add --no-repeated-offsets to help

unpack_armv6m: update comment headers and remove unneeded pushed register

Cargo.lock

@@ -62,6 +62,12 @@ version = "1.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e2abad23fbc42b3700f2f279844dc832adb2b2eb069b2df918f455c4e18cc646"
 
+[[package]]
+name = "lexopt"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "478ee9e62aaeaf5b140bd4138753d1f109765488581444218d3ddda43234f3e8"
+
 [[package]]
 name = "libc"
 version = "0.2.108"
@@ -89,12 +95,6 @@ dependencies = [
  "winapi",
 ]
 
-[[package]]
-name = "pico-args"
-version = "0.4.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "db8bcd96cb740d03149cbad5518db9fd87126a10ab519c011893b1754134c468"
-
 [[package]]
 name = "sacabase"
 version = "2.0.0"
@@ -117,12 +117,12 @@ dependencies = [
 
 [[package]]
 name = "upkr"
-version = "0.1.0"
+version = "0.2.0-pre1"
 dependencies = [
  "anyhow",
  "cdivsufsort",
+ "lexopt",
  "pbr",
- "pico-args",
 ]
 
 [[package]]

Cargo.toml

@@ -1,12 +1,12 @@
 [package]
 name = "upkr"
-version = "0.1.0"
+version = "0.2.0-pre2"
 edition = "2021"
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
 cdivsufsort = "2"
-pico-args = "0.4"
+lexopt = "0.2.1"
 anyhow = "1"
 pbr = "1"

asm_unpackers/.gitignore

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

asm_unpackers/Makefile

@@ -0,0 +1,51 @@
+build/unpack_riscv64: ../c_unpacker/main.c unpack_riscv.S
+	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_riscv64.o: unpack_riscv.S
+	mkdir -p build
+	riscv64-linux-gnu-gcc -c -o $@ $?
+
+build/unpack_riscv64.bin: build/unpack_riscv64.o
+	riscv64-linux-gnu-objcopy -O binary --only-section=.text $? $@
+
+disas-riscv64: build/unpack_riscv64.o
+	riscv64-linux-gnu-objdump -d $?
+
+build/unpack_riscv32.o: unpack_riscv.S
+	mkdir -p build
+	riscv64-linux-gnu-gcc -march=rv32imc -mabi=ilp32 -c -o $@ $?
+	
+build/unpack_riscv32.bin: build/unpack_riscv32.o
+	riscv64-linux-gnu-objcopy -O binary --only-section=.text $? $@
+
+disas-riscv32: build/unpack_riscv32.o
+	riscv64-linux-gnu-objdump -d $?
+
+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
+	mkdir -p build
+	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 build/unpack_riscv64.bin build/unpack_riscv32.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 }

