resampler 0.5.1

A small audio resampling library
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
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439

use core::arch::x86_64::*;

use super::{COS_2PI_7, COS_4PI_7, COS_6PI_7, SIN_2PI_7, SIN_4PI_7, SIN_6PI_7};
use crate::fft::{
    Complex32,
    butterflies::ops::{complex_mul_sse4_2, load_neg_imag_mask_sse4_2},
};

/// Performs a single radix-7 Stockham butterfly stage for stride=1 (out-of-place, SSE4.2).
///
/// This is a specialized version for the stride=1 case (first stage) that stores
/// outputs sequentially for better cache utilization.
#[target_feature(enable = "sse4.2")]
pub(super) unsafe fn butterfly_radix7_stride1_sse4_2(
    src: &[Complex32],
    dst: &mut [Complex32],
    stage_twiddles: &[Complex32],
) {
    let samples = src.len();
    let seventh_samples = samples / 7;
    let simd_iters = (seventh_samples >> 1) << 1;

    unsafe {
        let cos_2pi_7 = _mm_set1_ps(COS_2PI_7);
        let sin_2pi_7 = _mm_set1_ps(SIN_2PI_7);
        let cos_4pi_7 = _mm_set1_ps(COS_4PI_7);
        let sin_4pi_7 = _mm_set1_ps(SIN_4PI_7);
        let cos_6pi_7 = _mm_set1_ps(COS_6PI_7);
        let sin_6pi_7 = _mm_set1_ps(SIN_6PI_7);

        let neg_sin_6pi_7 = _mm_set1_ps(-SIN_6PI_7);
        let neg_sin_2pi_7 = _mm_set1_ps(-SIN_2PI_7);
        let neg_sin_4pi_7 = _mm_set1_ps(-SIN_4PI_7);

        let negate_im = load_neg_imag_mask_sse4_2();

        for i in (0..simd_iters).step_by(2) {
            // Load 2 complex numbers from each seventh.
            let z0_ptr = src.as_ptr().add(i) as *const f32;
            let z0 = _mm_loadu_ps(z0_ptr);

            let z1_ptr = src.as_ptr().add(i + seventh_samples) as *const f32;
            let z1 = _mm_loadu_ps(z1_ptr);

            let z2_ptr = src.as_ptr().add(i + seventh_samples * 2) as *const f32;
            let z2 = _mm_loadu_ps(z2_ptr);

            let z3_ptr = src.as_ptr().add(i + seventh_samples * 3) as *const f32;
            let z3 = _mm_loadu_ps(z3_ptr);

            let z4_ptr = src.as_ptr().add(i + seventh_samples * 4) as *const f32;
            let z4 = _mm_loadu_ps(z4_ptr);

            let z5_ptr = src.as_ptr().add(i + seventh_samples * 5) as *const f32;
            let z5 = _mm_loadu_ps(z5_ptr);

            let z6_ptr = src.as_ptr().add(i + seventh_samples * 6) as *const f32;
            let z6 = _mm_loadu_ps(z6_ptr);

            // Identity twiddles: t_k = (1+0i) * z_k = z_k (skip twiddle load and multiply)
            let t1 = z1;
            let t2 = z2;
            let t3 = z3;
            let t4 = z4;
            let t5 = z5;
            let t6 = z6;

            // Radix-7 DFT decomposition
            let sum_all = _mm_add_ps(
                _mm_add_ps(_mm_add_ps(t1, t2), _mm_add_ps(t3, t4)),
                _mm_add_ps(t5, t6),
            );

            // a1 = t1 + t6
            let a1 = _mm_add_ps(t1, t6);
            // a2 = t2 + t5
            let a2 = _mm_add_ps(t2, t5);
            // a3 = t3 + t4
            let a3 = _mm_add_ps(t3, t4);

            // b1_re = t1.im - t6.im, b1_im = t6.re - t1.re
            let t1_swap = _mm_shuffle_ps(t1, t1, 0b10_11_00_01);
            let t6_swap = _mm_shuffle_ps(t6, t6, 0b10_11_00_01);
            let b1_temp = _mm_sub_ps(t1_swap, t6_swap);
            let b1 = _mm_xor_ps(b1_temp, negate_im);

