audio_sample_format.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 "../utility/utility.hpp"
#include "../macros.hpp"
#include <hikocpu/hikocpu.hpp>
#include <bit>
 
hi_export_module(hikogui.audio.audio_sample_format);
 
hi_export namespace hi { inline namespace v1 {
 
hi_export struct audio_sample_format {
    uint8_t num_bytes;
 
    uint8_t num_guard_bits;
 
    uint8_t num_bits;
 
    bool is_float;
 
    std::endian endian;
 
    constexpr audio_sample_format() noexcept :
        num_bytes(0), num_guard_bits(0), num_bits(0), is_float(false), endian(std::endian::native)
    {
    }
 
    constexpr audio_sample_format(audio_sample_format const&) noexcept = default;
    constexpr audio_sample_format(audio_sample_format&&) noexcept = default;
    constexpr audio_sample_format& operator=(audio_sample_format const&) noexcept = default;
    constexpr audio_sample_format& operator=(audio_sample_format&&) noexcept = default;
 
    [[nodiscard]] constexpr audio_sample_format(
        uint8_t num_bytes,
        uint8_t num_guard_bits,
        uint8_t num_bits,
        bool is_float,
        std::endian endian) noexcept :
        num_bytes(num_bytes), num_guard_bits(num_guard_bits), num_bits(num_bits), is_float(is_float), endian(endian)
    {
        hi_axiom(holds_invariant());
    }
 
    [[nodiscard]] constexpr static audio_sample_format float32_le() noexcept
    {
        return {4, 8, 23, true, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format float32_be() noexcept
    {
        return {4, 8, 23, true, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format float32() noexcept
    {
        return {4, 8, 23, true, std::endian::native};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int16_le() noexcept
    {
        return {2, 0, 15, false, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int16_be() noexcept
    {
        return {2, 0, 15, false, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int16() noexcept
    {
        return {2, 0, 15, false, std::endian::native};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int20_le() noexcept
    {
        return {3, 0, 19, false, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int20_be() noexcept
    {
        return {3, 0, 19, false, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int20() noexcept
    {
        return {3, 0, 19, false, std::endian::native};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int24_le() noexcept
    {
        return {3, 0, 23, false, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int24_be() noexcept
    {
        return {3, 0, 23, false, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int24() noexcept
    {
        return {3, 0, 23, false, std::endian::native};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int32_le() noexcept
    {
        return {4, 0, 31, false, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int32_be() noexcept
    {
        return {4, 0, 31, false, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format int32() noexcept
    {
        return {4, 0, 31, false, std::endian::native};
    }
 
    [[nodiscard]] constexpr static audio_sample_format fix8_23_le() noexcept
    {
        return {4, 8, 23, false, std::endian::little};
    }
 
    [[nodiscard]] constexpr static audio_sample_format fix8_23_be() noexcept
    {
        return {4, 8, 23, false, std::endian::big};
    }
 
    [[nodiscard]] constexpr static audio_sample_format fix8_23() noexcept
    {
        return {4, 8, 23, false, std::endian::native};
    }
 
    constexpr explicit operator bool() const noexcept
    {
        return num_bytes != 0;
    }
 
    [[nodiscard]] float pack_multiplier() const noexcept
    {
        if (is_float) {
            return 1.0f;
 
        } else {
            // Find the maximum value of the fraction bits as a signed number.
            auto max_value = (1_uz << num_bits) - 1;
 
            // Align left inside an int32_t.
            hi_assert(num_bits + num_guard_bits <= 31);
            max_value <<= narrow_cast<size_t>(31 - num_bits - num_guard_bits);
 
            return static_cast<float>(max_value);
        }
    }
 
    [[nodiscard]] float unpack_multiplier() const noexcept
    {
        return 1.0f / pack_multiplier();
    }
 
    [[nodiscard]] std::size_t num_samples_per_chunk(std::size_t stride) const noexcept
    {
        auto r = narrow_cast<int>(std::bit_floor((((16u - num_bytes) / stride) & 3) + 1));
        hi_assert(r == 1 or r == 2 or r == 4);
        return r;
    }
 
