tamp 1.11.1

No project description provided

---

Documentation: https://tamp.readthedocs.io/en/latest/

Source Code: https://github.com/BrianPugh/tamp

Online Demo: https://brianpugh.github.io/tamp

Tamp is a low-memory, DEFLATE-inspired lossless compression library optimized for embedded and resource-constrained environments.

Tamp delivers the highest data compression ratios, while using the least amount of RAM and firmware storage.

Features

• Various language implementations available:
  • Pure Python reference:
    • tamp/__init__.py, tamp/compressor.py, tamp/decompressor.py
    • pip install tamp will use a python-bound C implementation optimized for speed.
  • Micropython:
    • Native Module (suggested micropython implementation).
      • mpy_bindings/
    • Viper.
      • tamp/__init__.py, tamp/compressor_viper.py, tamp/decompressor_viper.py
  • C library:
    • tamp/_c_src/
  • Javascript/Typescript via Emscripten WASM.
    • wasm/
  • Unofficial rust bindings.
    • See documentation here.
• High compression ratios, low memory use, and fast.
• Compact compression and decompression implementations.
  • Compiled C library is <4KB (compressor + decompressor).
• Mid-stream flushing.
  • Allows for submission of messages while continuing to compress subsequent data.
• Customizable dictionary for greater compression of small messages.
• Convenient CLI interface.

Installation

Tamp contains 4 implementations:

1. A reference desktop CPython implementation that is optimized for readability (and not speed).
2. A Micropython Native Module implementation (fast).
3. A Micropython Viper implementation (not recommended, please use Native Module).
4. A C implementation (with python bindings) for accelerated desktop use and to be used in C projects (very fast).

This section instructs how to install each implementation.

Desktop Python

The Tamp library and CLI requires Python >=3.9 and can be installed via:

pip install tamp

MicroPython

MicroPython Native Module

Tamp provides pre-compiled [native modules]{.title-ref} that are easy to install, are small, and are incredibly fast.

Download the appropriate .mpy file from the release page.

• Match the micropython version.
• Match the architecture to the microcontroller (e.g. armv6m for a pi pico).

Rename the file to tamp.mpy and transfer it to your board. If using Belay, tamp can be installed by adding the following to pyproject.toml.

[tool.belay.dependencies]
tamp = "https://github.com/BrianPugh/tamp/releases/download/v1.7.0/tamp-1.7.0-mpy1.23-armv6m.mpy"

MicroPython Viper

NOT RECOMMENDED, PLEASE USE NATIVE MODULE

For micropython use, there are 3 main files:

1. tamp/__init__.py - Always required.
2. tamp/decompressor_viper.py - Required for on-device decompression.
3. tamp/compressor_viper.py - Required for on-device compression.

For example, if on-device decompression isn't used, then do not include decompressor_viper.py. If manually installing, just copy these files to your microcontroller's /lib/tamp folder.

If using mip, tamp can be installed by specifying the appropriate package-*.json file.

mip install github:brianpugh/tamp  # Defaults to package.json: Compressor & Decompressor
mip install github:brianpugh/tamp/package-compressor.json  # Compressor only
mip install github:brianpugh/tamp/package-decompressor.json  # Decompressor only

If using Belay, tamp can be installed by adding the following to pyproject.toml.

[tool.belay.dependencies]
tamp = [
   "https://github.com/BrianPugh/tamp/blob/main/tamp/__init__.py",
   "https://github.com/BrianPugh/tamp/blob/main/tamp/compressor_viper.py",
   "https://github.com/BrianPugh/tamp/blob/main/tamp/decompressor_viper.py",
]

C

Copy the tamp/_c_src/tamp folder into your project. For more information, see the documentation.

Usage

Tamp works on desktop python and micropython. On desktop, Tamp is bundled with the tamp command line tool for compressing and decompressing tamp files.

CLI

Compression

Use tamp compress to compress a file or stream. If no input file is specified, data from stdin will be read. If no output is specified, the compressed output stream will be written to stdout.

$ tamp compress --help
Usage: tamp compress [ARGS] [OPTIONS]

Compress an input file or stream.

