mousiki 0.2.0

Pure Rust Opus codec.
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

use core::arch::arm::{
    vaddq_f32, vandq_u32, vcvtq_s32_f32, vdupq_n_f32, vdupq_n_u32, vget_high_f32, vget_lane_f32,
    vget_low_f32, vld1q_f32, vmax_f32, vmaxq_f32, vmin_f32, vminq_f32, vmulq_n_f32, vorrq_u32,
    vqmovn_s32, vreinterpretq_f32_u32, vreinterpretq_u32_f32, vst1_s16, vst1q_f32,
};

use crate::celt::float_cast;

#[inline]
unsafe fn vroundf(x: core::arch::arm::float32x4_t) -> core::arch::arm::int32x4_t {
    // Mirror opus-c arm/mathops_arm.h non-aarch64 fallback:
    // vcvtq_s32_f32(x + copysign(0.5, x))
    let sign = vandq_u32(vreinterpretq_u32_f32(x), vdupq_n_u32(0x8000_0000));
    let bias = vdupq_n_u32(0x3f00_0000);
    vcvtq_s32_f32(vaddq_f32(x, vreinterpretq_f32_u32(vorrq_u32(bias, sign))))
}

#[inline]
unsafe fn vminvf(a: core::arch::arm::float32x4_t) -> f32 {
    let xy = vmin_f32(vget_low_f32(a), vget_high_f32(a));
    let x = vget_lane_f32(xy, 0);
    let y = vget_lane_f32(xy, 1);
    if x < y { x } else { y }
}

#[inline]
unsafe fn vmaxvf(a: core::arch::arm::float32x4_t) -> f32 {
    let xy = vmax_f32(vget_low_f32(a), vget_high_f32(a));
    let x = vget_lane_f32(xy, 0);
    let y = vget_lane_f32(xy, 1);
    if x > y { x } else { y }
}

/// ARM NEON specialization for `opus_limit2_checkwithin1`.
pub(super) fn opus_limit2_checkwithin1(samples: &mut [f32]) -> bool {
    if samples.is_empty() {
        return true;
    }

    const HARDCLIP_MIN: f32 = -2.0;
    const HARDCLIP_MAX: f32 = 2.0;
    const BLOCK_SIZE: usize = 16;
    let blocked_size = samples.len() / BLOCK_SIZE * BLOCK_SIZE;

    let mut exceeding1 = false;
    let mut next_index = 0usize;

    if blocked_size > 0 {
        let mut min_all_0 = unsafe { vdupq_n_f32(0.0) };
        let mut min_all_1 = unsafe { vdupq_n_f32(0.0) };
        let mut max_all_0 = unsafe { vdupq_n_f32(0.0) };
        let mut max_all_1 = unsafe { vdupq_n_f32(0.0) };

        unsafe {
            let samples_ptr = samples.as_ptr();
            for i in (0..blocked_size).step_by(BLOCK_SIZE) {
                let orig_a = vld1q_f32(samples_ptr.add(i));
                let orig_b = vld1q_f32(samples_ptr.add(i + 4));
                let orig_c = vld1q_f32(samples_ptr.add(i + 8));
                let orig_d = vld1q_f32(samples_ptr.add(i + 12));

                max_all_0 = vmaxq_f32(max_all_0, vmaxq_f32(orig_a, orig_b));
                max_all_1 = vmaxq_f32(max_all_1, vmaxq_f32(orig_c, orig_d));
                min_all_0 = vminq_f32(min_all_0, vminq_f32(orig_a, orig_b));
                min_all_1 = vminq_f32(min_all_1, vminq_f32(orig_c, orig_d));
            }

            let max = vmaxvf(vmaxq_f32(max_all_0, max_all_1));
            let min = vminvf(vminq_f32(min_all_0, min_all_1));

            if min < HARDCLIP_MIN || max > HARDCLIP_MAX {
                let hardclip_min = vdupq_n_f32(HARDCLIP_MIN);
                let hardclip_max = vdupq_n_f32(HARDCLIP_MAX);
                let samples_ptr = samples.as_mut_ptr();
                for i in (0..blocked_size).step_by(BLOCK_SIZE) {
                    let orig_a = vld1q_f32(samples_ptr.add(i));
                    let orig_b = vld1q_f32(samples_ptr.add(i + 4));
                    let orig_c = vld1q_f32(samples_ptr.add(i + 8));
                    let orig_d = vld1q_f32(samples_ptr.add(i + 12));
                    let clipped_a = vminq_f32(hardclip_max, vmaxq_f32(orig_a, hardclip_min));
                    let clipped_b = vminq_f32(hardclip_max, vmaxq_f32(orig_b, hardclip_min));
                    let clipped_c = vminq_f32(hardclip_max, vmaxq_f32(orig_c, hardclip_min));
                    let clipped_d = vminq_f32(hardclip_max, vmaxq_f32(orig_d, hardclip_min));
                    vst1q_f32(samples_ptr.add(i), clipped_a);
                    vst1q_f32(samples_ptr.add(i + 4), clipped_b);
                    vst1q_f32(samples_ptr.add(i + 8), clipped_c);
                    vst1q_f32(samples_ptr.add(i + 12), clipped_d);
                }
            }

            exceeding1 = max > 1.0 || min < -1.0;
        }

        next_index = blocked_size;
    }

    for sample in &mut samples[next_index..] {
        let orig_val = *sample;
        *sample = orig_val.clamp(HARDCLIP_MIN, HARDCLIP_MAX);
        exceeding1 |= orig_val > 1.0 || orig_val < -1.0;
    }

    !exceeding1
}

/// ARM NEON specialization for `celt_float2int16`.
pub(super) fn celt_float2int16(input: &[f32], output: &mut [i16]) {
    const BLOCK_SIZE: usize = 16;
    let blocked_size = input.len() / BLOCK_SIZE * BLOCK_SIZE;

    unsafe {
        for i in (0..blocked_size).step_by(BLOCK_SIZE) {
            let input_ptr = input.as_ptr().add(i);
            let output_ptr = output.as_mut_ptr().add(i);

            let orig_a = vld1q_f32(input_ptr);
            let orig_b = vld1q_f32(input_ptr.add(4));
            let orig_c = vld1q_f32(input_ptr.add(8));
            let orig_d = vld1q_f32(input_ptr.add(12));

            let as_short_a = vqmovn_s32(vroundf(vmulq_n_f32(orig_a, float_cast::CELT_SIG_SCALE)));
            let as_short_b = vqmovn_s32(vroundf(vmulq_n_f32(orig_b, float_cast::CELT_SIG_SCALE)));
            let as_short_c = vqmovn_s32(vroundf(vmulq_n_f32(orig_c, float_cast::CELT_SIG_SCALE)));
            let as_short_d = vqmovn_s32(vroundf(vmulq_n_f32(orig_d, float_cast::CELT_SIG_SCALE)));

            vst1_s16(output_ptr, as_short_a);
            vst1_s16(output_ptr.add(4), as_short_b);
            vst1_s16(output_ptr.add(8), as_short_c);
            vst1_s16(output_ptr.add(12), as_short_d);
        }
    }

    super::celt_float2int16_scalar(&input[blocked_size..], &mut output[blocked_size..]);
}