LZMA2 In-Memory Compression/Decompression

lzma2-main.cpp

#include <Windows.h>
#include <iostream>
#include <fstream>
#include <vector>
#include "swapi.h" // Stopwatch
#include <fast-lzma2.h>

using namespace std;
using namespace timer;

const int LZMA2_COMPRESSION_LEVEL = 9;

#define FL2_CLEVEL_DEFAULT  6
#define FL2_MAX_CLEVEL      9

#define print(fmt,...) printf(fmt "\n",__VA_ARGS__);

void DumpBufferToDisk(const char* bufferName, void* bufferPtr, size_t bufferSize)
{
	ofstream	dumper;
	string bufferFileName(bufferName);
	bufferFileName += "-dumped.bin";
	dumper.open(bufferFileName.c_str(), ios::binary | ios::out);
	dumper.write((char*)bufferPtr, bufferSize);
	dumper.close();
}

Stopwatch sw_simple;
Stopwatch sw_advanced;

// simple
int simple_mt() 
{
	print("Initializing...");

	// App Code
	std::ifstream file_reader("DemoFile.bin", std::ios::binary);
	std::vector<unsigned char> file_buffer(std::istreambuf_iterator<char>(file_reader), {});

	// Compress
	std::vector<unsigned char> compressed;
	compressed.resize(file_buffer.size());

	print("Compressing...");
	sw_simple.Start();
	size_t compressed_size = 
		FL2_compressMt(compressed.data(), compressed.size(), file_buffer.data(), file_buffer.size(), LZMA2_COMPRESSION_LEVEL, 8);
	print("Compressed");

	print("Simple :: Compressed at %f ms with no error.", sw_simple.ElapsedMilliseconds());
	DumpBufferToDisk("DemoFile-Simple.lzma2", compressed.data(), compressed_size);

	// Decompress
	std::vector<unsigned char> decompressed;
	decompressed.resize(compressed_size * 100);

	print("Decompressing...");
	size_t decompressed_size = 
		FL2_decompressMt(decompressed.data(), decompressed.size(), compressed.data(), compressed.size(), 8);
	print("Decompressed");

	print("Simple :: Decompressed at %f ms with no error.", sw_simple.ElapsedMilliseconds());
	DumpBufferToDisk("DemoFile-Simple.decompressed", decompressed.data(), decompressed_size);

	print("Clean up...");
	std::vector<unsigned char>().swap(compressed);
	std::vector<unsigned char>().swap(decompressed);
	std::vector<unsigned char>().swap(file_buffer);

	print("Simple :: Executed in %f ms with no error.", sw_simple.ElapsedMilliseconds());
}

// advanced
int advanced_mt()
{
	print("Initializing...");

	// App Code
	std::ifstream file_reader("DemoFile.bin", std::ios::binary);
	std::vector<unsigned char> file_buffer(std::istreambuf_iterator<char>(file_reader), {});

	// Create Compressor Engine
	FL2_CCtx* cctx = FL2_createCCtxMt(8);
	FL2_CCtx_setParameter(cctx, FL2_p_compressionLevel, FL2_MAX_CLEVEL); // Maximum is 10 
	FL2_CCtx_setParameter(cctx, FL2_p_highCompression, 9); 
	FL2_CCtx_setParameter(cctx, FL2_p_dictionarySize, FL2_DICTSIZE_MAX);

	// Compress
	std::vector<unsigned char> compressed;
	compressed.resize(file_buffer.size());

	print("Compressing...");
	sw_advanced.Start();
	size_t compressed_size =
		FL2_compressCCtx(cctx, compressed.data(), compressed.size(), file_buffer.data(), file_buffer.size(), LZMA2_COMPRESSION_LEVEL);
	print("Compressed");

	print("Advanced :: Compressed at %f ms with no error.", sw_advanced.ElapsedMilliseconds());
	DumpBufferToDisk("DemoFile-Advanced.lzma2", compressed.data(), compressed_size);

	// Decompress
	std::vector<unsigned char> decompressed;
	decompressed.resize(compressed_size * 100);

	print("Decompressing...");
	size_t decompressed_size =
		FL2_decompressMt(decompressed.data(), decompressed.size(), compressed.data(), compressed.size(), 8);
	print("Decompressed");

	print("Advanced :: Decompressed at %f ms with no error.", sw_advanced.ElapsedMilliseconds());
	DumpBufferToDisk("DemoFile-Advanced.decompressed", decompressed.data(), decompressed_size);

	print("Clean up...");
	FL2_freeCCtx(cctx);
	std::vector<unsigned char>().swap(compressed);
	std::vector<unsigned char>().swap(decompressed);
	std::vector<unsigned char>().swap(file_buffer);

	print("Advanced :: Executed in %f ms with no error.", sw_advanced.ElapsedMilliseconds());
}

// main
int main() 
{
	simple_mt();
	advanced_mt();
	getchar();
}

This is the implementation of the FL2_compressMt()

FL2LIB_API size_t FL2LIB_CALL FL2_compressMt(void* dst, size_t dstCapacity,
    const void* src, size_t srcSize,
    int compressionLevel,
    unsigned nbThreads)
{
    FL2_CCtx* const cctx = FL2_createCCtxMt(nbThreads);
    if (cctx == NULL)
        return FL2_ERROR(memory_allocation);

    size_t const cSize = FL2_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);

    FL2_freeCCtx(cctx);

    return cSize;
}

As you can see, there is nothing extra that is going on, basically it's :

1. Create context
2. Compress,
3. Release context

Ok, after some playing, the impact is from the FL2_CCtx_setParameter(cctx, FL2_p_highCompression, 9);. Remove the line #85 one and the advanced is even faster. I have tested on Linux on ARM, but I assume will be same on Wondows on X86

Ok, after some playing, the impact is from the FL2_CCtx_setParameter(cctx, FL2_p_highCompression, 9);. Remove the line #85 one and the advanced is even faster. I have tested on Linux on ARM, but I assume will be same on Wondows on X86

Oh, they became same for me on windows, so how to set dic size to 1024 or can I even set it to higher values? I don't mind speed I want highest ratio, still with 7zip LZMA2 slow mode it becomes 1.95MB

If you are not chasing the speed, set dictionary to FL2_DICTSIZE_MIN, try also to play with FL2_p_resetInterval, set FL2_p_searchDepth to max, FL2_p_strategy=2

It's more of a trial and error, since the size of output also depends on the nature of the input. More randomized input, larger the compressed file. More uniform input, less compressed size.