asm_unpackers/unpack_riscv.S

@@ -0,0 +1,142 @@
+.section .text
+
+#define FRAME_SIZE (256+32*4+4)
+
+// x8 prob array ptr
+// x9 prev was literal
+// x10 out ptr
+// x11 in ptr
+// x12 offset
+// x13 state
+
+.global upkr_unpack
+.type upkr_unpack, %function
+upkr_unpack:
+	mv t4, ra
+	mv x17, x8
+	mv t6, x9
+	li x13, FRAME_SIZE
+	li x9, 128
+1:
+	addi sp, sp, -1
+	sb x9, 0(sp)
+	addi x13, x13, -1
+	bnez x13, 1b
+	
+.Lmainloop:
+	li x14, 0
+	jal upkr_decode_bit
+	beqz x15, .Lliteral
+	
+	slli x14, x14, 8
+	beqz x9, .Lread_offset_inc_x14
+	jal upkr_decode_bit
+	bnez x15, .Lread_offset
+
+.Lfinished_offset:
+	addi x14, x14, 64
+	jal t3, upkr_decode_number
+1:
+	add x14, x10, t0
+	lbu x14, (x14)
+.Lstore_byte:
+	sb x14, (x10)
+	addi x10, x10, 1
+	addi x9, x9, 1
+	blt x9, x0, 1b
+	j .Lmainloop
+	
+.Lliteral:
+	jal upkr_decode_bit
+	addi x14, x14, -1
+	slli x14, x14, 1
+	add x14, x14, x15
+	srli x9, x14, 8
+	beqz x9, .Lliteral
+	j .Lstore_byte
+
+.Lread_offset_inc_x14:
+	addi x14, x14, 1
+.Lread_offset:
+	jal t3, upkr_decode_number
+	addi t0, x9, 1
+	bnez t0, .Lfinished_offset
+.Ldone:
+	addi sp, sp, FRAME_SIZE
+	mv x8, x17
+	mv x9, t6
+	jr t4
+
+// x14 context index
+// return: x9 negtive decoded number 
+upkr_decode_number:
+	mv t5, x14
+	li x9, 0
+	li x8, -1
+1:
+	jal upkr_decode_bit
+	beqz x15, 1f
+	jal upkr_decode_bit
+	beqz x15, 2f
+	add x9, x9, x8
+2:
+	slli x8, x8, 1
+	j 1b
+1:
+	add x9, x9, x8
+	
+	mv x14, t5
+	jr t3
+
+upkr_load_byte:
+	lbu x15, 0(x11)
+	addi x11, x11, 1
+	slli x13, x13, 8
+	add x13, x13, x15
+// x8 prob array ptr
+// x11 in ptr
+// x13 state
+// x14 context index
+// return:
+//   x14 context index + 1
+//   x15 decoded bit
+upkr_decode_bit:
+	srli x15, x13, 12
+	beqz x15, upkr_load_byte
+
+	mv t1, x14
+	mv t2, x10
+	
+	add x14, x14, sp
+	lbu x12, 0(x14)
+	
+	andi x10, x13, 255
+	sltu x15, x10, x12
+	srli x13, x13, 8
+	beqz x15, .Lelse
+	
+	mul x13, x13, x12
+	add x13, x13, x10
+	li x10, 256 + 8
+	sub x10, x10, x12
+	srli x10, x10, 4
+	add x12, x12, x10
+	j .Lendif
+
+.Lelse:
+	li x16, 256
+	sub x16, x16, x12
+	mul x13, x13, x16
+	add x13, x13, x10
+	sub x13, x13, x12
+	addi x10, x12, 8
+	srli x10, x10, 4
+	sub x12, x12, x10
+
+.Lendif:
+
+	sb x12, 0(x14)
+	
+	addi x14, t1, 1
+	mv x10, t2
+	ret

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/context_state.rs

@@ -1,4 +1,7 @@
-use crate::rans::{PROB_BITS, ONE_PROB};
+use crate::{
+    rans::{ONE_PROB, PROB_BITS},
+    Config,
+};
 
 const INIT_PROB: u16 = 1 << (PROB_BITS - 1);
 const UPDATE_RATE: u32 = 4;
@@ -6,7 +9,9 @@ const UPDATE_ADD: u32 = 8;
 
 #[derive(Clone)]
 pub struct ContextState {
-    contexts: Vec<u16>,
+    contexts: Vec<u8>,
+    invert_bit_encoding: bool,
+    simplified_prob_update: bool,
 }
 
 pub struct Context<'a> {
@@ -15,9 +20,11 @@ pub struct Context<'a> {
 }
 
 impl ContextState {
-    pub fn new(size: usize) -> ContextState {
+    pub fn new(size: usize, config: &Config) -> ContextState {
         ContextState {
-            contexts: vec![INIT_PROB; size],
+            contexts: vec![INIT_PROB as u8; size],
+            invert_bit_encoding: config.invert_bit_encoding,
+            simplified_prob_update: config.simplified_prob_update,
         }
     }
 
@@ -28,15 +35,26 @@ impl ContextState {
 
 impl<'a> Context<'a> {
     pub fn prob(&self) -> u16 {
-        self.state.contexts[self.index]
+        self.state.contexts[self.index] as u16
     }
 
     pub fn update(&mut self, bit: bool) {
         let old = self.state.contexts[self.index];
-        self.state.contexts[self.index] = if bit {
+
-            old + ((ONE_PROB - old as u32 + UPDATE_ADD) >> UPDATE_RATE) as u16
+        self.state.contexts[self.index] = if self.state.simplified_prob_update {
+            let offset = if bit ^ self.state.invert_bit_encoding {
+                ONE_PROB as i32 >> UPDATE_RATE
+            } else {
+                0
+            };
+
+            (offset + old as i32 - ((old as i32 + UPDATE_ADD as i32) >> UPDATE_RATE)) as u8
         } else {
-            old - ((old + UPDATE_ADD as u16) >> UPDATE_RATE)
+            if bit ^ self.state.invert_bit_encoding {
+                old + ((ONE_PROB - old as u32 + UPDATE_ADD) >> UPDATE_RATE) as u8
+            } else {
+                old - ((old as u32 + UPDATE_ADD) >> UPDATE_RATE) as u8
+            }
         };
     }
 }

src/greedy_packer.rs

@@ -1,12 +1,16 @@
-use crate::lz;
 use crate::match_finder::MatchFinder;
 use crate::rans::RansCoder;
 use crate::ProgressCallback;
+use crate::{lz, Config};
 