            // b2_re = t2.im - t5.im, b2_im = t5.re - t2.re
            let t2_swap = _mm_shuffle_ps(t2, t2, 0b10_11_00_01);
            let t5_swap = _mm_shuffle_ps(t5, t5, 0b10_11_00_01);
            let b2_temp = _mm_sub_ps(t2_swap, t5_swap);
            let b2 = _mm_xor_ps(b2_temp, negate_im);

            // b3_re = t3.im - t4.im, b3_im = t4.re - t3.re
            let t3_swap = _mm_shuffle_ps(t3, t3, 0b10_11_00_01);
            let t4_swap = _mm_shuffle_ps(t4, t4, 0b10_11_00_01);
            let b3_temp = _mm_sub_ps(t3_swap, t4_swap);
            let b3 = _mm_xor_ps(b3_temp, negate_im);

            // out0 = z0 + sum_all
            let out0 = _mm_add_ps(z0, sum_all);

            // Compute outputs 1-6 using unrolled loop from scalar version

            // out1: cos1=COS_2PI_7, sin1=SIN_2PI_7, cos2=COS_4PI_7, sin2=SIN_4PI_7, cos3=COS_6PI_7, sin3=SIN_6PI_7
            let c1 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_2pi_7, a1), _mm_mul_ps(cos_4pi_7, a2)),
                    _mm_mul_ps(cos_6pi_7, a3),
                ),
            );
            let d1 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_2pi_7, b1), _mm_mul_ps(sin_4pi_7, b2)),
                _mm_mul_ps(sin_6pi_7, b3),
            );
            let out1 = _mm_add_ps(c1, d1);

            // out2: cos1=COS_4PI_7, sin1=SIN_4PI_7, cos2=COS_6PI_7, sin2=-SIN_6PI_7, cos3=COS_2PI_7, sin3=-SIN_2PI_7
            let c2 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_4pi_7, a1), _mm_mul_ps(cos_6pi_7, a2)),
                    _mm_mul_ps(cos_2pi_7, a3),
                ),
            );
            let d2 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_4pi_7, b1), _mm_mul_ps(neg_sin_6pi_7, b2)),
                _mm_mul_ps(neg_sin_2pi_7, b3),
            );
            let out2 = _mm_add_ps(c2, d2);

            // out3: cos1=COS_6PI_7, sin1=SIN_6PI_7, cos2=COS_2PI_7, sin2=-SIN_2PI_7, cos3=COS_4PI_7, sin3=SIN_4PI_7
            let c3 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_6pi_7, a1), _mm_mul_ps(cos_2pi_7, a2)),
                    _mm_mul_ps(cos_4pi_7, a3),
                ),
            );
            let d3 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_6pi_7, b1), _mm_mul_ps(neg_sin_2pi_7, b2)),
                _mm_mul_ps(sin_4pi_7, b3),
            );
            let out3 = _mm_add_ps(c3, d3);

            // out4: cos1=COS_6PI_7, sin1=-SIN_6PI_7, cos2=COS_2PI_7, sin2=SIN_2PI_7, cos3=COS_4PI_7, sin3=-SIN_4PI_7
            let c4 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_6pi_7, a1), _mm_mul_ps(cos_2pi_7, a2)),
                    _mm_mul_ps(cos_4pi_7, a3),
                ),
            );
            let d4 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_6pi_7, b1), _mm_mul_ps(sin_2pi_7, b2)),
                _mm_mul_ps(neg_sin_4pi_7, b3),
            );
            let out4 = _mm_add_ps(c4, d4);

            // out5: cos1=COS_4PI_7, sin1=-SIN_4PI_7, cos2=COS_6PI_7, sin2=SIN_6PI_7, cos3=COS_2PI_7, sin3=SIN_2PI_7
            let c5 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_4pi_7, a1), _mm_mul_ps(cos_6pi_7, a2)),
                    _mm_mul_ps(cos_2pi_7, a3),
                ),
            );
            let d5 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_4pi_7, b1), _mm_mul_ps(sin_6pi_7, b2)),
                _mm_mul_ps(sin_2pi_7, b3),
            );
            let out5 = _mm_add_ps(c5, d5);

