GY-63_MS5611/libraries/Adler
2022-06-16 09:46:28 +02:00
..
.github/workflows 0.1.0 Adler 2022-01-27 12:18:16 +01:00
examples 0.2.0 Adler 2022-06-16 09:46:28 +02:00
test 0.2.0 Adler 2022-06-16 09:46:28 +02:00
.arduino-ci.yml 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler16.cpp 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler16.h 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler32.cpp 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler32.h 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler.cpp 0.2.0 Adler 2022-06-16 09:46:28 +02:00
Adler.h 0.2.0 Adler 2022-06-16 09:46:28 +02:00
CHANGELOG.md 0.2.0 Adler 2022-06-16 09:46:28 +02:00
keywords.txt 0.2.0 Adler 2022-06-16 09:46:28 +02:00
library.json 0.2.0 Adler 2022-06-16 09:46:28 +02:00
library.properties 0.2.0 Adler 2022-06-16 09:46:28 +02:00
LICENSE 0.1.0 Adler 2022-01-27 12:18:16 +01:00
README.md 0.2.0 Adler 2022-06-16 09:46:28 +02:00

Arduino CI Arduino-lint JSON check License: MIT GitHub release

Adler

Arduino Library for Adler-32 and experimental Adler-16 checksum.

Description

This library provides a Adler32 checksum of a data array. Furthermore since 0.2.0 an experimental Adler-16 implementation is added. This one is often faster as it uses a smaller checksum than the Adler32, and the price is that it is less sensitive than the Adler32. Still it might have its niches where it will be useful.

Relates to https://github.com/RobTillaart/CRC

Relates to https://github.com/RobTillaart/Fletcher

Tested on Arduino UNO only.

0.2.0 is a breaking change, file names have been changed to be more in line with the CRC library.

  • Adler.h for the static functions
  • Adler32.h for the Adler32 class
  • Adler16.h for the Adler16 class.

Interface

Adler class

Use #include "Adler32.h" or #include "Adler16.h"

The interface for the Adler16 is very similar.

  • Adler32() Constructor, initializes internals.
  • void begin(uint8_t s1 = 1, uint8_t s2 = 0) resets the internals. optional setting start values for s1 and s2. Note this is not part of the standard. This allows a restart from a specific index in a buffer.
  • void add(uint8_t value) add a single value to the checksum.
  • void add(const uint8_t * array, uint8_t length) add an array of values to the checksum.
  • void addFast(const uint8_t * array, uint8_t length) add an array of values to the checksum. Is faster by trading PROGMEM for performance.
  • uint32_t getAdler() get the current checksum.
  • uint32_t count() get the number of items added. Merely a debugging feature, can overflow without affecting checksum.

The class is typically used for streaming very large blocks of data, optional with intermediate checksum tests.

Performance Adler32

Not tested ESP32 (and many other platforms) yet.

Numbers measured with Adler32_performance.ino.

add(value)

The add(value) adds one byte and does a subtraction instead of a modulo.

Version Function UNO 16 MHz ESP32 240 MHz
0.1.0 add 5.6 us
0.1.2 add 6.6 us
0.2.0 add 5.9 us

add(lorem) 868 chars

The add(array, length) is a straightforward loop over the array and has a small footprint.

Version Function UNO 16 MHz ESP32 240 MHz
0.1.0 add
0.1.2 add 6392 us
0.2.0 add 5748 us

Note: add() is about 6.6 us per byte.

addFast(lorem) 868 chars

The addFast(array, length) is faster than the reference add(array, length) and uses 108 bytes more. So the function has a larger footprint. Depending on your needs, choose performance or footprint.

See Adler32_performance_addFast.ino

Version Function UNO 16 MHz ESP32 240 MHz
0.1.0 addFast
0.1.2 addFast 1348 us
0.2.0 addFast 1348 us

Note: addFast() is less than 2 us per byte.

Performance Adler16

Not tested ESP32 (and many other platforms) yet.

Numbers measured with Adler16_performance.ino.

add(value)

The add(value) adds one byte and does a subtraction instead of a modulo.

Version Function UNO 16 MHz ESP32 240 MHz
0.2.0 add 4.0 us

The per byte performance of the Adler16 (on UNO) is faster than the Adler32 add(value). The reason is that a 16 bit subtraction on an UNO is faster than a 32 bit subtraction.

add(lorem) 868 chars

The add(array, length) is a straightforward loop over the array and has a small footprint.

Version Function UNO 16 MHz ESP32 240 MHz
0.2.0 add 4040 us

Note: add() is about 6.6 us per byte.

addFast(lorem) 868 chars

The addFast(array, length) is faster than the reference add(array, length).

Version Function UNO 16 MHz ESP32 240 MHz
0.2.0 addFast 1968 us

The gain of the faster 16 bit modulo meets the frequency of doing the modulo more often.

Interface static functions

The functions are straightforward.

Use #include "Adler.h"

  • uint32_t adler32(uint8_t *data, uint16_t length) length in units of 1 byte = 8 bits.
  • uint16_t adler16(uint8_t *data, uint16_t length) length in units of 1 byte = 8 bits.

The functions are typically used for an in memory buffer to calculate the checksum once. Think of packets in a network, records in a database, or a checksum for an configuration in EEPROM.

Performance

Not tested ESP32 (and many other platforms) yet.

Numbers measured with Adler_performance.ino.

Lorem Ipsum text = 868 bytes.

Version Function UNO 16 MHz ESP32 240 MHz
0.1.0 Adler32 1116 us
0.1.2 Adler32 1116 us
0.2.0 Adler32 1116 us
0.2.0 Adler16 1736 us

Adler32 average 1116 / 868 = 1.29 us per byte. Adler16 average 1736 / 868 = 2.00 us per byte. (~1.5x slower !)

Adler16 does more often the modulo math as it reaches halfway uint16_t faster than Adler32 reaches halfway uint32_t.

As the Adler16 is less performant as the Adler32, it is often the best to use the 32 bit version.

Operation

See examples.

Future

  • return values for add(array) and addFast(array)
    • updated checksum?
    • not for add(value) as that would create quite some overhead.
  • Adler64 ?
    • would need a large prime (which)

Wont

  • do the string wrappers need strlen() ? parameter.
    • yes, as string can be processed partially.