implement optional parity contexts

unpack_armv6m: update comment headers and remove unneeded pushed register

asm_unpackers/.gitignore

@@ -0,0 +1 @@
+/build/

asm_unpackers/Makefile

@@ -0,0 +1,30 @@
+build/unpack_riscv64: ../c_unpacker/main.c ../c_unpacker/unpack.c
+	mkdir -p build
+	riscv64-linux-gnu-gcc -g -static -o $@ $^
+
+test_riscv64: build/unpack_riscv64
+	qemu-riscv64 $< test_data.upk /tmp/out.bin
+	cmp test_data.bin /tmp/out.bin
+
+build/unpack_armv6m: ../c_unpacker/main.c unpack_armv6m.S
+	mkdir -p build
+	arm-linux-gnueabihf-gcc -g -static -o $@ $^
+
+test_armv6m: build/unpack_armv6m
+	qemu-arm $< test_data.upk /tmp/out.bin
+	cmp test_data.bin /tmp/out.bin
+
+build/unpack_armv6m.bin: unpack_armv6m.S
+	arm-none-eabi-gcc -march=armv6-m -c -o build/unpack_armv6m.o $?
+	arm-none-eabi-objcopy -O binary --only-section=.text build/unpack_armv6m.o $@
+	
+build/unpack_c: ../c_unpacker/main.c ../c_unpacker/unpack.c
+	mkdir -p build
+	gcc -g -o $@ $^
+
+test_c: build/unpack_c
+	$< test_data.upk /tmp/out.bin
+	cmp test_data.bin /tmp/out.bin
+
+sizes: build/unpack_armv6m.bin
+	ls -l build/*.bin

asm_unpackers/test_data.bin

@@ -0,0 +1,99 @@
+typedef unsigned char u8;
+typedef unsigned short u16;
+typedef unsigned long u32;
+
+u8* upkr_data_ptr;
+u8 upkr_probs[1 + 255 + 1 + 2*32 + 2*32]; 
+#ifdef UPKR_BITSTREAM
+u16 upkr_state;
+u8 upkr_current_byte;
+int upkr_bits_left;
+#else
+u32 upkr_state;
+#endif
+
+int upkr_decode_bit(int context_index) {
+#ifdef UPKR_BITSTREAM
+    while(upkr_state < 32768) {
+        if(upkr_bits_left == 0) {
+            upkr_current_byte = *upkr_data_ptr++;
+            upkr_bits_left = 8;
+        }
+        upkr_state = (upkr_state << 1) + (upkr_current_byte & 1);
+        upkr_current_byte >>= 1;
+        --upkr_bits_left;
+    }
+#else
+    while(upkr_state < 4096) {
+        upkr_state = (upkr_state << 8) | *upkr_data_ptr++;
+    }
+#endif
+   
+    int prob = upkr_probs[context_index];
+    int bit = (upkr_state & 255) < prob ? 1 : 0;
+    
+    int tmp = prob;
+    if(!bit) {
+        tmp = 256 - tmp;
+    }
+    upkr_state = tmp * (upkr_state >> 8) + (upkr_state & 255);
+    tmp += (256 - tmp + 8) >> 4;
+    if(!bit) {
+        upkr_state -= prob;
+        tmp = 256 - tmp;
+    }
+    upkr_probs[context_index] = tmp;
+
+    return bit;
+}
+
+int upkr_decode_length(int context_index) {
+    int length = 0;
+    int bit_pos = 0;
+    while(upkr_decode_bit(context_index)) {
+        length |= upkr_decode_bit(context_index + 1) << bit_pos++;
+        context_index += 2;
+    }
+    return length | (1 << bit_pos);
+}
+
+void* upkr_unpack(void* destination, void* compressed_data) {
+    upkr_data_ptr = (u8*)compressed_data;
+    upkr_state = 0;
+#ifdef UPKR_BITSTREAM
+    upkr_bits_left = 0;
+#endif
+    for(int i = 0; i < sizeof(upkr_probs); ++i)
+        upkr_probs[i] = 128;
+    
+    u8* write_ptr = (u8*)destination;
+    
+    int prev_was_match = 0;
+    int offset = 0;
+    for(;;) {
+        if(upkr_decode_bit(0)) {
+            if(prev_was_match || upkr_decode_bit(256)) {
+                offset = upkr_decode_length(257) - 1;
+                if(offset == 0) {
+                    break;
+                }
+            }
+            int length = upkr_decode_length(257 + 64);
+            while(length--) {
+                *write_ptr = write_ptr[-offset];
+                ++write_ptr;
+            }
+            prev_was_match = 1;
+        } else {
+            int byte = 1;
+            while(byte < 256) {
+                int bit = upkr_decode_bit(byte);
+                byte = (byte << 1) + bit;
+            }
+            *write_ptr++ = byte;
+            prev_was_match = 0;
+        }
+    }
+    
+    return write_ptr;
+}