-pub fn pack(data: &[u8], mut progress_callback: Option<ProgressCallback>) -> Vec<u8> {
+pub fn pack(
+    data: &[u8],
+    config: &Config,
+    mut progress_callback: Option<ProgressCallback>,
+) -> Vec<u8> {
     let mut match_finder = MatchFinder::new(data);
-    let mut rans_coder = RansCoder::new();
+    let mut rans_coder = RansCoder::new(config);
-    let mut state = lz::CoderState::new();
+    let mut state = lz::CoderState::new(config);
 
     let mut pos = 0;
     while pos < data.len() {
@@ -22,7 +26,7 @@ pub fn pack(data: &[u8], mut progress_callback: Option<ProgressCallback>) -> Vec
                     offset: offset as u32,
                     len: m.length as u32,
                 }
-                .encode(&mut rans_coder, &mut state);
+                .encode(&mut rans_coder, &mut state, config);
                 pos += m.length;
                 encoded_match = true;
             }
@@ -41,7 +45,7 @@ pub fn pack(data: &[u8], mut progress_callback: Option<ProgressCallback>) -> Vec
                         offset: offset as u32,
                         len: length as u32,
                     }
-                    .encode(&mut rans_coder, &mut state);
+                    .encode(&mut rans_coder, &mut state, config);
                     pos += length;
                     encoded_match = true;
                 }
@@ -49,11 +53,11 @@ pub fn pack(data: &[u8], mut progress_callback: Option<ProgressCallback>) -> Vec
         }
 
         if !encoded_match {
-            lz::Op::Literal(data[pos]).encode(&mut rans_coder, &mut state);
+            lz::Op::Literal(data[pos]).encode(&mut rans_coder, &mut state, config);
             pos += 1;
         }
     }
 
-    lz::encode_eof(&mut rans_coder, &mut state);
+    lz::encode_eof(&mut rans_coder, &mut state, config);
     rans_coder.finish()
 }

src/lib.rs

@@ -2,17 +2,65 @@ 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 struct Config {
-    if level == 0 {
+    pub use_bitstream: bool,
-        greedy_packer::pack(data, progress_callback)
+    pub parity_contexts: usize,
-    } else {
+
-        parsing_packer::pack(data, level, progress_callback)
+    pub invert_bit_encoding: bool,
+    pub is_match_bit: bool,
+    pub new_offset_bit: bool,
+    pub continue_value_bit: bool,
+
+    pub bitstream_is_big_endian: bool,
+    pub simplified_prob_update: bool,
+
+    pub no_repeated_offsets: bool,
+}
+
+impl Default for Config {
+    fn default() -> Config {
+        Config {
+            use_bitstream: false,
+            parity_contexts: 1,
+
+            invert_bit_encoding: false,
+            is_match_bit: true,
+            new_offset_bit: true,
+            continue_value_bit: true,
+
+            bitstream_is_big_endian: false,
+            simplified_prob_update: false,
+
+            no_repeated_offsets: false,
+        }
     }
 }
+
+pub fn pack(
+    data: &[u8],
+    level: u8,
+    config: Config,
+    progress_callback: Option<ProgressCallback>,
+) -> Vec<u8> {
+    if level == 0 {
+        greedy_packer::pack(data, &config, progress_callback)
+    } else {
+        parsing_packer::pack(data, level, &config, 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

@@ -1,5 +1,6 @@
 use crate::context_state::ContextState;
 use crate::rans::{EntropyCoder, RansDecoder};
+use crate::Config;
 
