audio_sample_packer.hpp

// Copyright Take Vos 2021.
// Distributed under the Boost Software License, Version 1.0.
// (See accompanying file LICENSE_1_0.txt or copy at https://www.boost.org/LICENSE_1_0.txt)
 
#pragma once
 
#include "audio_sample_format.hpp"
#include "../utility/utility.hpp"
#include "../SIMD/module.hpp"
#include "../random/random.hpp"
#include "../macros.hpp"
#include <cstddef>
#include <bit>
 
hi_export_module(hikogui.audio.audio_sample_packer);
 
namespace hi { inline namespace v1 {
 
hi_export class audio_sample_packer {
public:
    audio_sample_packer(audio_sample_format format, std::size_t stride) noexcept :
        _dither(format.num_bits), _format(format), _stride(stride)
    {
        _store_shuffle_indices = format.store_shuffle_indices(stride);
        _concat_shuffle_indices = format.concat_shuffle_indices(stride);
 
        _multiplier = f32x4::broadcast(format.pack_multiplier());
 
        _num_chunks_per_quad = format.num_chunks_per_quad(stride);
        _chunk_stride = format.chunk_stride(stride);
 
        _direction = format.endian == std::endian::little ? 1 : -1;
        _start_byte = format.endian == std::endian::little ? 0 : format.num_bytes - 1;
        _align_shift = 32 - format.num_bytes * 8;
    }
 
    void operator()(float const *hi_restrict src, std::byte *hi_restrict dst, std::size_t num_samples) const noexcept
    {
        hi_assert(src != nullptr);
        hi_assert(dst != nullptr);
 
        // Calculate a conservative number of samples that can be copied quickly
        // without overflowing the dst buffer.
        hilet src_end = src + num_samples;
        hilet src_fast_end = src + _format.num_fast_quads(_stride, num_samples) * 4;
 
        hilet store_shuffle_indices = _store_shuffle_indices;
        hilet concat_shuffle_indices = _concat_shuffle_indices;
        hilet num_chunks_per_quad = _num_chunks_per_quad;
        hilet chunk_stride = _chunk_stride;
 
        if (_format.is_float) {
            while (src != src_fast_end) {
                hilet float_samples = load_samples(src);
                hilet int_samples = i8x16::cast_from(float_samples);
                store_samples(int_samples, dst, store_shuffle_indices, concat_shuffle_indices, num_chunks_per_quad, chunk_stride);
            }
            while (src != src_end) {
                hilet float_sample = load_sample(src);
                hilet int_sample = std::bit_cast<int32_t>(float_sample);
                store_sample(int_sample, dst, _stride, _format.num_bytes, _direction, _start_byte, _align_shift);
            }
 
        } else {
            hilet multiplier = _multiplier;
            hilet one = f32x4::broadcast(1);
            hilet min_one = f32x4::broadcast(-1);
 
            auto dither = _dither;
 
            while (src != src_fast_end) {
                hilet dither_value = dither.next();
 
                auto float_samples = load_samples(src);
                float_samples += dither_value;
                float_samples = min(float_samples, one);
                float_samples = max(float_samples, min_one);
                float_samples *= multiplier;
                hilet int_samples = i8x16::cast_from(static_cast<i32x4>(float_samples));
                store_samples(int_samples, dst, store_shuffle_indices, concat_shuffle_indices, num_chunks_per_quad, chunk_stride);
            }
            while (src != src_end) {
                hilet dither_value = dither.next();
 
                auto float_sample = f32x4::broadcast(load_sample(src));
                float_sample += dither_value;
                float_sample = min(float_sample, one);
                float_sample = max(float_sample, min_one);
                float_sample *= multiplier;
                hilet int_sample = get<0>(static_cast<i32x4>(float_sample));
                store_sample(int_sample, dst, _stride, _format.num_bytes, _direction, _start_byte, _align_shift);
            }
 
            _dither = dither;
        }
    }
 
private:
    i8x16 _store_shuffle_indices;
    i8x16 _concat_shuffle_indices;
    f32x4 _multiplier;
    mutable dither _dither;
    audio_sample_format _format;
    std::size_t _num_chunks_per_quad;
    std::size_t _stride;
    std::size_t _chunk_stride;
    int _direction;
    int _start_byte;
    int _align_shift;
 
    static void store_sample(
        int32_t int_sample,
        std::byte *& dst,
        std::size_t stride,
        int num_bytes,
        int direction,
        int start_byte,
        int align_shift) noexcept
    {
        int_sample >>= align_shift;
 
        hi_axiom(dst != nullptr);
        auto p = dst + start_byte;
        do {
            *p = static_cast<std::byte>(int_sample);
            p += direction;
            int_sample >>= 8;
        } while (--num_bytes);
 
        dst += stride;
    }
 
    [[nodiscard]] static void
    store_samples(i8x16 int_samples, std::byte *& dst, i8x16 store_shuffle_indices, std::size_t stride) noexcept
    {
        hi_axiom(dst != nullptr);
        hi_axiom(stride > 0);
 
        // Read out the samples from the other channels, that where packed before.
        auto tmp = i8x16::load(dst);
 
        hilet packed_samples = permute(int_samples, store_shuffle_indices);
 
        // When the shuffle-index is -1 use the samples from the other channels.
        tmp = blend(packed_samples, tmp, store_shuffle_indices);
 
        // Store back the samples from this channel and from the other channel.
        tmp.store(dst);
 
        dst += stride;
    }
 
    [[nodiscard]] static void store_samples(
        i8x16 int_samples,
        std::byte *& dst,
        i8x16 store_shuffle_indices,
        i8x16 concat_shuffle_indices,
        std::size_t num_chunks,
        std::size_t stride) noexcept
    {
        hi_assert(dst != nullptr);
        hi_assert(num_chunks > 0 and num_chunks <= 4);
        hi_assert(stride > 0);
 
        do {
            store_samples(int_samples, dst, store_shuffle_indices, stride);
            int_samples = permute(int_samples, concat_shuffle_indices);
            // The result of the last shuffle is not used, so will be pipelined by the CPU.
        } while (--num_chunks);
    }
 
    [[nodiscard]] static float load_sample(float const *& src) noexcept
    {
        hi_axiom(src != nullptr);
        return *(src++);
    }
 
    [[nodiscard]] static f32x4 load_samples(float const *& src) noexcept
    {
        hilet r = f32x4::load(src);
        src += 4;
        return r;
    }
};
 
}} // namespace hi::inline v1