Compression using miniz

jubueche
Posts: 32
Joined: Sat Jan 19, 2019 4:03 pm

Compression using miniz

Postby jubueche » Fri Feb 08, 2019 10:26 am

Hi,

I am trying to run the test sketch for miniz compression. However, my heap space is too small (available size is bigger, but fragmented) and I was wondering how to reduce the size used by

Code: Select all

tdefl_compressor
without breaking the algorithm.

Here is the short code that I am trying to run:

Code: Select all

#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "rom/miniz.h"
#include "unity.h" //testing
#include "stdlib.h"
#include "stdio.h"


#define DATASIZE (1024*1)

extern "C"{
#include "wifi_synch.h"
}

extern "C" void app_main()
{


	int x;
	char b;
	char *inbuf, *outbuf;
	tdefl_compressor *comp;
	tinfl_decompressor *decomp;
	tdefl_status status;
	size_t inbytes = 0, outbytes = 0, inpos = 0, outpos = 0, compsz;

	printf("Allocating compressor & outbuf (%d bytes)\n", sizeof(tdefl_compressor));
	comp = (tdefl_compressor *) malloc(sizeof(tdefl_compressor));
	TEST_ASSERT(comp != NULL);


	printf("Allocating data buffer and filling it with semi-random data\n");
	inbuf = (char *) malloc(DATASIZE*sizeof(char));
	TEST_ASSERT(inbuf != NULL);
	srand(0);
	for (x = 0; x < DATASIZE; x++) {
		inbuf[x] = (x & 1) ? rand() & 0xff : 0;
	}

	int free_space = heap_caps_get_free_size(MALLOC_CAP_8BIT);
	printf("Free space is %d\n", free_space);

	outbuf = (char *) malloc(DATASIZE);
	TEST_ASSERT(outbuf != NULL);
	printf("Compressing...\n");
	status = tdefl_init(comp, NULL, NULL, TDEFL_WRITE_ZLIB_HEADER | 1500);
	TEST_ASSERT(status == TDEFL_STATUS_OKAY);
	while (inbytes != DATASIZE) {
		outbytes = DATASIZE - outpos;
		inbytes = DATASIZE - inpos;
		tdefl_compress(comp, &inbuf[inpos], &inbytes, &outbuf[outpos], &outbytes, TDEFL_FINISH);
		printf("...Compressed %d into %d bytes\n", inbytes, outbytes);
		inpos += inbytes; outpos += outbytes;
	}
	compsz = outpos;
	free(comp);
	//Kill inbuffer
	for (x = 0; x < DATASIZE; x++) {
		inbuf[x] = 0;
	}
	free(inbuf);

	inbuf = outbuf;
	outbuf = (char *) malloc(DATASIZE);
	TEST_ASSERT(outbuf != NULL);
	printf("Reinflating...\n");
	decomp = (tinfl_decompressor *) malloc(sizeof(tinfl_decompressor));
	TEST_ASSERT(decomp != NULL);
	tinfl_init(decomp);
	inpos = 0; outpos = 0;
	while (inbytes != compsz) {
		outbytes = DATASIZE - outpos;
		inbytes = compsz - inpos;
		tinfl_decompress(decomp, (const mz_uint8 *)&inbuf[inpos], &inbytes, (uint8_t *)outbuf, (mz_uint8 *)&outbuf[outpos], &outbytes, TINFL_FLAG_PARSE_ZLIB_HEADER);
		printf("...Decompressed %d into %d bytes\n", inbytes, outbytes);
		inpos += inbytes; outpos += outbytes;
	}
	printf("Checking if same...\n");
	srand(0);
	for (x = 0; x < DATASIZE; x++) {
		b = (x & 1) ? rand() & 0xff : 0;
		if (outbuf[x] != b) {
			printf("Pos %x: %hhx!=%hhx\n", x, outbuf[x], b);
			TEST_ASSERT(0);
		}
	}
	printf("Great Success!\n");
	free(inbuf);
	free(outbuf);
	free(decomp);


	while(1){
		vTaskDelay(100);
	}
}
I am also looking into zlib.
I am trying to compress sensor readings. My plan is to continuously read sensor data into heap memory that is say 32kB and each time I hit say 16kB, I compress this chunk and write it to external SPI flash. OR I can write everything to external flash and compress ~16MB worth of data before uploading to the cloud. What do you think is better? I don't know how long it takes to compress 16MB worth of data.

Also what compression rate at normal compression level (6 for zlib) can be expected?

User avatar
fasani
Posts: 197
Joined: Wed Jan 30, 2019 12:00 pm
Location: Barcelona
Contact:

Re: Compression using miniz

Postby fasani » Tue May 14, 2019 8:18 am

Dear @jubueche
did you found a way to this and make it work ?