asm_unpackers/unpack_armv6m.S

@@ -0,0 +1,162 @@
+// armv6-m upkr unpacker by yrlf
+// some optimizations by exoticorn
+
+.syntax unified
+.thumb
+
+.section .text
+
+#define ALIGNUP(n, align) (((n) + (align) - 1) & ~((align) - 1))
+#define PROB_LEN (1 + 255 + 1 + 2*32 + 2*32)
+#define FRAME_SIZE ALIGNUP(PROB_LEN, 4)
+
+// auto upkr_unpack(uint8_t * out, uint8_t * in) -> tuple<uint8_t *, uint8_t *>
+.global upkr_unpack
+.type upkr_unpack, %function
+// r0 .. out_ptr (returned)
+// r1 .. in_ptr (returned)
+// r2 .. state
+// r3 .. offset
+// r4 .. prev_was_literal / decode_length ret
+// r5 .. subroutine arg (preserved)
+// r6 .. decode_bit ret
+// r7 .. probs ptr
+upkr_unpack:
+    push { r4, r5, r6, r7, lr }
+    sub sp, sp, #FRAME_SIZE
+
+    mov r7, sp
+    movs r2, #255
+    adds r2, r2, #(PROB_LEN - 255)
+    movs r3, #128
+.Lclear:
+    subs r2, r2, #1
+    strb r3, [r7, r2]
+    bne .Lclear
+
+.Lloop:
+    movs r5, #0
+    bl upkr_decode_bit
+    beq .Ldata
+.Lmatch:
+    // r6 = 1
+    lsls r5, r6, #8
+    cmp r4, #0
+    beq 1f
+
+    bl upkr_decode_bit
+    beq 2f
+
+1:
+    bl upkr_decode_length
+    adds r3, r4, #1
+    beq .Lend
+2:
+
+    adds r5, r5, #64
+    bl upkr_decode_length
+.Lcopy_loop:
+    ldrb r5, [r0, r3]
+.Lstore:
+    strb r5, [r0]
+    adds r0, r0, #1
+    adds r4, r4, #1
+    blt .Lcopy_loop
+    b .Lloop
+
+.Ldata:
+    movs r5, #1
+
+.Ldata_loop:
+    bl upkr_decode_bit
+    adcs r5, r5, r5
+    lsrs r4, r5, #8
+    beq .Ldata_loop
+    b .Lstore
+
+.Lend:
+    add sp, sp, #FRAME_SIZE
+    pop { r4, r5, r6, r7, pc }
+
+.type upkr_decode_length, %function
+// r0 .. -length tmp (saved)
+// r1 ..
+// r2 ..
+// r3 ..
+// r4 .. -length (returned)
+// r5 .. context index (saved)
+// r6 .. (saved)
+// r7 ..
+upkr_decode_length:
+    push { r0, r5, r6, lr }
+
+    movs r0, #0
+    subs r4, r0, #1
+.Lbit_loop:
+    adds r5, r5, #1
+    bl upkr_decode_bit
+    beq 1f
+
+    adds r5, r5, #1
+    bl upkr_decode_bit
+    beq 2f
+    adds r0, r0, r4
+2:
+    lsls r4, r4, #1
+    b .Lbit_loop
+1:
+    adds r4, r4, r0
+
+    pop { r0, r5, r6, pc }
+
+.type upkr_decode_bit, %function
+// r0 .. tmp / prob (saved)
+// r1 .. in_ptr (modified)
+// r2 .. state (modified)
+// r3 .. scratch (saved)
+// r4 ..
+// r5 .. context index (preserved)
+// r6 .. bit (returned)
+// r7 .. probs ptr (preserved)
+upkr_fill_state:
+    lsls r2, r2, #8
+    ldrb r6, [r1]
+    adds r1, r1, #1
+    orrs r2, r2, r6
+
+upkr_decode_bit:
+    lsrs r6, r2, #12
+    beq upkr_fill_state
+
+    push { r0, r1, r3, lr }
+
+    ldrb r0, [r7, r5]
+
+    lsrs r3, r2, #8
+    uxtb r1, r2
+
+    subs r6, r1, r0
+    blt 1f
+
+    subs r1, r2, r0
+    rsbs r0, r0, #0
+1:
+
+    muls r3, r3, r0
+    adds r2, r1, r3
+
+    rsbs r3, r0, #0
+    uxtb r3, r3
+    lsrs r3, r3, #4
+    adcs r0, r0, r3
+
+    cmp r6, #0
+    blt 1f
+
+    rsbs r0, r0, #0
+1:
+
+    strb r0, [r7, r5]
+
+    lsrs r6, r6, #31
+    pop { r0, r1, r3, pc }

