PPMd Compression

PPMd (Prediction by Partial Matching, variant D) is a compression method based on context modeling and statistical prediction. It provides excellent compression ratios for text and other highly structured data.

Method ID

ID: 03 04 01 (hex) The PPMd method is identified by the 3-byte sequence 03 04 01 in 7z archive format.

Overview

7-Zip includes two PPMd implementations:

PPMd Variant H (PPMd7)

Original PPMd algorithm from 2001

Based on Dmitry Shkarin’s PPMd var.H
Used in 7z archives
Order: 2-64

PPMd Variant I (PPMd8)

Improved PPMd algorithm from 2002

Based on Dmitry Shkarin’s PPMd var.I
Better compression ratio
Order: 2-16

PPMd is particularly effective for:

Text files (source code, logs, documents)
Structured data (CSV, JSON, XML)
Database dumps
Configuration files

For these file types, PPMd often achieves 5-20% better compression than LZMA.

How PPMd Works

PPMd uses Prediction by Partial Matching:

Context Modeling: Analyzes previous bytes to build statistical models
Probability Prediction: Predicts the probability of each possible next byte
Arithmetic Coding: Encodes bytes using their predicted probabilities
Adaptive Learning: Updates models as more data is processed

Context Orders

PPMd maintains multiple context models of different lengths (orders):

Order 0: Current byte independent of previous bytes
Order 1: Based on 1 previous byte
Order 2: Based on 2 previous bytes
...
Order N: Based on N previous bytes (max order)

Higher orders provide better compression for structured data but require more memory.

Escape Mechanism

When a byte is not found in the current context, PPMd uses an “escape” mechanism:

// From Ppmd7.c:15
static const Byte PPMD7_kExpEscape[16] = 
  { 25, 14, 9, 7, 5, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 2 };

This allows the algorithm to fall back to lower-order contexts when needed.

PPMd7 (Variant H)

PPMd var.H is the original PPMd implementation used in 7z archives.

Parameters

Order (maxOrder)

Type: unsigned
Range: PPMD7_MIN_ORDER (2) to PPMD7_MAX_ORDER (64)
Default: 6

// From Ppmd7.h:14-15
#define PPMD7_MIN_ORDER 2
#define PPMD7_MAX_ORDER 64

Maximum context length for prediction.

Low order (2-4): Faster, less memory, lower compression
Medium order (6-8): Balanced (recommended)
High order (16-64): Best compression for structured text

For most text files, order 6-8 provides the best balance. For highly structured data like source code, try order 10-16.

Memory Size (mem)

Type: UInt32
Range: PPMD7_MIN_MEM_SIZE (2 KB) to PPMD7_MAX_MEM_SIZE (~4 GB)
Default: 16 MB

// From Ppmd7.h:17-18
#define PPMD7_MIN_MEM_SIZE (1 << 11)
#define PPMD7_MAX_MEM_SIZE (0xFFFFFFFF - 12 * 3)

Amount of memory allocated for context models.Typical settings:

1-4 MB: Fast, lower compression
16-64 MB: Good balance (recommended)
128-256 MB: Maximum compression for large text files

Both encoder and decoder require the same amount of memory. Large memory settings may cause issues on memory-constrained systems.

Memory Requirements

PPMd7 memory usage:

Memory = mem_size + alignment + state_structures
       ≈ mem_size + 4 KB

Encoding: Same as memory size parameter
Decoding: Same as memory size parameter
Stack: Minimal (< 1 KB)

Example configurations:

Order 6, 16 MB: ~16 MB RAM (both compress/decompress)
Order 8, 64 MB: ~64 MB RAM (both compress/decompress)
Order 16, 256 MB: ~256 MB RAM (both compress/decompress)

API Usage

#include "Ppmd7.h"
#include "Ppmd7Enc.h"

// Allocate PPMd7
CPpmd7 ppmd;
Ppmd7_Construct(&ppmd);

UInt32 memSize = 16 << 20;  // 16 MB
unsigned maxOrder = 6;

if (!Ppmd7_Alloc(&ppmd, memSize, &g_Alloc))
  return SZ_ERROR_MEM;

// Initialize for encoding
Ppmd7_Init(&ppmd, maxOrder);
Ppmd7z_Init_RangeEnc(&ppmd);

// Encode data
Ppmd7z_EncodeSymbols(&ppmd, buf, buf + size);

// Flush encoder
Ppmd7z_Flush_RangeEnc(&ppmd);

// Free memory
Ppmd7_Free(&ppmd, &g_Alloc);