I'm looking here for some working example reading a small GZ from SPIFFs:

https://www.esp32.com/viewtopic.php?f=1 ... =10#p43154


Pointers to existing #esp32 miniz gzip compression / decompression examples
https://github.com/espressif/esp-idf/bl ... st_miniz.c ESP-IDF test
https://www.esp32.com/viewtopic.php?f=13&t=1076 Discussion in esp32 forum
epdiy collaborator | http://fasani.de Fan of Espressif MCUs and electronic design

jubueche
Posts: 32
Joined: Sat Jan 19, 2019 4:03 pm

Re: Compression using miniz

Postby jubueche » Tue May 14, 2019 9:18 am

Code: Select all

#include "compression.h"
#include "assert.h"
#include "zlib.h"
#include <string.h>
#include "stdio.h"
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "stdlib.h"
#include "esp_log.h"

const char * TAG = "Chaze-Compression";

/**
 * @brief When recording sensor data, multiple tasks need to acquire a mutex in order to read from the bus and
 * write to the buffer. The buffer is a struct that holds a counter (number of bytes written) and a malloced
 * char array. If the number of bytes written equals to the BUFFER SIZE, this function is called and the
 * data is compressed and saved to the flash.
 * @param Buffer number to read from and compress. Either 0 or 1.
 */
void compress_and_save(uint8_t buff_num)
{
	ESP_LOGI(TAG, "Using buffer number %d", buff_num);
	//Get the pointer to the current buffer
	buffer_t * current_buffer = buffers[buff_num];
	assert(current_buffer->counter == BUFFER_SIZE); //The buffer must be filled completely.

	//Data written to flash.
	char * out = (char *) malloc(BUFFER_SIZE * sizeof(char));

	if(out == NULL){
		ESP_LOGE(TAG, "Could not allocate enough space");
		//Need to write the raw, uncompressed data to the flash instead
		return;
	} else{

		ESP_LOGI(TAG, "Start compressing.");
		//Writes 'written' many chars to the out buffer previously allocated
		uint32_t written = def(current_buffer->data, out, DEFAULT_COMPRESSION_LEVEL);

		//Write data to flash
		write_data_to_flash(out, written);
		free(out);
	}
}

/**
 * @brief Deflates the char array given by source and writes to the char array dest.
 * @param source
 * @param dest
 * @param level Level of compression.
 * @return Number of written bytes to dest
 */
uint32_t def(char * source, char * dest, uint8_t level)
{
	int ret, flush;
	unsigned have;
	z_stream strm;

	unsigned char * in = (unsigned char *) malloc(CHUNK * sizeof(char));
	unsigned char * out = (unsigned char *) malloc(CHUNK * sizeof(char));

	strm.zalloc = Z_NULL;
	strm.zfree = Z_NULL;
	strm.opaque = Z_NULL;
	ret = deflateInit2(&strm, level,Z_DEFLATED, WINDOW_SIZE ,MEM_LEVEL,Z_DEFAULT_STRATEGY);
	zerr(ret);

	uint32_t dest_offset = 0;
    do {
        memcpy(in, source, CHUNK);
        strm.avail_in = CHUNK; /*We assert that the buffer is full and since chunk size = buffer size, we can read that many bytes*/
        flush = Z_FINISH; //Immediately flush since we only compress one chunk.
        strm.next_in = in;

        do {
            strm.avail_out = CHUNK;
            strm.next_out = out;
            ret = deflate(&strm, flush);
            assert(ret != Z_STREAM_ERROR);
            zerr(ret);
            have = CHUNK - strm.avail_out;
            ESP_LOGI(TAG, "Have: %d", have);
            for(int i=0;i<have;i++)
            	printf("%c", out[i]);

            memcpy(dest+dest_offset, out, have);
            dest_offset += have;
        } while (strm.avail_out == 0);
        assert(strm.avail_in == 0);