c_unpacker/.gitignore

@@ -0,0 +1,5 @@
+unpack
+unpack_bitstream
+unpack_debug
+*.upk
+

c_unpacker/Makefile

@@ -0,0 +1,10 @@
+all: unpack unpack_bitstream
+
+unpack: main.c unpack.c
+	cc -O2 -o unpack main.c unpack.c
+
+unpack_bitstream: main.c unpack.c
+	cc -O2 -D UPKR_BITSTREAM -o unpack_bitstream main.c unpack.c
+	
+unpack_debug: main.c unpack.c
+	cc -g -o unpack_debug main.c unpack.c

c_unpacker/decode_bit_alt.c

@@ -0,0 +1,33 @@
+int upkr_decode_bit(int context_index) {
+#ifdef UPKR_BITSTREAM
+    while(upkr_state < 32768) {
+        if(upkr_bits_left == 0) {
+            upkr_current_byte = *upkr_data_ptr++;
+            upkr_bits_left = 8;
+        }
+        upkr_state = (upkr_state << 1) + (upkr_current_byte & 1);
+        upkr_current_byte >>= 1;
+        --upkr_bits_left;
+    }
+#else
+    while(upkr_state < 4096) {
+        upkr_state = (upkr_state << 8) | *upkr_data_ptr++;
+    }
+#endif
+   
+    int prob = upkr_probs[context_index];
+    int bit = (upkr_state & 255) < prob ? 1 : 0;
+    
+    if(bit) {
+        prob = 256 - prob;
+    }
+    upkr_state -= prob * ((upkr_state >> 8) + (bit ^ 1));
+    prob -= (prob + 8) >> 4;
+    if(bit) {
+        prob = -prob;
+    }
+    upkr_probs[context_index] = prob;
+
+    return bit;
+}
+

c_unpacker/main.c

@@ -0,0 +1,26 @@
+#include <stdio.h>
+#include <stdlib.h>
+
+void* upkr_unpack(void* destination, void* compressed_data);
+
+int main(int argn, char** argv) {
+  void* input_buffer = malloc(1024*1024);
+  void* output_buffer = malloc(1024*1024);
+  
+  FILE* in_file = fopen(argv[1], "rb");
+  int in_size = fread(input_buffer, 1, 1024*1024, in_file);
+  fclose(in_file);
+  
+  printf("Compressed size: %d\n", in_size);
+  
+  void* end_ptr = upkr_unpack(output_buffer, input_buffer);
+  int out_size = (char*)end_ptr - (char*)output_buffer;
+  
+  printf("Uncompressed size: %d\n", out_size);
+  
+  FILE* out_file = fopen(argv[2], "wb");
+  fwrite(output_buffer, 1, out_size, out_file);
+  fclose(out_file);
+  
+  return 0;
+}

c_unpacker/readme.txt

@@ -0,0 +1,4 @@
+a very simple unpacker in c, as a reference for people wanting to implement their own unpacker.
+absolutely not production ready, it makes no effort to ensure the output buffer can actually
+hold the uncompressed data.
+!!! Never run on untrusted input !!!