╭─ Parameters ───────────────────────────────────────────────────────────────────────────────╮
│ INPUT,--input    -i  Input file to compress. Defaults to stdin.                            │
│ OUTPUT,--output  -o  Output compressed file. Defaults to stdout.                           │
│ --window         -w  Number of bits used to represent the dictionary window. [default: 10] │
│ --literal        -l  Number of bits used to represent a literal. [default: 8]              │
╰────────────────────────────────────────────────────────────────────────────────────────────╯

Example usage:

tamp compress enwik8 -o enwik8.tamp  # Compress a file
echo "hello world" | tamp compress | wc -c  # Compress a stream and print the compressed size.

The following options can impact compression ratios and memory usage:

• window - 2^window plaintext bytes to look back to try and find a pattern. A larger window size will increase the chance of finding a longer pattern match, but will use more memory, increase compression time, and cause each pattern-token to take up more space. Try smaller window values if compressing highly repetitive data, or short messages.
• literal - Number of bits used in each plaintext byte. For example, if all input data is 7-bit ASCII, then setting this to 7 will improve literal compression ratios by 11.1%. The default, 8-bits, can encode any binary data.

Decompression

Use tamp decompress to decompress a file or stream. If no input file is specified, data from stdin will be read. If no output is specified, the compressed output stream will be written to stdout.

$ tamp decompress --help
Usage: tamp decompress [ARGS] [OPTIONS]

Decompress an input file or stream.

╭─ Parameters ───────────────────────────────────────────────────────────────────────────────╮
│ INPUT,--input    -i  Input file to decompress. Defaults to stdin.                          │
│ OUTPUT,--output  -o  Output decompressed file. Defaults to stdout.                         │
╰────────────────────────────────────────────────────────────────────────────────────────────╯

Example usage:

tamp decompress enwik8.tamp -o enwik8
echo "hello world" | tamp compress | tamp decompress

Python

The python library can perform one-shot compression, as well as operate on files/streams.

import tamp

# One-shot compression
string = b"I scream, you scream, we all scream for ice cream."
compressed_data = tamp.compress(string)
reconstructed = tamp.decompress(compressed_data)
assert reconstructed == string

# Streaming compression
with tamp.open("output.tamp", "wb") as f:
    for _ in range(10):
        f.write(string)

# Streaming decompression
with tamp.open("output.tamp", "rb") as f:
    reconstructed = f.read()

Benchmark

In the following section, we compare Tamp against:

• zlib, a python builtin gzip-compatible DEFLATE compression library.
• heatshrink, a data compression library for embedded/real-time systems. Heatshrink has similar goals as Tamp.

All of these are LZ-based compression algorithms, and tests were performed using a 1KB (10 bit) window. Since zlib already uses significantly more memory by default, the lowest memory level (memLevel=1) was used in these benchmarks. It should be noted that higher zlib memory levels will having greater compression ratios than Tamp. Currently, there is no micropython-compatible zlib or heatshrink compression implementation, so these numbers are provided simply as a reference.

Compression Ratio

The following table shows compression algorithm performance over a variety of input data sourced from the Silesia Corpus and Enwik8. This should give a general idea of how these algorithms perform over a variety of input data types.

dataset	raw	tamp	tamp (LazyMatching)	zlib	heatshrink
enwik8	100,000,000	51,635,633	51,252,113	56,205,166	56,110,394
silesia/dickens	10,192,446	5,546,761	5,511,604	6,049,169	6,155,768
silesia/mozilla	51,220,480	25,121,385	24,936,067	25,104,966	25,435,908
silesia/mr	9,970,564	5,027,032	4,886,272	4,864,734	5,442,180
silesia/nci	33,553,445	8,643,610	8,645,299	5,765,521	8,247,487
silesia/ooffice	6,152,192	3,814,938	3,798,261	4,077,277	3,994,589
silesia/osdb	10,085,684	8,520,835	8,506,443	8,625,159	8,747,527
silesia/reymont	6,627,202	2,847,981	2,820,870	2,897,661	2,910,251
silesia/samba	21,606,400	9,102,594	9,060,692	8,862,423	9,223,827
silesia/sao	7,251,944	6,137,755	6,101,744	6,506,417	6,400,926
silesia/webster	41,458,703	18,694,172	18,567,288	20,212,235	19,942,817
silesia/x-ray	8,474,240	7,510,606	7,405,814	7,351,750	8,059,723
silesia/xml	5,345,280	1,681,687	1,672,660	1,586,985	1,665,179

Tamp usually out-performs heatshrink, and is generally very competitive with zlib. While trying to be an apples-to-apples comparison, zlib still uses significantly more memory during both compression and decompression (see next section). Tamp accomplishes competitive performance while using around 10x less memory.

