Butterfly3::default())).unwrap();
        bf.execute(&mut input).unwrap();
        input
            .iter()
            .zip(reference_input.iter())
            .enumerate()
            .for_each(|(i, (&src, &r0))| {
                if (src - r0).abs() > 1e-1 {
                    println!(
                        "Difference must be < {}, but it was {}, at position {i}",
                        1e-1,
                        (src - r0).abs()
                    )
                }
            });
    }
}