    } while (flush != Z_FINISH);
    assert(ret == Z_STREAM_END);

    zerr(deflateEnd(&strm));
    free(in);
    free(out);
    return dest_offset;
}

//For now just inflates data to check if was deflated correctly
void write_data_to_flash(char * data, uint32_t n)
{
	int ret;
	unsigned have;
	z_stream strm;

	unsigned char * in = (unsigned char *) malloc(CHUNK * sizeof(char));
	unsigned char * out = (unsigned char *) malloc(CHUNK * sizeof(char));


	char * to_print = (char *) malloc(CHUNK*sizeof(char));

	if(to_print == NULL || in == NULL || out == NULL)
	{
		ESP_LOGE(TAG, "Could not allocate enough space");
		return;
	}

	strm.zalloc = Z_NULL;
	strm.zfree = Z_NULL;
	strm.opaque = Z_NULL;
	strm.avail_in = 0;
	strm.next_in = Z_NULL;

	ret = inflateInit2(&strm,0); //Use window size in compressed stream
	zerr(ret);
	uint32_t to_print_offset = 0;

	do {
			memcpy(in, data, n);
			strm.avail_in = n;

			if (strm.avail_in == 0)
				break;
			strm.next_in = in;

			do {
				strm.avail_out = CHUNK;
				strm.next_out = out;

				ret = inflate(&strm, Z_NO_FLUSH);
				assert(ret != Z_STREAM_ERROR);
				zerr(ret);

				have = CHUNK - strm.avail_out;
				memcpy(to_print+to_print_offset, out, have);
				to_print_offset+=have;

			} while (strm.avail_out == 0);
		} while (ret != Z_STREAM_END);

	(void)inflateEnd(&strm);
	free(in);
	free(out);

	for(int i=0;i<CHUNK;i++)
		printf("%c", to_print[i]);
}


void zerr(int ret)
{
    switch (ret) {
    case Z_ERRNO:
        ESP_LOGE(TAG, "Z err no");
        break;
    case Z_STREAM_ERROR:
    	ESP_LOGE(TAG, "Invalid compression level");
        break;
    case Z_DATA_ERROR:
    	ESP_LOGE(TAG, "Data error");
        break;
    case Z_MEM_ERROR:
    	ESP_LOGE(TAG, "Mem error");
        break;
    case Z_VERSION_ERROR:
    	ESP_LOGE(TAG, "Version error");
    }
}

Code: Select all

#ifndef COMPRESSION_H
#define COMPRESSION_H

#include "stdint.h"


#define BUFFER_SIZE /*16384*/ 8192
#define CHUNK 8192 /*Must be same size as BUFFER_SIZE*/
#define DEFAULT_COMPRESSION_LEVEL 6
#define WINDOW_SIZE 9
#define MEM_LEVEL 1
#define GZIP_ENCODING 16


typedef struct {
	int32_t counter;
	char * data;
}buffer_t;

buffer_t * buffers[2];

void compress_and_save(uint8_t);
void write_data_to_flash(char *, uint32_t);
uint32_t def(char *, char *, uint8_t);
void zerr(int32_t);