            // out6: cos1=COS_2PI_7, sin1=-SIN_2PI_7, cos2=COS_4PI_7, sin2=-SIN_4PI_7, cos3=COS_6PI_7, sin3=-SIN_6PI_7
            let c6 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_2pi_7, a1), _mm_mul_ps(cos_4pi_7, a2)),
                    _mm_mul_ps(cos_6pi_7, a3),
                ),
            );
            let d6 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_2pi_7, b1), _mm_mul_ps(neg_sin_4pi_7, b2)),
                _mm_mul_ps(neg_sin_6pi_7, b3),
            );
            let out6 = _mm_add_ps(c6, d6);

            // Sequential 128-bit stores for stride=1 (matching radix-2/radix-4 pattern).
            // Using 7x 128-bit stores instead of 14x 64-bit stores
            let j = 14 * i; // In units of f32 (2 f32 per complex)
            let dst_ptr = dst.as_mut_ptr() as *mut f32;

            let out0_pd = _mm_castps_pd(out0);
            let out1_pd = _mm_castps_pd(out1);
            let out2_pd = _mm_castps_pd(out2);
            let out3_pd = _mm_castps_pd(out3);
            let out4_pd = _mm_castps_pd(out4);
            let out5_pd = _mm_castps_pd(out5);
            let out6_pd = _mm_castps_pd(out6);

            // Interleave outputs for sequential 128-bit stores
            let out01_lo = _mm_castpd_ps(_mm_unpacklo_pd(out0_pd, out1_pd)); // [out0[0], out1[0]]
            let out23_lo = _mm_castpd_ps(_mm_unpacklo_pd(out2_pd, out3_pd)); // [out2[0], out3[0]]
            let out45_lo = _mm_castpd_ps(_mm_unpacklo_pd(out4_pd, out5_pd)); // [out4[0], out5[0]]
            let out60_cross = _mm_castpd_ps(_mm_shuffle_pd(out6_pd, out0_pd, 0b10)); // [out6[0], out0[1]]
            let out12_hi = _mm_castpd_ps(_mm_unpackhi_pd(out1_pd, out2_pd)); // [out1[1], out2[1]]
            let out34_hi = _mm_castpd_ps(_mm_unpackhi_pd(out3_pd, out4_pd)); // [out3[1], out4[1]]
            let out56_hi = _mm_castpd_ps(_mm_unpackhi_pd(out5_pd, out6_pd)); // [out5[1], out6[1]]

            _mm_storeu_ps(dst_ptr.add(j), out01_lo);
            _mm_storeu_ps(dst_ptr.add(j + 4), out23_lo);
            _mm_storeu_ps(dst_ptr.add(j + 8), out45_lo);
            _mm_storeu_ps(dst_ptr.add(j + 12), out60_cross);
            _mm_storeu_ps(dst_ptr.add(j + 16), out12_hi);
            _mm_storeu_ps(dst_ptr.add(j + 20), out34_hi);
            _mm_storeu_ps(dst_ptr.add(j + 24), out56_hi);
        }
    }

    super::butterfly_radix7_scalar::<2>(src, dst, stage_twiddles, 1, simd_iters);
}

/// Performs a single radix-7 Stockham butterfly stage for p>1 (out-of-place, SSE4.2).
///
/// This is the generic version for p>1 cases.
#[target_feature(enable = "sse4.2")]
pub(super) unsafe fn butterfly_radix7_generic_sse4_2(
    src: &[Complex32],
    dst: &mut [Complex32],
    stage_twiddles: &[Complex32],
    stride: usize,
) {
    // We convince the compiler here that stride can't be 0 to optimize better.
    if stride == 0 {
        return;
    }

    let samples = src.len();
    let seventh_samples = samples / 7;
    let simd_iters = (seventh_samples >> 1) << 1;

    unsafe {
        let cos_2pi_7 = _mm_set1_ps(COS_2PI_7);
        let sin_2pi_7 = _mm_set1_ps(SIN_2PI_7);
        let cos_4pi_7 = _mm_set1_ps(COS_4PI_7);
        let sin_4pi_7 = _mm_set1_ps(SIN_4PI_7);
        let cos_6pi_7 = _mm_set1_ps(COS_6PI_7);
        let sin_6pi_7 = _mm_set1_ps(SIN_6PI_7);

        let neg_sin_6pi_7 = _mm_set1_ps(-SIN_6PI_7);
        let neg_sin_2pi_7 = _mm_set1_ps(-SIN_2PI_7);
        let neg_sin_4pi_7 = _mm_set1_ps(-SIN_4PI_7);

        let negate_im = load_neg_imag_mask_sse4_2();

        for i in (0..simd_iters).step_by(2) {
            let k = i % stride;
            let k0 = k;
            let k1 = k + 1 - ((k + 1 >= stride) as usize) * stride;