c_unpacker/unpack.c

@@ -0,0 +1,96 @@
+typedef unsigned char u8;
+typedef unsigned short u16;
+typedef unsigned long u32;
+
+u8* upkr_data_ptr;
+u8 upkr_probs[1 + 255 + 1 + 2*32 + 2*32]; 
+#ifdef UPKR_BITSTREAM
+u16 upkr_state;
+u8 upkr_current_byte;
+int upkr_bits_left;
+#else
+u32 upkr_state;
+#endif
+
+int upkr_decode_bit(int context_index) {
+#ifdef UPKR_BITSTREAM
+    while(upkr_state < 32768) {
+        if(upkr_bits_left == 0) {
+            upkr_current_byte = *upkr_data_ptr++;
+            upkr_bits_left = 8;
+        }
+        upkr_state = (upkr_state << 1) + (upkr_current_byte & 1);
+        upkr_current_byte >>= 1;
+        --upkr_bits_left;
+    }
+#else
+    while(upkr_state < 4096) {
+        upkr_state = (upkr_state << 8) | *upkr_data_ptr++;
+    }
+#endif
+   
+    int prob = upkr_probs[context_index];
+    int bit = (upkr_state & 255) < prob ? 1 : 0;
+    
+    if(bit) {
+        upkr_state = prob * (upkr_state >> 8) + (upkr_state & 255);
+        prob += (256 - prob + 8) >> 4;
+    } else {
+        upkr_state = (256 - prob) * (upkr_state >> 8) + (upkr_state & 255) - prob;
+        prob -= (prob + 8) >> 4;
+    }
+    upkr_probs[context_index] = prob;
+
+    return bit;
+}
+
+int upkr_decode_length(int context_index) {
+    int length = 0;
+    int bit_pos = 0;
+    while(upkr_decode_bit(context_index)) {
+        length |= upkr_decode_bit(context_index + 1) << bit_pos++;
+        context_index += 2;
+    }
+    return length | (1 << bit_pos);
+}
+
+void* upkr_unpack(void* destination, void* compressed_data) {
+    upkr_data_ptr = (u8*)compressed_data;
+    upkr_state = 0;
+#ifdef UPKR_BITSTREAM
+    upkr_bits_left = 0;
+#endif
+    for(int i = 0; i < sizeof(upkr_probs); ++i)
+        upkr_probs[i] = 128;
+    
+    u8* write_ptr = (u8*)destination;
+    
+    int prev_was_match = 0;
+    int offset = 0;
+    for(;;) {
+        if(upkr_decode_bit(0)) {
+            if(prev_was_match || upkr_decode_bit(256)) {
+                offset = upkr_decode_length(257) - 1;
+                if(offset == 0) {
+                    break;
+                }
+            }
+            int length = upkr_decode_length(257 + 64);
+            while(length--) {
+                *write_ptr = write_ptr[-offset];
+                ++write_ptr;
+            }
+            prev_was_match = 1;
+        } else {
+            int byte = 1;
+            while(byte < 256) {
+                int bit = upkr_decode_bit(byte);
+                byte = (byte << 1) + bit;
+            }
+            *write_ptr++ = byte;
+            prev_was_match = 0;
+        }
+    }
+    
+    return write_ptr;
+}

src/greedy_packer.rs

@@ -3,10 +3,15 @@ use crate::match_finder::MatchFinder;
 use crate::rans::RansCoder;
 use crate::ProgressCallback;
 