 #[derive(Copy, Clone, Debug)]
 pub enum Op {
@@ -8,40 +9,66 @@ pub enum Op {
 }
 
 impl Op {
-    pub fn encode(&self, coder: &mut dyn EntropyCoder, state: &mut CoderState) {
+    pub fn encode(&self, coder: &mut dyn EntropyCoder, state: &mut CoderState, config: &Config) {
+        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, !config.is_match_bit);
                 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, config.is_match_bit);
-                if !state.prev_was_match {
+                if !state.prev_was_match && !config.no_repeated_offsets {
-                    encode_bit(coder, state, 256, offset != state.last_offset);
+                    encode_bit(
+                        coder,
+                        state,
+                        256 * state.parity_contexts,
+                        (offset != state.last_offset) == config.new_offset_bit,
+                    );
                 } else {
-                    assert!(offset != state.last_offset);
+                    assert!(offset != state.last_offset || config.no_repeated_offsets);
                 }
-                if offset != state.last_offset {
+                if offset != state.last_offset || config.no_repeated_offsets {
-                    encode_length(coder, state, 257, offset + 1);
+                    encode_length(
+                        coder,
+                        state,
+                        256 * state.parity_contexts + 1,
+                        offset + 1,
+                        config,
+                    );
                     state.last_offset = offset;
                 }
-                encode_length(coder, state, 257 + 64, len);
+                encode_length(coder, state, 256 * state.parity_contexts + 65, len, config);
                 state.prev_was_match = true;
+                state.pos += len as usize;
             }
         }
     }
 }
 
-pub fn encode_eof(coder: &mut dyn EntropyCoder, state: &mut CoderState) {
+pub fn encode_eof(coder: &mut dyn EntropyCoder, state: &mut CoderState, config: &Config) {
-    encode_bit(coder, state, 0, true);
+    encode_bit(
-    encode_bit(coder, state, 256, true);
+        coder,
-    encode_length(coder, state, 257, 1);
+        state,
+        state.pos % state.parity_contexts * 256,
+        config.is_match_bit,
+    );
+    if !state.prev_was_match {
+        encode_bit(
+            coder,
+            state,
+            256 * state.parity_contexts,
+            config.new_offset_bit,
+        );
+    }
+    encode_length(coder, state, 256 * state.parity_contexts + 1, 1, config);
 }
 
 fn encode_bit(
@@ -58,32 +85,37 @@ fn encode_length(
     state: &mut CoderState,
     context_start: usize,
     mut value: u32,
+    config: &Config,
 ) {
     assert!(value >= 1);
 
     let mut context_index = context_start;
     while value >= 2 {
-        encode_bit(coder, state, context_index, true);
+        encode_bit(coder, state, context_index, config.continue_value_bit);
         encode_bit(coder, state, context_index + 1, value & 1 != 0);
         context_index += 2;
         value >>= 1;
     }
-    encode_bit(coder, state, context_index, false);
+    encode_bit(coder, state, context_index, !config.continue_value_bit);
 }
 
 #[derive(Clone)]
 pub struct CoderState {
     contexts: ContextState,
     last_offset: u32,
-    prev_was_match: bool
+    prev_was_match: bool,
+    pos: usize,
+    parity_contexts: usize,
 }
 
 impl CoderState {
-    pub fn new() -> CoderState {
+    pub fn new(config: &Config) -> CoderState {
         CoderState {
-            contexts: ContextState::new(1 + 255 + 1 + 64 + 64),
+            contexts: ContextState::new((1 + 255) * config.parity_contexts + 1 + 64 + 64, config),
             last_offset: 0,
-            prev_was_match: false
+            prev_was_match: false,
+            pos: 0,
+            parity_contexts: config.parity_contexts,
         }
     }
 
@@ -92,9 +124,9 @@ impl CoderState {
     }
 }
 
-pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
+pub fn unpack(packed_data: &[u8], config: Config) -> Vec<u8> {
-    let mut decoder = RansDecoder::new(packed_data);
+    let mut decoder = RansDecoder::new(packed_data, &config);
-    let mut contexts = ContextState::new(1 + 255 + 1 + 64 + 64);
+    let mut contexts = ContextState::new((1 + 255) * config.parity_contexts + 1 + 64 + 64, &config);
     let mut result = vec![];
     let mut offset = 0;
     let mut prev_was_match = false;
@@ -103,10 +135,13 @@ pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
         decoder: &mut RansDecoder,
         contexts: &mut ContextState,
         mut context_index: usize,
+        config: &Config,
     ) -> usize {
         let mut length = 0;
         let mut bit_pos = 0;
-        while decoder.decode_with_context(&mut contexts.context_mut(context_index)) {
+        while decoder.decode_with_context(&mut contexts.context_mut(context_index))
+            == config.continue_value_bit
+        {
             length |= (decoder.decode_with_context(&mut contexts.context_mut(context_index + 1))
                 as usize)
                 << bit_pos;
@@ -117,14 +152,32 @@ pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
     }
 