    [[nodiscard]] std::size_t chunk_stride(std::size_t stride) const noexcept
    {
        return stride * num_samples_per_chunk(stride);
    }
 
    [[nodiscard]] std::size_t num_chunks_per_quad(std::size_t stride) const noexcept
    {
        return 4 / num_samples_per_chunk(stride);
    }
 
    [[nodiscard]] std::size_t num_fast_quads(std::size_t stride, std::size_t num_samples) const noexcept
    {
        auto const src_buffer_size = (num_samples - 1) * stride + num_bytes;
        if (src_buffer_size < 16) {
            return 0;
        }
 
        auto const num_chunks = (src_buffer_size - 16) / chunk_stride(stride) + 1;
        return num_chunks / num_chunks_per_quad(stride);
    }
 
    [[nodiscard]] i8x16 load_shuffle_indices(std::size_t stride) const noexcept
    {
        auto const num_samples = num_samples_per_chunk(stride);
 
        // Indices set to -1 result in a zero after a byte shuffle.
        auto r = i8x16::broadcast(-1);
        for (int sample_nr = 0; sample_nr != num_samples; ++sample_nr) {
            auto const sample_src_offset = sample_nr * stride;
 
            // Offset the samples to the highest elements in the i32x4 vector.
            // By shifting the samples from high to low together with 'OR' we can
            // concatenate 1, 2, or 4 loads into a single 4 samples vector.
            // Where the sample in the lowest index is the first sample in memory.
            auto const sample_dst_offset = (sample_nr + (4 - num_samples)) * 4;
 
            // Bytes are ordered least to most significant.
            for (int byte_nr = 0; byte_nr != num_bytes; ++byte_nr) {
                auto const src_offset = sample_src_offset + (endian == std::endian::little ? byte_nr : num_bytes - byte_nr - 1);
 
                // Offset the bytes so they become aligned to the left.
                auto const dst_offset = sample_dst_offset + byte_nr + (4 - num_bytes);
 
                r[dst_offset] = narrow_cast<int8_t>(src_offset);
            }
        }
 
        return r;
    }
 
    [[nodiscard]] i8x16 store_shuffle_indices(std::size_t stride) const noexcept
    {
        auto const num_samples = num_samples_per_chunk(stride);
 
        // Indices set to -1 result in a zero after a byte shuffle.
        auto r = i8x16::broadcast(-1);
        for (int sample_nr = 0; sample_nr != num_samples; ++sample_nr) {
            auto const sample_dst_offset = sample_nr * stride;
 
            // Offset the samples to the lowest elements in the i32x4 vector.
            // By shifting the samples from high to low we can extract 1, 2, or 4 stores
            // from a single 4 samples vector.
            // Where the sample at the lowest index becomes the first sample in memory.
            auto const sample_src_offset = sample_nr * 4;
 
            // Bytes are ordered least to most significant.
            for (int byte_nr = 0; byte_nr != num_bytes; ++byte_nr) {
                auto const dst_offset = sample_dst_offset + (endian == std::endian::little ? byte_nr : num_bytes - byte_nr - 1);
 
                // Offset the bytes so they become aligned to the left.
                auto const src_offset = sample_src_offset + byte_nr + (4 - num_bytes);
 
                r[dst_offset] = narrow_cast<int8_t>(src_offset);
            }
        }
 
        return r;
    }
 
    [[nodiscard]] i8x16 concat_shuffle_indices(std::size_t stride) const noexcept
    {
        auto const num_samples = num_samples_per_chunk(stride);
 
        // The bytes are shifted right.
        auto const byte_shift = (4 - num_samples) * 4;
 
        return i8x16::byte_srl_shuffle_indices(narrow_cast<unsigned int>(byte_shift));
    }
 
    [[nodiscard]] constexpr bool holds_invariant() const noexcept
    {
        return (num_bytes >= 1 && num_bytes <= 4) && (num_bits + num_guard_bits <= num_bytes * 8) &&
            (endian == std::endian::little || endian == std::endian::big);
    }
};
 
}} // namespace hi::inline v1