#endif


//Working main.cpp file.
//Creates task simulating sensor readings with mutex, compresses if buffer full, switches buffer and writes to next one
/*
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "stdlib.h"
#include "stdio.h"
#include <stdio.h>
#include <string.h>
extern "C" {
#include "compression.h"
}
#include "freertos/semphr.h"
#include "esp_log.h"
SemaphoreHandle_t xSemaphore = NULL;
volatile uint8_t buff_idx = 0;
static const char * TAG = "Chaze-Compression-Main";
void sample_hr(void * pvParams)
{
	char dummy[] = "Sorem ipsum lorem ich habe hier einen schönen\nAusblich wäre ich hier irgendwas rein schreibe. Ich sitze an meinem Schreibtisch\nund drehe ein bisschen Daeumchen und so weiter und so fort.\n";
	for(;;){
		//Try to aquire the mutex for a given amount of time then back-off
		//For real life, need priorites in low sample rate sensors, random back-off time etc.
		buffer_t * curr_buff = buffers[buff_idx];
		if(xSemaphore != NULL)
		{
			if(xSemaphoreTake(xSemaphore,(TickType_t) 100) == pdTRUE)
			{
				ESP_LOGI(TAG, "Acquired the lock");
				if(BUFFER_SIZE-curr_buff->counter >= strlen(dummy)){
					ESP_LOGI(TAG, "Enough space to write");
					//Enough space, we can write
					memcpy(curr_buff->data+curr_buff->counter, dummy, strlen(dummy));
					curr_buff->counter += strlen(dummy);
				} else{
					//Fill up the buffer with 'f' and set the buff_idx to buff_idx XOR 1
					ESP_LOGI(TAG, "Buffer almost full, fill up");
					for(int i=curr_buff->counter;i<BUFFER_SIZE-curr_buff->counter;i++)
						curr_buff->data[i] = 'f';
					curr_buff->counter += BUFFER_SIZE-curr_buff->counter;
					ESP_LOGI(TAG, "Counter is %d", curr_buff->counter);
					if(buff_idx==0)
						buff_idx=1;
					else buff_idx=0;
					ESP_LOGI(TAG, "Call compress");
					//Call compress. Switch the bit back since we want to compress the buffer that is full
					compress_and_save(!buff_idx);
					curr_buff->counter = 0; //Reset the compressed buffer
				}
				ESP_LOGI(TAG, "Release mutex");
				//Release mutex
				xSemaphoreGive(xSemaphore);
			}
			vTaskDelay(100 / portTICK_PERIOD_MS); //Back off a little longer
		}
	}
}
extern "C" void app_main()
{
	 xSemaphore = xSemaphoreCreateMutex();
	//Initialize buffers 1 and 2
	buffer_t buffer0 = {};
	buffer0.data = (char *) malloc(BUFFER_SIZE);
	buffer0.counter = 0;
	buffer_t buffer1 = {};
	buffer1.data = (char *) malloc(BUFFER_SIZE);
	buffer1.counter = 0;
	buffers[0] = &buffer0;
	buffers[1] = &buffer1;
	if (xTaskCreate(&sample_hr, "sample_hr", 1024 * 8, NULL, 5, NULL) != pdPASS )
	{
		printf("Synch task create failed.\n");
	} else ESP_LOGI(TAG, "Created task");
	while(1){
		vTaskDelay(10000);
	}
}
/*
extern "C" void app_main()
{
	esp_vfs_fat_mount_config_t mount_config = {};
	mount_config.max_files = 4;
	mount_config.format_if_mount_failed = true;
	mount_config.allocation_unit_size = CONFIG_WL_SECTOR_SIZE;
	esp_err_t err = esp_vfs_fat_spiflash_mount(base_path, "storage", &mount_config, &s_wl_handle);
	if (err != ESP_OK) {
		ESP_LOGE(TAG, "Failed to mount FATFS (%s)", esp_err_to_name(err));
		return;
	}
	print_dir();
	remove_folder();
	print_dir();
	ESP_LOGI(TAG, "Opening file");
	foo = fopen("/spiflash/foo.bin", "wb");
	if (foo == NULL) {
		ESP_LOGE(TAG, "Failed to open file for writing");
		return;
	}
	//int ret = fprintf(foo, "written using ESP-IDF\n");
	const char * tmp = esp_get_idf_version();
	int ret = fwrite(tmp,1,strlen(tmp),foo);
	if(ret < strlen(tmp)){
		ESP_LOGE(TAG, "Error writing file");
	}
	if(fclose(foo) == EOF){
		ESP_LOGE(TAG, "Error closing the file.");
	}
	else ESP_LOGI(TAG, "File written");
	comp = fopen("/spiflash/c.bin", "wb");
	if (comp == NULL) {
		ESP_LOGE(TAG, "Failed to open file for writing");
		return;
	}
	foo = fopen("/spiflash/foo.bin", "rb");
	if(foo == NULL){
		ESP_LOGE(TAG, "Error opening the file.");
	}
	print_dir();
	uint8_t level = 6;
	zerr(def(foo, comp, level));
	ESP_LOGI(TAG, "Compressed.");
	if(fclose(foo) == EOF){
		ESP_LOGE(TAG, "Error closing file");
	}
	else ESP_LOGI(TAG, "Closed foo.");
	if(fclose(comp) == EOF){
		ESP_LOGE(TAG, "Error closing file: %s", strerror(errno));
	}
	else ESP_LOGI(TAG, "Closed comp.");
	//Read comp
	char line[128];
	char *pos;
	// Open file for reading
	ESP_LOGI(TAG, "Reading file");
	comp = fopen("/spiflash/c.bin", "rb");
	if (comp == NULL) {
		ESP_LOGE(TAG, "Failed to open file for reading");
	}
	fgets(line, sizeof(line), foo);
	fclose(comp);
	// strip newline
	pos = strchr(line, '\n');
	if (pos) {
		*pos = '\0';
	}
	ESP_LOGI(TAG, "Read from file: '%s'", line);
	//Inflate
	recon = fopen("/spiflash/rec.bin", "wb");
	comp = fopen("/spiflash/c.bin", "rb");
	if(recon == NULL || comp == NULL){
		ESP_LOGE(TAG, "Error opening file before inflating.");
	}
	ESP_LOGI(TAG, "Start inflating");
	zerr(inf(comp, recon));
	ESP_LOGI(TAG, "Stopped inflating");
	if(fclose(comp) == EOF){
		ESP_LOGE(TAG, "Error closing file");
	}
	ESP_LOGI(TAG, "Closed comp");
	if(fclose(recon) == EOF){
		ESP_LOGE(TAG, "Error closing file");
	}
	ESP_LOGI(TAG, "Closed recon");
	// Open file for reading
	ESP_LOGI(TAG, "Reading file");
	recon = fopen("/spiflash/rec.bin", "rb");
	if (recon == NULL) {
		ESP_LOGE(TAG, "Failed to open file for reading");
	}
	fgets(line, sizeof(line), recon);
	fclose(recon);
	// strip newline
	pos = strchr(line, '\n');
	if (pos) {
		*pos = '\0';
	}
	ESP_LOGI(TAG, "Read from file: '%s'", line);
	// Open file for reading
	ESP_LOGI(TAG, "Reading file");
	foo = fopen("/spiflash/foo.bin", "rb");
	if (foo == NULL) {
		ESP_LOGE(TAG, "Failed to open file for reading");
		return;
	}
	fgets(line, sizeof(line), foo);
	fclose(foo);
	// strip newline
	pos = strchr(line, '\n');
	if (pos) {
		*pos = '\0';
	}
	ESP_LOGI(TAG, "Read from file: '%s'", line);
	// Unmount FATFS
	ESP_LOGI(TAG, "Unmounting FAT filesystem");
	ESP_ERROR_CHECK( esp_vfs_fat_spiflash_unmount(base_path, s_wl_handle));
	ESP_LOGI(TAG, "Done");
	while(1){
		vTaskDelay(1000);
	}
}
*/