-pub fn pack(data: &[u8], mut progress_callback: Option<ProgressCallback>) -> Vec<u8> {
+pub fn pack(
+    data: &[u8],
+    use_bitstream: bool,
+    parity_contexts: usize,
+    mut progress_callback: Option<ProgressCallback>,
+) -> Vec<u8> {
     let mut match_finder = MatchFinder::new(data);
-    let mut rans_coder = RansCoder::new();
-    let mut state = lz::CoderState::new();
+    let mut rans_coder = RansCoder::new(use_bitstream);
+    let mut state = lz::CoderState::new(parity_contexts);
 
     let mut pos = 0;
     while pos < data.len() {

src/lib.rs

@@ -2,17 +2,38 @@ mod context_state;
 mod greedy_packer;
 mod lz;
 mod match_finder;
-mod rans;
 mod parsing_packer;
+mod rans;
 
 pub use lz::unpack;
 
 pub type ProgressCallback<'a> = &'a mut dyn FnMut(usize);
 
-pub fn pack(data: &[u8], level: u8, progress_callback: Option<ProgressCallback>) -> Vec<u8> {
+pub fn pack(
+    data: &[u8],
+    level: u8,
+    use_bitstream: bool,
+    parity_contexts: usize,
+    progress_callback: Option<ProgressCallback>,
+) -> Vec<u8> {
     if level == 0 {
-        greedy_packer::pack(data, progress_callback)
+        greedy_packer::pack(data, use_bitstream, parity_contexts, progress_callback)
     } else {
-        parsing_packer::pack(data, level, progress_callback)
+        parsing_packer::pack(
+            data,
+            level,
+            use_bitstream,
+            parity_contexts,
+            progress_callback,
+        )
     }
 }
+
+pub fn compressed_size(mut data: &[u8]) -> f32 {
+    let mut state = 0;
+    while state < 4096 {
+        state = (state << 8) | data[0] as u32;
+        data = &data[1..];
+    }
+    data.len() as f32 + (state as f32).log2() / 8.
+}

src/lz.rs

@@ -9,41 +9,49 @@ pub enum Op {
 
 impl Op {
     pub fn encode(&self, coder: &mut dyn EntropyCoder, state: &mut CoderState) {
+        let literal_base = state.pos % state.parity_contexts * 256;
         match self {
             &Op::Literal(lit) => {
-                encode_bit(coder, state, 0, false);
+                encode_bit(coder, state, literal_base, false);
                 let mut context_index = 1;
                 for i in (0..8).rev() {
                     let bit = (lit >> i) & 1 != 0;
-                    encode_bit(coder, state, context_index, bit);
+                    encode_bit(coder, state, literal_base + context_index, bit);
                     context_index = (context_index << 1) | bit as usize;
                 }
                 state.prev_was_match = false;
+                state.pos += 1;
             }
             &Op::Match { offset, len } => {
-                encode_bit(coder, state, 0, true);
+                encode_bit(coder, state, literal_base, true);
                 if !state.prev_was_match {
-                    encode_bit(coder, state, 256, offset != state.last_offset);
+                    encode_bit(
+                        coder,
+                        state,
+                        256 * state.parity_contexts,
+                        offset != state.last_offset,
+                    );
                 } else {
                     assert!(offset != state.last_offset);
                 }
                 if offset != state.last_offset {
-                    encode_length(coder, state, 257, offset + 1);
+                    encode_length(coder, state, 256 * state.parity_contexts + 1, offset + 1);
                     state.last_offset = offset;
                 }
-                encode_length(coder, state, 257 + 64, len);
+                encode_length(coder, state, 256 * state.parity_contexts + 65, len);
                 state.prev_was_match = true;
+                state.pos += len as usize;
             }
         }
     }
 }
 
 pub fn encode_eof(coder: &mut dyn EntropyCoder, state: &mut CoderState) {
-    encode_bit(coder, state, 0, true);
+    encode_bit(coder, state, state.pos % state.parity_contexts * 256, true);
     if !state.prev_was_match {
-        encode_bit(coder, state, 256, true);
+        encode_bit(coder, state, 256 * state.parity_contexts, true);
     }
-    encode_length(coder, state, 257, 1);
+    encode_length(coder, state, 256 * state.parity_contexts + 1, 1);
 }
 
 fn encode_bit(
@@ -76,16 +84,20 @@ fn encode_length(
 #[derive(Clone)]
 pub struct CoderState {
     contexts: ContextState,
+    parity_contexts: usize,
     last_offset: u32,
-    prev_was_match: bool
+    prev_was_match: bool,
+    pos: usize,
 }
 
 impl CoderState {
-    pub fn new() -> CoderState {
+    pub fn new(parity_contexts: usize) -> CoderState {
         CoderState {
-            contexts: ContextState::new(1 + 255 + 1 + 64 + 64),
+            contexts: ContextState::new((1 + 255) * parity_contexts + 1 + 64 + 64),
             last_offset: 0,
-            prev_was_match: false
+            parity_contexts,
+            prev_was_match: false,
+            pos: 0,
         }
     }
 