Lazy Matching is a simple technique to improve compression ratios at the expense of CPU while requiring very little code. One can expect 50-75% more CPU usage for modest compression gains (around 0.5 - 2.0%). Because of this poor trade-off, it is disabled by default; however, in applications where we want to compress once on a powerful machine (like a desktop/server) and decompress on an embedded device, it may be worth it to spend a bit more compute. Lazy matched compressed data is the exact same format; it appears no different to the tamp decoder.

One might wonder "Why did Tamp perform so much worse than zlib on the nci dataset?" The nci dataset contains highly compressible data with long patterns. For example, the following 49-character text appears repeatedly in the dataset:

    0.0000 H   0  0  0  0  0  0  0  0  0  0  0  0

Tamp's maximum pattern length peaks at around 15 characters, meaning that these 49 characters has to be compressed as 4 pattern-matches. Zlib can handle patterns with a maximum length of 258, meaning that it can encode this highly repeating data more efficiently. Given Tamp's excellent performance in most of the other data compression benchmark files, this is a good tradeoff for most real-world scenarios.

Memory Usage

The following table shows approximately how much memory each algorithm uses during compression and decompression.

	Compression	Decompression
Tamp	(1 << windowBits)	(1 << windowBits)
ZLib	(1 << (windowBits + 2)) + 7KB	(1 << windowBits) + 7KB
Heatshrink	(1 << (windowBits + 1))	(1 << (windowBits + 1))
Deflate (micropython)	(1 << windowBits)	(1 << windowBits)

All libraries have a few dozen bytes of overhead in addition to the primary window buffer, but are implementation-specific and ignored for clarity here. Tamp uses significantly less memory than ZLib, and half the memory of Heatshrink.

Runtime

As a rough benchmark, here is the performance (in seconds) of these different compression algorithms on the 100MB enwik8 dataset. These tests were performed on an M3 Macbook Air.

	Compression (s)	Decompression (s)
Tamp (Pure Python Reference)	136.2	105.0
Tamp (C bindings)	5.56	0.544
ZLib	3.65	0.578
Heatshrink (with index)	4.42	0.67
Heatshrink (without index)	27.40	0.67

Heatshrink v0.4.1 was used in these benchmarks. When heathshrink uses an index, an additional (1 << (windowBits + 1)) bytes of memory are used, resulting in 4x more memory-usage than Tamp. Tamp could use a similar indexing to increase compression speed, but has chosen not to to focus on the primary goal of a low-memory compressor.

To give an idea of Tamp's speed on an embedded device, the following table shows compression/decompression in bytes/second of the first 100KB of enwik8 on a pi pico (rp2040) at the default 125MHz clock rate. The C benchmark does not use a filesystem nor dynamic memory allocation, so it represents the maximum speed Tamp can achieve. In all tests, a 1KB window (10 bit) was used.

	Compression (bytes/s)	Decompression (bytes/s)
Tamp (MicroPython Viper)	4,300	42,000
Tamp (Micropython Native Module)	31,192	1,086,957
Tamp (C)	36,127	1,400,600
Deflate (micropython builtin)	6,885	294,985

Tamp resulted in a 51637 byte archive, while Micropython's builtin deflate resulted in a larger, 59442 byte archive.

Binary Size

To give an idea on the resulting binary sizes, Tamp and other libraries were compiled for the Pi Pico (armv6m). All libraries were compiled with -O3. Numbers reported in bytes. Tamp sizes were measured using arm-none-eabi-gcc 15.2.1 and MicroPython v1.27, and can be regenerated with make binary-size.

	Compressor	Decompressor	Compressor + Decompressor
Tamp (MicroPython Viper)	4676	4372	7917
Tamp (MicroPython Native)	3896	3559	6616
Tamp (C, -DTAMP_STREAM=0)	2028	1992	3900
Tamp (C)	2472	2444	4796
Heatshrink (C)	2956	3876	6832
uzlib (C)	2355	3963	6318

Tamp C includes a high-level stream API by default. Even with -DTAMP_STREAM=0, Tamp includes buffer-looping functions (like tamp_compressor_compress) that Heatshrink lacks (Heatshrink only provides poll/sink primitives). In an apples-to-apples comparison, Tamp would be even smaller.

Acknowledgement

• Thanks @BitsForPeople for the esp32-optimized compressor implementation.