 */


This is working for me. I am using zlib.
I remember there were some errors when initially compiling the code, but they could be resolved easily by simply reading the error messages.
Also note that compression.h has an example main.cpp file that works at the end.

Best,
Julian

User avatar
gunar.kroeger
Posts: 143
Joined: Fri Jul 27, 2018 6:48 pm

Re: Compression using miniz

Postby gunar.kroeger » Tue May 14, 2019 5:19 pm

I have miniz working as a component. Works great.
Had to change some defines to reduce RAM usage.

Code: Select all

    if(!mz_zip_writer_init_file(&zip_archive, tempNameWithDir, 0))
    {
        ESP_LOGE(TAG, "Error while initializing ZIP file archive\n");
        return false;
    }

    mz_bool rc;
    ESP_LOGD(TAG, "mz_zip_writer_add_file");
    ESP_LOGI(TAG, "Max alloc size = %d", heap_caps_get_largest_free_block(MALLOC_CAP_SPIRAM));
    printf("sizeof(tdefl_compressor) = %d\n", sizeof(tdefl_compressor));
    uint32_t millis = esp_log_timestamp();
    rc = mz_zip_writer_add_file(&zip_archive, fileName, fileNameWithDir, NULL, 0, MZ_DEFAULT_COMPRESSION);
    if (rc == MZ_FALSE)
    {
        ESP_LOGE(TAG, "Could not add file: %s", mz_zip_get_error_string(mz_zip_get_last_error(&zip_archive)));
        return false;
    }

    ESP_LOGD(TAG, "mz_zip_writer_finalize_archive");
    rc = mz_zip_writer_finalize_archive( &zip_archive );
    if ( !rc )
        return false;

    ESP_LOGD(TAG, "mz_zip_writer_end");
    rc = mz_zip_writer_end( &zip_archive );
    if ( !rc )
        return false;
Attachments
miniz.zip
(80.77 KiB) Downloaded 1159 times
"Running was invented in 1612 by Thomas Running when he tried to walk twice at the same time."

User avatar
fasani
Posts: 197
Joined: Wed Jan 30, 2019 12:00 pm
Location: Barcelona
Contact:

Re:Compression using brotli

Postby fasani » Wed May 15, 2019 6:07 pm

Thanks both Jubu and Gunar,
I found an alternative library reading an article in tech tutorials that is a bit complicated since involves some headers renaming and found time to make a fork of Brotli library for the ESP32 that works out if the box:
https://github.com/martinberlin/brotli

In tests folder I left an example to compile and try. Now that I have this great answers I would like to do some benchmarking since I’m sending animations to addressable Leds and time counts a lot.
Great thing is that brotli algorithm that at the beginning I though was broccoli since could not memorize the name compresses quite much more than gzip. A 1.4k sample of json was 85 bytes with gzip and about 57 in brotli
Last edited by fasani on Thu May 16, 2019 7:07 pm, edited 1 time in total.
epdiy collaborator | http://fasani.de Fan of Espressif MCUs and electronic design

jubueche
Posts: 32
Joined: Sat Jan 19, 2019 4:03 pm

Re: Compression using miniz

Postby jubueche » Thu May 16, 2019 7:36 am

That’s great news. I will check it out. I haven’t done any benchmarking yet.

Who is online

Users browsing this forum: No registered users and 64 guests