@@ -94,9 +106,9 @@ impl CoderState {
     }
 }
 
-pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
-    let mut decoder = RansDecoder::new(packed_data);
-    let mut contexts = ContextState::new(1 + 255 + 1 + 64 + 64);
+pub fn unpack(packed_data: &[u8], use_bitstream: bool, parity_contexts: usize) -> Vec<u8> {
+    let mut decoder = RansDecoder::new(packed_data, use_bitstream);
+    let mut contexts = ContextState::new((1 + 255) * parity_contexts + 1 + 64 + 64);
     let mut result = vec![];
     let mut offset = 0;
     let mut prev_was_match = false;
@@ -119,14 +131,17 @@ pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
     }
 
     loop {
-        if decoder.decode_with_context(&mut contexts.context_mut(0)) {
-            if prev_was_match || decoder.decode_with_context(&mut contexts.context_mut(256)) {
-                offset = decode_length(&mut decoder, &mut contexts, 257) - 1;
+        let literal_base = result.len() % parity_contexts * 256;
+        if decoder.decode_with_context(&mut contexts.context_mut(literal_base)) {
+            if prev_was_match
+                || decoder.decode_with_context(&mut contexts.context_mut(256 * parity_contexts))
+            {
+                offset = decode_length(&mut decoder, &mut contexts, 256 * parity_contexts + 1) - 1;
                 if offset == 0 {
                     break;
                 }
             }
-            let length = decode_length(&mut decoder, &mut contexts, 257 + 64);
+            let length = decode_length(&mut decoder, &mut contexts, 256 * parity_contexts + 65);
             for _ in 0..length {
                 result.push(result[result.len() - offset]);
             }
@@ -135,7 +150,8 @@ pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
             let mut context_index = 1;
             let mut byte = 0;
             for i in (0..8).rev() {
-                let bit = decoder.decode_with_context(&mut contexts.context_mut(context_index));
+                let bit = decoder
+                    .decode_with_context(&mut contexts.context_mut(literal_base + context_index));
                 context_index = (context_index << 1) | bit as usize;
                 byte |= (bit as u8) << i;
             }

src/main.rs

@@ -1,5 +1,6 @@
 use anyhow::{bail, Result};
 use std::io::prelude::*;
+use std::process;
 use std::{fs::File, path::PathBuf};
 
 fn main() -> Result<()> {
@@ -9,6 +10,15 @@ fn main() -> Result<()> {
         None => print_help(),
         Some("pack") => {
             let level = args.opt_value_from_str(["-l", "--level"])?.unwrap_or(2u8);
+            let use_bitstream = args.contains(["-b", "--bitstream"]);
+            let parity_contexts = args
+                .opt_value_from_str(["-p", "--parity"])?
+                .unwrap_or(1usize);
+
+            if parity_contexts != 1 && parity_contexts != 2 && parity_contexts != 4 {
+                eprintln!("--parity has to be 1, 2 or 4");
+                process::exit(1);
+            }
 
             let infile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
             let outfile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
@@ -21,6 +31,8 @@ fn main() -> Result<()> {
             let packed_data = upkr::pack(
                 &data,
                 level,
+                use_bitstream,
+                parity_contexts,
                 Some(&mut |pos| {
                     pb.set(pos as u64);
                 }),
@@ -36,12 +48,22 @@ fn main() -> Result<()> {
             File::create(outfile)?.write_all(&packed_data)?;
         }
         Some("unpack") => {
+            let use_bitstream = args.contains(["-b", "--bitstream"]);
+            let parity_contexts = args
+                .opt_value_from_str(["-p", "--parity"])?
+                .unwrap_or(1usize);
+
+            if parity_contexts != 1 && parity_contexts != 2 && parity_contexts != 4 {
+                eprintln!("--parity has to be 1, 2 or 4");
+                process::exit(1);
+            }
+
             let infile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
             let outfile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
 