Decoding

#include "Ppmd7.h"
#include "Ppmd7Dec.h"

// Allocate and initialize decoder
CPpmd7 ppmd;
Ppmd7_Construct(&ppmd);
Ppmd7_Alloc(&ppmd, memSize, &g_Alloc);
Ppmd7_Init(&ppmd, maxOrder);
Ppmd7z_RangeDec_Init(&ppmd.rc.dec);

// Decode symbols
for (;;) {
  int sym = Ppmd7z_DecodeSymbol(&ppmd);
  if (sym < 0)
    break;  // PPMD7_SYM_END or PPMD7_SYM_ERROR
  *dest++ = (Byte)sym;
}

Ppmd7_Free(&ppmd, &g_Alloc);

PPMd8 (Variant I)

PPMd var.I is an improved version with better compression ratio.

Key Differences from PPMd7

Maximum Order Limit

PPMd8 has a lower maximum order:

// From Ppmd8.h:14-15
#define PPMD8_MIN_ORDER 2
#define PPMD8_MAX_ORDER 16

This is due to improved context modeling that provides better compression with lower orders.

Restore Method

PPMd8 supports different memory restoration methods:

// From Ppmd8.h:56-64
enum {
  PPMD8_RESTORE_METHOD_RESTART,      // Default
  PPMD8_RESTORE_METHOD_CUT_OFF       // Alternative
  // PPMD8_RESTORE_METHOD_FREEZE     // Disabled (compatibility)
};

RESTART: More aggressive memory recycling
CUT_OFF: Conservative memory management

FREEZE mode is disabled due to compatibility issues between PPMdI rev.1 and rev.2.

Improved Context Statistics

PPMd8 uses different statistical models:

// From Ppmd8.c:15-17
static const Byte PPMD8_kExpEscape[16] = 
  { 25, 14, 9, 7, 5, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 2 };
static const UInt16 PPMD8_kInitBinEsc[] = 
  { 0x3CDD, 0x1F3F, 0x59BF, 0x48F3, 0x64A1, 0x5ABC, 0x6632, 0x6051 };

These provide better probability estimation for certain data patterns.

API Usage

#include "Ppmd8.h"
#include "Ppmd8Enc.h"

// Allocate PPMd8
CPpmd8 ppmd;
Ppmd8_Construct(&ppmd);

UInt32 memSize = 16 << 20;  // 16 MB
unsigned maxOrder = 6;
unsigned restoreMethod = PPMD8_RESTORE_METHOD_RESTART;

Ppmd8_Alloc(&ppmd, memSize, &g_Alloc);

// Initialize
Ppmd8_Init(&ppmd, maxOrder, restoreMethod);
Ppmd8_Init_RangeEnc(&ppmd);

// Encode
for (size_t i = 0; i < size; i++)
  Ppmd8_EncodeSymbol(&ppmd, buf[i]);

Ppmd8_Flush_RangeEnc(&ppmd);
Ppmd8_Free(&ppmd, &g_Alloc);

Performance Characteristics

Typical performance on modern hardware (Intel Core i7, 3.5 GHz):Compression:

Order 6, 16 MB: ~1-2 MB/s
Order 8, 64 MB: ~0.5-1 MB/s
Order 16, 256 MB: ~0.3-0.5 MB/s

Decompression:

Order 6: ~5-10 MB/s
Order 8: ~3-7 MB/s
Order 16: ~2-5 MB/s

Compression ratio (vs LZMA):

Text files: 5-20% better
Source code: 10-25% better
Binary executables: Similar or worse
Multimedia: Much worse (use LZMA)

Compression Ratio Comparison

Text Files (Source Code, Logs)

Method	Compression Ratio	Speed
PPMd (order 8)	100% (best)	Slow
PPMd (order 6)	102-105%	Medium
LZMA2	110-120%	Medium
BZip2	115-125%	Fast
Deflate	130-150%	Very Fast

For large text archives, PPMd can save 10-20% space compared to LZMA.

Structured Data (XML, JSON, CSV)

Method	Compression Ratio	Speed
PPMd (order 10)	100% (best)	Very Slow
PPMd (order 6)	103-107%	Slow
LZMA2	115-130%	Medium
BZip2	120-135%	Fast
Deflate	140-160%	Very Fast

Binary Executables

Method	Compression Ratio	Speed
LZMA2	100% (best)	Medium
PPMd	105-115%	Slow
BZip2	110-120%	Fast
Deflate	120-140%	Very Fast

PPMd is not recommended for binary executables. Use LZMA2 with BCJ filter instead.