            // Load 2 complex numbers from each seventh.
            let z0_ptr = src.as_ptr().add(i) as *const f32;
            let z0 = _mm_loadu_ps(z0_ptr);

            let z1_ptr = src.as_ptr().add(i + seventh_samples) as *const f32;
            let z1 = _mm_loadu_ps(z1_ptr);

            let z2_ptr = src.as_ptr().add(i + seventh_samples * 2) as *const f32;
            let z2 = _mm_loadu_ps(z2_ptr);

            let z3_ptr = src.as_ptr().add(i + seventh_samples * 3) as *const f32;
            let z3 = _mm_loadu_ps(z3_ptr);

            let z4_ptr = src.as_ptr().add(i + seventh_samples * 4) as *const f32;
            let z4 = _mm_loadu_ps(z4_ptr);

            let z5_ptr = src.as_ptr().add(i + seventh_samples * 5) as *const f32;
            let z5 = _mm_loadu_ps(z5_ptr);

            let z6_ptr = src.as_ptr().add(i + seventh_samples * 6) as *const f32;
            let z6 = _mm_loadu_ps(z6_ptr);

            // Load prepackaged twiddles directly (no shuffle needed).
            let tw_ptr = stage_twiddles.as_ptr().add(i * 6) as *const f32;
            let w1 = _mm_loadu_ps(tw_ptr); // w1[i], w1[i+1]
            let w2 = _mm_loadu_ps(tw_ptr.add(4)); // w2[i], w2[i+1]
            let w3 = _mm_loadu_ps(tw_ptr.add(8)); // w3[i], w3[i+1]
            let w4 = _mm_loadu_ps(tw_ptr.add(12)); // w4[i], w4[i+1]
            let w5 = _mm_loadu_ps(tw_ptr.add(16)); // w5[i], w5[i+1]
            let w6 = _mm_loadu_ps(tw_ptr.add(20)); // w6[i], w6[i+1]

            // Complex multiply.
            let t1 = complex_mul_sse4_2(w1, z1);
            let t2 = complex_mul_sse4_2(w2, z2);
            let t3 = complex_mul_sse4_2(w3, z3);
            let t4 = complex_mul_sse4_2(w4, z4);
            let t5 = complex_mul_sse4_2(w5, z5);
            let t6 = complex_mul_sse4_2(w6, z6);

            // Radix-7 DFT decomposition
            let sum_all = _mm_add_ps(
                _mm_add_ps(_mm_add_ps(t1, t2), _mm_add_ps(t3, t4)),
                _mm_add_ps(t5, t6),
            );

            let a1 = _mm_add_ps(t1, t6);
            let a2 = _mm_add_ps(t2, t5);
            let a3 = _mm_add_ps(t3, t4);

            let t1_swap = _mm_shuffle_ps(t1, t1, 0b10_11_00_01);
            let t6_swap = _mm_shuffle_ps(t6, t6, 0b10_11_00_01);
            let b1_temp = _mm_sub_ps(t1_swap, t6_swap);
            let b1 = _mm_xor_ps(b1_temp, negate_im);

            let t2_swap = _mm_shuffle_ps(t2, t2, 0b10_11_00_01);
            let t5_swap = _mm_shuffle_ps(t5, t5, 0b10_11_00_01);
            let b2_temp = _mm_sub_ps(t2_swap, t5_swap);
            let b2 = _mm_xor_ps(b2_temp, negate_im);

            let t3_swap = _mm_shuffle_ps(t3, t3, 0b10_11_00_01);
            let t4_swap = _mm_shuffle_ps(t4, t4, 0b10_11_00_01);
            let b3_temp = _mm_sub_ps(t3_swap, t4_swap);
            let b3 = _mm_xor_ps(b3_temp, negate_im);