             let mut data = vec![];
             File::open(infile)?.read_to_end(&mut data)?;
-            let packed_data = upkr::unpack(&data);
+            let packed_data = upkr::unpack(&data, use_bitstream, parity_contexts);
             File::create(outfile)?.write_all(&packed_data)?;
         }
         Some(other) => {
@@ -54,7 +76,11 @@ fn main() -> Result<()> {
 
 fn print_help() {
     eprintln!("Usage:");
-    eprintln!("  upkr pack [-l level(0-9)] <infile> <outfile>");
-    eprintln!("  upkr unpack <infile> <outfile>");
-    std::process::exit(1);
+    eprintln!("  upkr pack [-b] [-l level(0-9)] [-p N] <infile> <outfile>");
+    eprintln!("  upkr unpack [-b] [-p N] <infile> <outfile>");
+    eprintln!();
+    eprintln!(" -b, --bitstream     bitstream mode");
+    eprintln!(" -l, --level N       compression level 0-9");
+    eprintln!(" -p, --parity N      use N (2/4) parity contexts");
+    process::exit(1);
 }

src/parsing_packer.rs

@@ -6,15 +6,26 @@ use crate::match_finder::MatchFinder;
 use crate::rans::{CostCounter, RansCoder};
 use crate::{lz, ProgressCallback};
 
-pub fn pack(data: &[u8], level: u8, progress_cb: Option<ProgressCallback>) -> Vec<u8> {
-    let mut parse = parse(data, Config::from_level(level), progress_cb);
+pub fn pack(
+    data: &[u8],
+    level: u8,
+    use_bitstream: bool,
+    parity_contexts: usize,
+    progress_cb: Option<ProgressCallback>,
+) -> Vec<u8> {
+    let mut parse = parse(
+        data,
+        Config::from_level(level),
+        parity_contexts,
+        progress_cb,
+    );
     let mut ops = vec![];
     while let Some(link) = parse {
         ops.push(link.op);
         parse = link.prev.clone();
     }
-    let mut state = lz::CoderState::new();
-    let mut coder = RansCoder::new();
+    let mut state = lz::CoderState::new(parity_contexts);
+    let mut coder = RansCoder::new(use_bitstream);
     for op in ops.into_iter().rev() {
         op.encode(&mut coder, &mut state);
     }
@@ -38,6 +49,7 @@ type Arrivals = HashMap<usize, Vec<Arrival>>;
 fn parse(
     data: &[u8],
     config: Config,
+    parity_contexts: usize,
     mut progress_cb: Option<ProgressCallback>,
 ) -> Option<Rc<Parse>> {
     let mut match_finder = MatchFinder::new(data)
@@ -129,7 +141,7 @@ fn parse(
         0,
         Arrival {
             parse: None,
-            state: lz::CoderState::new(),
+            state: lz::CoderState::new(parity_contexts),
             cost: 0.0,
         },
         max_arrivals,

src/rans.rs

@@ -1,6 +1,5 @@
 use crate::context_state::Context;
 
-const L_BITS: u32 = 12;
 pub const PROB_BITS: u32 = 8;
 pub const ONE_PROB: u32 = 1 << PROB_BITS;
 
@@ -13,43 +12,75 @@ pub trait EntropyCoder {
     }
 }
 
-pub struct RansCoder(Vec<u16>);
+pub struct RansCoder {
+    bits: Vec<u16>,
+    use_bitstream: bool,
+}
 
 impl EntropyCoder for RansCoder {
     fn encode_bit(&mut self, bit: bool, prob: u16) {
         assert!(prob < 32768);
-        self.0.push(prob | ((bit as u16) << 15));
+        self.bits.push(prob | ((bit as u16) << 15));
     }
 }
 
 impl RansCoder {
-    pub fn new() -> RansCoder {
-        RansCoder(Vec::new())
+    pub fn new(use_bitstream: bool) -> RansCoder {
+        RansCoder {
+            bits: Vec::new(),
+            use_bitstream,
+        }
     }
 
     pub fn finish(self) -> Vec<u8> {
         let mut buffer = vec![];
-        let mut state = 1 << L_BITS;
+        let l_bits: u32 = if self.use_bitstream { 15 } else { 12 };
+        let mut state = 1 << l_bits;
 