Command Line Usage

# Compress with PPMd (default: order 6, 16 MB)
7z a -m0=PPMd archive.7z file.txt

# Set order and memory size
7z a -m0=PPMd -mo=8 -mmem=64m archive.7z file.txt

# Maximum compression for text
7z a -m0=PPMd -mo=16 -mmem=256m archive.7z documents/

# Fast PPMd compression
7z a -m0=PPMd -mo=4 -mmem=8m archive.7z file.txt

# Compare with LZMA
7z a -m0=LZMA2 lzma.7z file.txt
7z a -m0=PPMd -mo=8 -mmem=64m ppmd.7z file.txt

Recommended settings by file size:

Small files (< 1 MB): -mo=4 -mmem=4m
Medium files (1-10 MB): -mo=6 -mmem=16m
Large files (10-100 MB): -mo=8 -mmem=64m
Huge files (> 100 MB): -mo=10 -mmem=128m

Best Practices

Choosing Order

Higher order = better compression but slower and more memory:

Start with order 6 (default)
For highly structured data, try order 8-10
For maximum compression, test up to order 16
Monitor compression time and memory usage

# Test different orders
for order in 4 6 8 10 12; do
  7z a -m0=PPMd -mo=$order -mmem=64m test-$order.7z file.txt
done

Memory Size Selection

Memory should be:

At least 1-2x the file size for small files
10-20% of file size for large files
Minimum 4 MB for order 6+
Maximum what both encoder and decoder can allocate

Decoder needs the same memory as encoder. Don’t use 256 MB if decoder only has 64 MB available.

When NOT to Use PPMd

Avoid PPMd for:

Already compressed data: JPEG, PNG, MP3, video files
Binary executables: Use LZMA2 with BCJ filter
Random data: Any method will fail
Memory-constrained systems: Use LZMA2 or Deflate
Need for speed: Use LZMA2 with multithreading

# Wrong: PPMd for multimedia
7z a -m0=PPMd archive.7z photos/*.jpg  # Will be larger!

# Right: Store only or use LZMA2
7z a -m0=Copy archive.7z photos/*.jpg
7z a archive.7z photos/*.jpg  # Auto-detects, uses LZMA2

Error Codes

Symbol	Value	Description
`PPMD7_SYM_END`	-1	End of payload marker
`PPMD7_SYM_ERROR`	-2	Data corruption error
`PPMD8_SYM_END`	-1	End of payload marker
`PPMD8_SYM_ERROR`	-2	Data corruption error

PPMd7 vs PPMd8

Feature	PPMd7 (var.H)	PPMd8 (var.I)
Year	2001	2002
Max Order	64	16
Compression	Excellent	Slightly better
Speed	Medium	Slightly slower
Memory	Same as setting	Same as setting
Restore method	Simple	Multiple options
7z default	Yes	No
Compatibility	Wider	Limited

For most use cases, PPMd7 (var.H) is recommended as it’s the standard PPMd implementation in 7z archives.

Getting Started

Command Reference

Compression Methods

Archive Formats

Advanced Usage

PPMd Compression

Method ID

Overview

PPMd Variant H (PPMd7)

PPMd Variant I (PPMd8)

How PPMd Works

PPMd7 (Variant H)

Parameters

Memory Requirements

API Usage

Decoding

PPMd8 (Variant I)

Key Differences from PPMd7

API Usage

Performance Characteristics

Compression Ratio Comparison

Command Line Usage

Best Practices

Error Codes

PPMd7 vs PPMd8

See Also

Build docs developers (and LLMs) love

Getting Started

Command Reference

Compression Methods

Archive Formats

Advanced Usage

​Method ID

​Overview

PPMd Variant H (PPMd7)

PPMd Variant I (PPMd8)

​How PPMd Works

​PPMd7 (Variant H)

​Parameters

​Memory Requirements

​API Usage

​Decoding

​PPMd8 (Variant I)

​Key Differences from PPMd7

​API Usage

​Performance Characteristics

​Compression Ratio Comparison

​Command Line Usage

​Best Practices

​Error Codes

​PPMd7 vs PPMd8

​See Also

Build docs developers (and LLMs) love

Method ID

Overview

How PPMd Works

PPMd7 (Variant H)

Parameters

Memory Requirements

API Usage

Decoding

PPMd8 (Variant I)

Key Differences from PPMd7

API Usage

Performance Characteristics

Compression Ratio Comparison

Command Line Usage

Best Practices

Error Codes

PPMd7 vs PPMd8

See Also