     loop {
-        if decoder.decode_with_context(&mut contexts.context_mut(0)) {
+        let literal_base = result.len() % config.parity_contexts * 256;
-            if prev_was_match || decoder.decode_with_context(&mut contexts.context_mut(256)) {
+        if decoder.decode_with_context(&mut contexts.context_mut(literal_base))
-                offset = decode_length(&mut decoder, &mut contexts, 257) - 1;
+            == config.is_match_bit
+        {
+            if config.no_repeated_offsets
+                || prev_was_match
+                || decoder
+                    .decode_with_context(&mut contexts.context_mut(256 * config.parity_contexts))
+                    == config.new_offset_bit
+            {
+                offset = decode_length(
+                    &mut decoder,
+                    &mut contexts,
+                    256 * config.parity_contexts + 1,
+                    &config,
+                ) - 1;
                 if offset == 0 {
                     break;
                 }
             }
-            let length = decode_length(&mut decoder, &mut contexts, 257 + 64);
+            let length = decode_length(
+                &mut decoder,
+                &mut contexts,
+                256 * config.parity_contexts + 65,
+                &config,
+            );
             for _ in 0..length {
                 result.push(result[result.len() - offset]);
             }
@@ -133,7 +186,8 @@ 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

@@ -1,60 +1,151 @@
-use anyhow::{bail, Result};
+use anyhow::Result;
+use std::ffi::OsStr;
 use std::io::prelude::*;
+use std::process;
 use std::{fs::File, path::PathBuf};
 
 fn main() -> Result<()> {
-    let mut args = pico_args::Arguments::from_env();
+    let mut config = upkr::Config::default();
+    let mut reverse = false;
+    let mut unpack = false;
+    let mut level = 2;
+    let mut infile: Option<PathBuf> = None;
+    let mut outfile: Option<PathBuf> = None;
 
-    match args.subcommand()?.as_ref().map(|s| s.as_str()) {
+    let mut parser = lexopt::Parser::from_env();
-        None => print_help(),
+    while let Some(arg) = parser.next()? {
-        Some("pack") => {
+        use lexopt::prelude::*;
-            let level = args.opt_value_from_str(["-l", "--level"])?.unwrap_or(2u8);
+        match arg {
+            Short('b') | Long("bitstream") => config.use_bitstream = true,
+            Short('p') | Long("parity") => config.parity_contexts = parser.value()?.parse()?,
+            Short('r') | Long("reverse") => reverse = true,
+            Long("invert-is-match-bit") => config.is_match_bit = false,
+            Long("invert-new-offset-bit") => config.new_offset_bit = false,
+            Long("invert-continue-value-bit") => config.continue_value_bit = false,
+            Long("invert-bit-encoding") => config.invert_bit_encoding = true,
+            Long("simplified-prob-update") => config.simplified_prob_update = true,
+            Long("big-endian-bitstream") => {
+                config.use_bitstream = true;
+                config.bitstream_is_big_endian = true;
+            }
+            Long("no-repeated-offsets") => config.no_repeated_offsets = true,
 
-            let infile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+            Long("z80") => {
-            let outfile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+                config.use_bitstream = true;
+                config.bitstream_is_big_endian = true;
+                config.invert_bit_encoding = true;
+                config.simplified_prob_update = true;
+            }
+            Long("x86") => {
+                config.use_bitstream = true;
+                config.continue_value_bit = false;
+                config.is_match_bit = false;
+            }
 