-        const MAX_STATE_FACTOR: u32 = 1 << (L_BITS + 8 - PROB_BITS);
-        for step in self.0.into_iter().rev() {
+        let mut byte = 0u8;
+        let mut bit = 8;
+        let mut flush_state: Box<dyn FnMut(&mut u32)> = if self.use_bitstream {
+            Box::new(|state: &mut u32| {
+                bit -= 1;
+                byte |= ((*state & 1) as u8) << bit;
+                if bit == 0 {
+                    buffer.push(byte);
+                    byte = 0;
+                    bit = 8;
+                }
+                *state >>= 1;
+            })
+        } else {
+            Box::new(|state: &mut u32| {
+                buffer.push(*state as u8);
+                *state >>= 8;
+            })
+        };
+
+        let num_flush_bits = if self.use_bitstream { 1 } else { 8 };
+        let max_state_factor: u32 = 1 << (l_bits + num_flush_bits - PROB_BITS);
+        for step in self.bits.into_iter().rev() {
             let prob = step as u32 & 32767;
             let (start, prob) = if step & 32768 != 0 {
                 (0, prob)
             } else {
                 (prob, ONE_PROB - prob)
             };
-            let max_state = MAX_STATE_FACTOR * prob;
+            let max_state = max_state_factor * prob;
             while state >= max_state {
-                buffer.push(state as u8);
-                state >>= 8;
+                flush_state(&mut state);
             }
             state = ((state / prob) << PROB_BITS) + (state % prob) + start;
         }
 
         while state > 0 {
-            buffer.push(state as u8);
-            state >>= 8;
+            flush_state(&mut state);
+        }
+
+        drop(flush_state);
+
+        if self.use_bitstream && byte != 0 {
+            buffer.push(byte);
         }
 
         buffer.reverse();
@@ -99,14 +130,22 @@ impl EntropyCoder for CostCounter {
 pub struct RansDecoder<'a> {
     data: &'a [u8],
     state: u32,
+    use_bitstream: bool,
+    byte: u8,
+    bits_left: u8,
 }
 
 const PROB_MASK: u32 = ONE_PROB - 1;
-const L: u32 = 1 << L_BITS;
 
 impl<'a> RansDecoder<'a> {
-    pub fn new(data: &'a [u8]) -> RansDecoder<'a> {
-        RansDecoder { data, state: 0 }
+    pub fn new(data: &'a [u8], use_bitstream: bool) -> RansDecoder<'a> {
+        RansDecoder {
+            data,
+            state: 0,
+            use_bitstream,
+            byte: 0,
+            bits_left: 0,
+        }
     }
 
     pub fn decode_with_context(&mut self, context: &mut Context) -> bool {
@@ -117,9 +156,22 @@ impl<'a> RansDecoder<'a> {
 
     pub fn decode_bit(&mut self, prob: u16) -> bool {
         let prob = prob as u32;
-        while self.state < L {
-            self.state = (self.state << 8) | self.data[0] as u32;
-            self.data = &self.data[1..];
+        if self.use_bitstream {
+            while self.state < 32768 {
+                if self.bits_left == 0 {
+                    self.byte = self.data[0];
+                    self.data = &self.data[1..];
+                    self.bits_left = 8;
+                }
+                self.state = (self.state << 1) | (self.byte & 1) as u32;
+                self.byte >>= 1;
+                self.bits_left -= 1;
+            }
+        } else {
+            while self.state < 4096 {
+                self.state = (self.state << 8) | self.data[0] as u32;
+                self.data = &self.data[1..];
+            }
         }
 
         let bit = (self.state & PROB_MASK) < prob;