Jungfraujoch/compression/JFJochDecompress.h

// SPDX-FileCopyrightText: 2024 Filip Leonarski, Paul Scherrer Institute <filip.leonarski@psi.ch>
// SPDX-License-Identifier: GPL-3.0-only

#ifndef JUNGFRAUJOCH_JFJOCHDECOMPRESS_H
#define JUNGFRAUJOCH_JFJOCHDECOMPRESS_H

#include <vector>
#include <cstring>

#include <bitshuffle/bitshuffle.h>
#include <bitshuffle/bitshuffle_internals.h>
#include <bitshuffle_hperf/bitshuffle.h>
#include <lz4/lz4.h>
#include <zstd.h>

#include "../compression/CompressionAlgorithmEnum.h"

#include "../common/JFJochException.h"
#include "../common/CompressedImage.h"

extern "C" {
    uint64_t bshuf_read_uint64_BE(const void* buf);
};

inline size_t JFJochDecompressHperfPtr(uint8_t *output,
                                       CompressionAlgorithm algorithm,
                                       const uint8_t *source,
                                       size_t source_size,
                                       size_t nelements,
                                       size_t elem_size,
                                       size_t block_size) {
    if ((algorithm != CompressionAlgorithm::BSHUF_LZ4) &&
        (algorithm != CompressionAlgorithm::BSHUF_ZSTD) &&
        (algorithm != CompressionAlgorithm::BSHUF_ZSTD_RLE))
        throw JFJochException(JFJochExceptionCategory::Compression, "Algorithm not supported by hperf decompressor");

    if ((block_size % BSHUF_BLOCKED_MULT) != 0)
        throw JFJochException(JFJochExceptionCategory::Compression, "Invalid block size");

    std::vector<char> decompressed_block(block_size * elem_size);
    std::vector<char> scratch(block_size * elem_size);

    const uint8_t *src_ptr = source;
    uint8_t *dst_ptr = output;

    const size_t num_full_blocks = nelements / block_size;
    const size_t reminder_size = nelements - num_full_blocks * block_size;
    const size_t last_block_size = reminder_size - reminder_size % BSHUF_BLOCKED_MULT;

    auto decode_block = [&](size_t current_nelements) {
        const auto compressed_size = static_cast<size_t>(bshuf_read_uint32_BE(src_ptr));
        src_ptr += 4;

        const size_t expected_size = current_nelements * elem_size;
        size_t decompressed_size = 0;

        switch (algorithm) {
            case CompressionAlgorithm::BSHUF_LZ4: {
                const int ret = LZ4_decompress_safe(reinterpret_cast<const char *>(src_ptr),
                                                    decompressed_block.data(),
                                                    static_cast<int>(compressed_size),
                                                    static_cast<int>(expected_size));
                if (ret < 0 || static_cast<size_t>(ret) != expected_size)
                    throw JFJochException(JFJochExceptionCategory::Compression, "LZ4 decompression error");
                decompressed_size = static_cast<size_t>(ret);
                break;
            }
            case CompressionAlgorithm::BSHUF_ZSTD:
            case CompressionAlgorithm::BSHUF_ZSTD_RLE: {
                const size_t ret = ZSTD_decompress(decompressed_block.data(),
                                                   expected_size,
                                                   src_ptr,
                                                   compressed_size);
                if (ZSTD_isError(ret) || ret != expected_size)
                    throw JFJochException(JFJochExceptionCategory::Compression, "ZSTD decompression error");
                decompressed_size = ret;
                break;
            }
            default:
                throw JFJochException(JFJochExceptionCategory::Compression, "Algorithm not supported");
        }

        if (bitshuf_decode_block(reinterpret_cast<char *>(dst_ptr),
                                 decompressed_block.data(),
                                 scratch.data(),
                                 current_nelements,
                                 elem_size) < 0)
            throw JFJochException(JFJochExceptionCategory::Compression, "bitshuffle_hperf decode error");

        src_ptr += compressed_size;
        dst_ptr += decompressed_size;
    };

    for (size_t i = 0; i < num_full_blocks; ++i)
        decode_block(block_size);

    if (last_block_size > 0)
        decode_block(last_block_size);

    const size_t leftover_bytes = (reminder_size % BSHUF_BLOCKED_MULT) * elem_size;
    if (leftover_bytes > 0) {
        memcpy(dst_ptr, src_ptr, leftover_bytes);
        src_ptr += leftover_bytes;
    }

    return static_cast<size_t>(src_ptr - source);
}

inline void JFJochDecompressPtr(uint8_t *output,
                      CompressionAlgorithm algorithm,
                      const uint8_t *source,
                      size_t source_size,
                      size_t nelements,
                      size_t elem_size,
                      bool use_hperf = true) {
    size_t block_size;
    if (algorithm != CompressionAlgorithm::NO_COMPRESSION) {
        if (bshuf_read_uint64_BE(const_cast<uint8_t *>(source)) != nelements * elem_size)
            throw JFJochException(JFJochExceptionCategory::Compression, "Mismatch in size");
        auto tmp = bshuf_read_uint32_BE(source + 8);
        block_size = tmp / elem_size;
    }

    switch (algorithm) {
        case CompressionAlgorithm::NO_COMPRESSION:
            if (source_size != nelements * elem_size)
                throw JFJochException(JFJochExceptionCategory::Compression, "Mismatch in size");
            memcpy(output, source, source_size);
            break;
        case CompressionAlgorithm::BSHUF_LZ4:
            if (use_hperf) {
                if (JFJochDecompressHperfPtr(output, algorithm, source + 12, source_size - 12,
                                             nelements, elem_size, block_size) != source_size - 12)
                    throw JFJochException(JFJochExceptionCategory::Compression, "Decompression error");
            } else {
                if (bshuf_decompress_lz4(source + 12, output, nelements,
                                         elem_size, block_size) != source_size - 12)
                    throw JFJochException(JFJochExceptionCategory::Compression, "Decompression error");
            }
            break;
        case CompressionAlgorithm::BSHUF_ZSTD_RLE:
        case CompressionAlgorithm::BSHUF_ZSTD:
            if (use_hperf) {
                if (JFJochDecompressHperfPtr(output, algorithm, source + 12, source_size - 12,
                                             nelements, elem_size, block_size) != source_size - 12)
                    throw JFJochException(JFJochExceptionCategory::Compression, "Decompression error");
            } else {
                if (bshuf_decompress_zstd(source + 12, output, nelements,
                                          elem_size, block_size) != source_size - 12)
                    throw JFJochException(JFJochExceptionCategory::Compression, "Decompression error");
            }
            break;
        default:
            throw JFJochException(JFJochExceptionCategory::Compression, "Not implemented algorithm");
    }
}

template <class Td, class Ts>
void JFJochDecompress(std::vector<Td> &output, CompressionAlgorithm algorithm, const Ts *source_v, size_t source_size,
                      size_t nelements, bool use_hperf = true) {
    output.resize(nelements);
    JFJochDecompressPtr((uint8_t *) output.data(), algorithm, (uint8_t *) source_v, source_size,
                        nelements, sizeof(Td), use_hperf);
}

template <class Td, class Ts>
void JFJochDecompress(std::vector<Td> &output, CompressionAlgorithm algorithm, const std::vector<Ts> source_v,
                      size_t nelements, bool use_hperf = true) {
    JFJochDecompress(output, algorithm, source_v.data(), source_v.size() * sizeof(Ts), nelements, use_hperf);
}

#endif //JUNGFRAUJOCH_JFJOCHDECOMPRESS_H