-            let mut data = vec![];
+            Short('u') | Long("unpack") => unpack = true,
-            File::open(infile)?.read_to_end(&mut data)?;
+            Short('l') | Long("level") => level = parser.value()?.parse()?,
-            
+            Short('h') | Long("help") => print_help(0),
-            let mut pb = pbr::ProgressBar::new(data.len() as u64);
+            Value(val) if infile.is_none() => infile = Some(val.try_into()?),
-            pb.set_units(pbr::Units::Bytes);
+            Value(val) if outfile.is_none() => outfile = Some(val.try_into()?),
-            let packed_data = upkr::pack(
+            _ => return Err(arg.unexpected().into()),
-                &data,
-                level,
-                Some(&mut |pos| {
-                    pb.set(pos as u64);
-                }),
-            );
-            pb.finish();
-
-            println!(
-                "Compressed {} bytes to {} bytes ({}%)",
-                data.len(),
-                packed_data.len(),
-                packed_data.len() as f32 * 100. / data.len() as f32
-            );
-            File::create(outfile)?.write_all(&packed_data)?;
         }
-        Some("unpack") => {
+    }
-            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![];
+    let infile = infile.unwrap_or_else(|| print_help(1));
-            File::open(infile)?.read_to_end(&mut data)?;
+    let outfile = outfile.unwrap_or_else(|| {
-            let packed_data = upkr::unpack(&data);
+        let mut name = infile.clone();
-            File::create(outfile)?.write_all(&packed_data)?;
+        if unpack {
+            if name.extension().filter(|&e| e == "upk").is_some() {
+                name.set_extension("");
+            } else {
+                name.set_extension("bin");
+            }
+        } else {
+            let mut filename = name
+                .file_name()
+                .unwrap_or_else(|| OsStr::new(""))
+                .to_os_string();
+            filename.push(".upk");
+            name.set_file_name(filename);
         }
-        Some(other) => {
+        name
-            bail!("Unknown subcommand '{}'", other);
+    });
+
+    if config.parity_contexts != 1 && config.parity_contexts != 2 && config.parity_contexts != 4 {
+        eprintln!("--parity has to be 1, 2, or 4");
+        process::exit(1);
+    }
+
+    if !unpack {
+        let mut data = vec![];
+        File::open(infile)?.read_to_end(&mut data)?;
+        if reverse {
+            data.reverse();
         }
+
+        let mut pb = pbr::ProgressBar::new(data.len() as u64);
+        pb.set_units(pbr::Units::Bytes);
+        let mut packed_data = upkr::pack(
+            &data,
+            level,
+            config,
+            Some(&mut |pos| {
+                pb.set(pos as u64);
+            }),
+        );
+        pb.finish();
+
+        if reverse {
+            packed_data.reverse();
+        }
+
+        println!(
+            "Compressed {} bytes to {} bytes ({}%)",
+            data.len(),
+            packed_data.len(),
+            packed_data.len() as f32 * 100. / data.len() as f32
+        );
+        File::create(outfile)?.write_all(&packed_data)?;
+    } else {
+        let mut data = vec![];
+        File::open(infile)?.read_to_end(&mut data)?;
+        if reverse {
+            data.reverse();
+        }
+        let mut unpacked_data = upkr::unpack(&data, config);
+        if reverse {
+            unpacked_data.reverse();
+        }
+        File::create(outfile)?.write_all(&unpacked_data)?;
     }
 
     Ok(())
 }
 
-fn print_help() {
+fn print_help(exit_code: i32) -> ! {
     eprintln!("Usage:");
-    eprintln!("  upkr pack [-l level(0-9)] <infile> <outfile>");
+    eprintln!("  upkr [-l level(0-9)] [config options] <infile> [<outfile>]");
-    eprintln!("  upkr unpack <infile> <outfile>");
+    eprintln!("  upkr -u [config options] <infile> [<outfile>]");
-    std::process::exit(1);
+    eprintln!();
+    eprintln!(" -l, --level N       compression level 0-9");
+    eprintln!(" -u, --unpack        unpack infile");
+    eprintln!();
+    eprintln!("Config presets for specific unpackers:");
+    eprintln!(" --z80               --big-endian-bitstream --invert-bit-encoding --simplified-prob-update");
+    eprintln!(" --x86               --bitstream --invert-is-match-bit --invert-continue-value-bit");
+    eprintln!();
+    eprintln!("Config options (need to match when packing/unpacking):");
+    eprintln!(" -b, --bitstream     bitstream mode");
+    eprintln!(" -p, --parity N      use N (2/4) parity contexts");
+    eprintln!(" -r, --reverse       reverse input & output");
+    eprintln!();
+    eprintln!("Config options to tailor output to specific optimized unpackers:");
+    eprintln!(" --invert-is-match-bit");
+    eprintln!(" --invert-new-offset-bit");
+    eprintln!(" --invert-continue-value-bit");
+    eprintln!(" --invert-bit-encoding");
+    eprintln!(" --simplified-prob-update");
+    eprintln!(" --big-endian-bitstream   (implies --bitstream)");
+    eprintln!(" --no-repeated-offsets");
+    process::exit(exit_code);
 }