            let out0 = _mm_add_ps(z0, sum_all);

            let c1 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_2pi_7, a1), _mm_mul_ps(cos_4pi_7, a2)),
                    _mm_mul_ps(cos_6pi_7, a3),
                ),
            );
            let d1 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_2pi_7, b1), _mm_mul_ps(sin_4pi_7, b2)),
                _mm_mul_ps(sin_6pi_7, b3),
            );
            let out1 = _mm_add_ps(c1, d1);

            let c2 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_4pi_7, a1), _mm_mul_ps(cos_6pi_7, a2)),
                    _mm_mul_ps(cos_2pi_7, a3),
                ),
            );
            let d2 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_4pi_7, b1), _mm_mul_ps(neg_sin_6pi_7, b2)),
                _mm_mul_ps(neg_sin_2pi_7, b3),
            );
            let out2 = _mm_add_ps(c2, d2);

            let c3 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_6pi_7, a1), _mm_mul_ps(cos_2pi_7, a2)),
                    _mm_mul_ps(cos_4pi_7, a3),
                ),
            );
            let d3 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(sin_6pi_7, b1), _mm_mul_ps(neg_sin_2pi_7, b2)),
                _mm_mul_ps(sin_4pi_7, b3),
            );
            let out3 = _mm_add_ps(c3, d3);

            let c4 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_6pi_7, a1), _mm_mul_ps(cos_2pi_7, a2)),
                    _mm_mul_ps(cos_4pi_7, a3),
                ),
            );
            let d4 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_6pi_7, b1), _mm_mul_ps(sin_2pi_7, b2)),
                _mm_mul_ps(neg_sin_4pi_7, b3),
            );
            let out4 = _mm_add_ps(c4, d4);

            let c5 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_4pi_7, a1), _mm_mul_ps(cos_6pi_7, a2)),
                    _mm_mul_ps(cos_2pi_7, a3),
                ),
            );
            let d5 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_4pi_7, b1), _mm_mul_ps(sin_6pi_7, b2)),
                _mm_mul_ps(sin_2pi_7, b3),
            );
            let out5 = _mm_add_ps(c5, d5);

            let c6 = _mm_add_ps(
                z0,
                _mm_add_ps(
                    _mm_add_ps(_mm_mul_ps(cos_2pi_7, a1), _mm_mul_ps(cos_4pi_7, a2)),
                    _mm_mul_ps(cos_6pi_7, a3),
                ),
            );
            let d6 = _mm_add_ps(
                _mm_add_ps(_mm_mul_ps(neg_sin_2pi_7, b1), _mm_mul_ps(neg_sin_4pi_7, b2)),
                _mm_mul_ps(neg_sin_6pi_7, b3),
            );
            let out6 = _mm_add_ps(c6, d6);

            // Calculate output indices.
            let j0 = 7 * i - 6 * k0;
            let j1 = 7 * (i + 1) - 6 * k1;

            // Direct SIMD stores.
            let out0_pd = _mm_castps_pd(out0);
            let out1_pd = _mm_castps_pd(out1);
            let out2_pd = _mm_castps_pd(out2);
            let out3_pd = _mm_castps_pd(out3);
            let out4_pd = _mm_castps_pd(out4);
            let out5_pd = _mm_castps_pd(out5);
            let out6_pd = _mm_castps_pd(out6);

            let dst_ptr = dst.as_mut_ptr() as *mut f64;

            _mm_storel_pd(dst_ptr.add(j0), out0_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride), out1_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride * 2), out2_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride * 3), out3_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride * 4), out4_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride * 5), out5_pd);
            _mm_storel_pd(dst_ptr.add(j0 + stride * 6), out6_pd);

            _mm_storeh_pd(dst_ptr.add(j1), out0_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride), out1_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride * 2), out2_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride * 3), out3_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride * 4), out4_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride * 5), out5_pd);
            _mm_storeh_pd(dst_ptr.add(j1 + stride * 6), out6_pd);
        }
    }

    super::butterfly_radix7_scalar::<2>(src, dst, stage_twiddles, stride, simd_iters);
}