src/parsing_packer.rs

@@ -6,19 +6,24 @@ 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> {
+pub fn pack(
-    let mut parse = parse(data, Config::from_level(level), progress_cb);
+    data: &[u8],
+    level: u8,
+    config: &crate::Config,
+    progress_cb: Option<ProgressCallback>,
+) -> Vec<u8> {
+    let mut parse = parse(data, Config::from_level(level), config, 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 state = lz::CoderState::new(config);
-    let mut coder = RansCoder::new();
+    let mut coder = RansCoder::new(config);
     for op in ops.into_iter().rev() {
-        op.encode(&mut coder, &mut state);
+        op.encode(&mut coder, &mut state, config);
     }
-    lz::encode_eof(&mut coder, &mut state);
+    lz::encode_eof(&mut coder, &mut state, config);
     coder.finish()
 }
 
@@ -38,6 +43,7 @@ type Arrivals = HashMap<usize, Vec<Arrival>>;
 fn parse(
     data: &[u8],
     config: Config,
+    encoding_config: &crate::Config,
     mut progress_cb: Option<ProgressCallback>,
 ) -> Option<Rc<Parse>> {
     let mut match_finder = MatchFinder::new(data)
@@ -102,6 +108,7 @@ fn parse(
         length: usize,
         arrival: &Arrival,
         max_arrivals: usize,
+        config: &crate::Config,
     ) {
         cost_counter.reset();
         let mut state = arrival.state.clone();
@@ -109,7 +116,7 @@ fn parse(
             offset: offset as u32,
             len: length as u32,
         };
-        op.encode(cost_counter, &mut state);
+        op.encode(cost_counter, &mut state, config);
         add_arrival(
             arrivals,
             pos + length,
@@ -129,13 +136,13 @@ fn parse(
         0,
         Arrival {
             parse: None,
-            state: lz::CoderState::new(),
+            state: lz::CoderState::new(encoding_config),
             cost: 0.0,
         },
         max_arrivals,
     );
 
-    let cost_counter = &mut CostCounter::new();
+    let cost_counter = &mut CostCounter::new(encoding_config);
     let mut best_per_offset = HashMap::new();
     for pos in 0..data.len() {
         let match_length = |offset: usize| {
@@ -185,6 +192,7 @@ fn parse(
                     m.length,
                     &arrival,
                     max_arrivals,
+                    encoding_config,
                 );
                 if m.length >= config.greedy_size {
                     break 'arrival_loop;
@@ -208,6 +216,7 @@ fn parse(
                     length,
                     &arrival,
                     max_arrivals,
+                    encoding_config,
                 );
                 found_last_offset |= offset as u32 == arrival.state.last_offset();
                 if offset < near_matches.len() {
@@ -228,6 +237,7 @@ fn parse(
                         length,
                         &arrival,
                         max_arrivals,
+                        encoding_config,
                     );
                 }
             }
@@ -235,7 +245,7 @@ fn parse(
             cost_counter.reset();
             let mut state = arrival.state;
             let op = lz::Op::Literal(data[pos]);
-            op.encode(cost_counter, &mut state);
+            op.encode(cost_counter, &mut state, encoding_config);
             add_arrival(
                 &mut arrivals,
                 pos + 1,

src/rans.rs

@@ -1,6 +1,5 @@
-use crate::context_state::Context;
+use crate::{context_state::Context, Config};
 
-const L_BITS: u32 = 12;
 pub const PROB_BITS: u32 = 8;
 pub const ONE_PROB: u32 = 1 << PROB_BITS;
 
@@ -13,43 +12,93 @@ pub trait EntropyCoder {
     }
 }
 
-pub struct RansCoder(Vec<u16>);
+pub struct RansCoder {
+    bits: Vec<u16>,
+    use_bitstream: bool,
+    bitstream_is_big_endian: bool,
+    invert_bit_encoding: 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 ^ self.invert_bit_encoding) as u16) << 15));
     }
 }
 
 impl RansCoder {
-    pub fn new() -> RansCoder {
+    pub fn new(config: &Config) -> RansCoder {
-        RansCoder(Vec::new())
+        RansCoder {
+            bits: Vec::new(),
+            use_bitstream: config.use_bitstream,
+            bitstream_is_big_endian: config.bitstream_is_big_endian,
+            invert_bit_encoding: config.invert_bit_encoding,
+        }
     }
 
     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);
+        let mut byte = 0u8;
-        for step in self.0.into_iter().rev() {
+        let mut bit = if self.bitstream_is_big_endian { 0 } else { 8 };
+        let mut flush_state: Box<dyn FnMut(&mut u32)> = if self.use_bitstream {
+            if self.bitstream_is_big_endian {
+                Box::new(|state: &mut u32| {
+                    byte |= ((*state & 1) as u8) << bit;
+                    bit += 1;
+                    if bit == 8 {
+                        buffer.push(byte);
+                        byte = 0;
+                        bit = 0;
+                    }
+                    *state >>= 1;
+                })
+            } else {
+                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);
+                flush_state(&mut state);
-                state >>= 8;
             }
             state = ((state / prob) << PROB_BITS) + (state % prob) + start;
         }
 
         while state > 0 {
-            buffer.push(state as u8);
+            flush_state(&mut state);
-            state >>= 8;
+        }
+
+        drop(flush_state);
+
+        if self.use_bitstream && byte != 0 {
+            buffer.push(byte);
         }
 
         buffer.reverse();
@@ -60,10 +109,11 @@ impl RansCoder {
 pub struct CostCounter {
     cost: f64,
     log2_table: Vec<f64>,
+    invert_bit_encoding: bool,
 }
 
 impl CostCounter {
-    pub fn new() -> CostCounter {
+    pub fn new(config: &Config) -> CostCounter {
         let log2_table = (0..ONE_PROB)
             .map(|prob| {
                 let inv_prob = ONE_PROB as f64 / prob as f64;
@@ -73,6 +123,7 @@ impl CostCounter {
         CostCounter {
             cost: 0.0,
             log2_table,
+            invert_bit_encoding: config.invert_bit_encoding,
         }
     }
 
@@ -87,7 +138,7 @@ impl CostCounter {
 
 impl EntropyCoder for CostCounter {
     fn encode_bit(&mut self, bit: bool, prob: u16) {
-        let prob = if bit {
+        let prob = if bit ^ self.invert_bit_encoding {
             prob as u32
         } else {
             ONE_PROB - prob as u32
@@ -99,14 +150,26 @@ impl EntropyCoder for CostCounter {
 pub struct RansDecoder<'a> {
     data: &'a [u8],
     state: u32,
+    use_bitstream: bool,
+    byte: u8,
+    bits_left: u8,
+    invert_bit_encoding: bool,
+    bitstream_is_big_endian: bool,
 }
 
 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> {
+    pub fn new(data: &'a [u8], config: &Config) -> RansDecoder<'a> {
-        RansDecoder { data, state: 0 }
+        RansDecoder {
+            data,
+            state: 0,
+            use_bitstream: config.use_bitstream,
+            byte: 0,
+            bits_left: 0,
+            invert_bit_encoding: config.invert_bit_encoding,
+            bitstream_is_big_endian: config.bitstream_is_big_endian,
+        }
     }
 
     pub fn decode_with_context(&mut self, context: &mut Context) -> bool {
@@ -117,9 +180,27 @@ impl<'a> RansDecoder<'a> {
 
     pub fn decode_bit(&mut self, prob: u16) -> bool {
         let prob = prob as u32;
-        while self.state < L {
+        if self.use_bitstream {
-            self.state = (self.state << 8) | self.data[0] as u32;
+            while self.state < 32768 {
-            self.data = &self.data[1..];
+                if self.bits_left == 0 {
+                    self.byte = self.data[0];
+                    self.data = &self.data[1..];
+                    self.bits_left = 8;
+                }
+                if self.bitstream_is_big_endian {
+                    self.state = (self.state << 1) | (self.byte >> 7) as u32;
+                    self.byte <<= 1;
+                } else {
+                    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;
@@ -131,6 +212,6 @@ impl<'a> RansDecoder<'a> {
         };
         self.state = prob * (self.state >> PROB_BITS) + (self.state & PROB_MASK) - start;
 
-        bit
+        bit ^ self.invert_bit_encoding
     }
 }