better parity contexts integration for literal length

missing 512 byte table generator, which doesn't look trivial to do
(especially in terms of code size).
Not tested, but looks as decent speed up.

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"
@@ -90,10 +96,22 @@ dependencies = [
 ]
 
 [[package]]
-name = "pico-args"
+name = "proc-macro2"
-version = "0.4.2"
+version = "1.0.44"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "db8bcd96cb740d03149cbad5518db9fd87126a10ab519c011893b1754134c468"
+checksum = "7bd7356a8122b6c4a24a82b278680c73357984ca2fc79a0f9fa6dea7dced7c58"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbe448f377a7d6961e30f5955f9b8d106c3f5e449d493ee1b125c1d43c2b5179"
+dependencies = [
+ "proc-macro2",
+]
 
 [[package]]
 name = "sacabase"
@@ -104,6 +122,37 @@ dependencies = [
  "num-traits",
 ]
 
+[[package]]
+name = "syn"
+version = "1.0.101"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e90cde112c4b9690b8cbe810cba9ddd8bc1d7472e2cae317b69e9438c1cba7d2"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "thiserror"
+version = "1.0.36"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0a99cb8c4b9a8ef0e7907cd3b617cc8dc04d571c4e73c8ae403d80ac160bb122"
+dependencies = [
+ "thiserror-impl",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "1.0.36"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a891860d3c8d66fec8e73ddb3765f90082374dbaaa833407b904a94f1a7eb43"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
 [[package]]
 name = "time"
 version = "0.1.44"
@@ -115,14 +164,21 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "unicode-ident"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dcc811dc4066ac62f84f11307873c4850cb653bfa9b1719cee2bd2204a4bc5dd"
+
 [[package]]
 name = "upkr"
-version = "0.1.0"
+version = "0.2.0-pre3"
 dependencies = [
  "anyhow",
  "cdivsufsort",
+ "lexopt",
  "pbr",
- "pico-args",
+ "thiserror",
 ]
 
 [[package]]

Cargo.toml

@@ -1,12 +1,14 @@
 [package]
 name = "upkr"
-version = "0.1.0"
+version = "0.2.0-pre3"
 edition = "2021"
 
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+[profile.release]
+strip = "debuginfo"
 
 [dependencies]
 cdivsufsort = "2"
-pico-args = "0.4"
+lexopt = "0.2.1"
 anyhow = "1"
+thiserror = "1.0.36"
 pbr = "1"

asm_unpackers/.gitignore

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

asm_unpackers/Makefile

@@ -0,0 +1,64 @@
+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_arm32: ../c_unpacker/main.c unpack_arm32.S
+	mkdir -p build
+	arm-linux-gnueabihf-gcc -g -static -o $@ $^
+
+test_arm32: build/unpack_arm32
+	qemu-arm $< test_data.upk /tmp/out.bin
+	cmp test_data.bin /tmp/out.bin
+
+build/unpack_arm32.bin: unpack_arm32.S
+	mkdir -p build
+	arm-none-eabi-gcc -c -o build/unpack_arm32.o $?
+	arm-none-eabi-objcopy -O binary --only-section=.text build/unpack_arm32.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 build/unpack_arm32.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_arm32.S

@@ -0,0 +1,100 @@
+.arm
+
+.section .text
+
+.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 .. context index
+// r6 .. decode_length temp
+// r7 .. probs ptr
+// r8-r11 .. decode_bit temp
+// r12 .. decode_length return address
+upkr_unpack:
+    push { r3-r11, lr }
+
+	mov r2, #384
+    mov r3, #128
+.Lclear:
+    subs r2, r2, #1
+    strb r3, [sp, -r2]
+    bne .Lclear
+
+.Lloop:
+    mov r5, #0
+    bl upkr_decode_bit
+    bcc .Ldata
+.Lmatch:
+	mov r5, #256
+    rsbs r6, r4, #0
+    blcc upkr_decode_bit
+    bcc .Lskip_offset
+
+    bl upkr_decode_length
+    adds r3, r4, #1
+    popeq { r3-r11, pc }
+.Lskip_offset:
+
+	mov r5, #256+64
+    bl upkr_decode_length
+.Lcopy_loop:
+    ldrb r5, [r0, r3]
+.Lstore:
+    strb r5, [r0], #1
+    adds r4, r4, #1
+    blt .Lcopy_loop
+    b .Lloop
+
+.Ldata:
+    mov r5, #1
+
+.Ldata_loop:
+    bl upkr_decode_bit
+    adc r5, r5, r5
+	movs r4, r5, lsr #8
+    beq .Ldata_loop
+	b .Lstore
+
+.type upkr_decode_length, %function
+upkr_decode_length:
+	mov r12, lr
+
+    mov r4, #0
+	mvn r6, #0
+.Lbit_loop:
+    bl upkr_decode_bit_inc
+	addcc r4, r4, r6
+	movcc pc, r12
+
+    bl upkr_decode_bit_inc
+    addcs r4, r4, r6
+	mov r6, r6, lsl #1
+    b .Lbit_loop
+
+.type upkr_decode_bit, %function
+upkr_decode_bit_inc:
+	add r5, r5, #1
+upkr_decode_bit:
+	cmp r2, #4096
+	ldrltb r8, [r1], #1
+	orrlt r2, r8, r2, lsl#8
+	blt upkr_decode_bit
+
+    ldrb r8, [sp, -r5]
+	and r9, r2, #255
+	add r9, r9, #1
+	cmp r8, r9
+	rsbcs r8, r8, #256
+	mvn r9, r2, lsr#8
+	addcs r9, r9, #1
+	mla r2, r8, r9, r2
+	add r9, r8, #8
+	sub r8, r8, r9, lsr#4
+	rsbcs r8, r8, #256
+	strb r8, [sp, -r5]
+	mov pc, r14
+

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 +1,131 @@
+.section .text
+
+// x9 prev was literal
+// x10 out ptr
+// x11 in ptr
+// x12 offset
+// x13 state
+// x14 context index
+
+.global upkr_unpack
+.type upkr_unpack, %function
+upkr_unpack:
+	mv t4, ra
+	mv x17, x8
+	mv t6, x9
+	li x9, 256 + 128
+	mv x13, x9
+1:
+	sub x8, sp, x13
+	sb x9, 0(x8)
+	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
+	jalr ra // jal 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:
+	jalr ra // jal upkr_decode_number
+	addi t0, x9, 1
+	bnez t0, .Lfinished_offset
+.Ldone:
+	mv x8, x17
+	mv x9, t6
+	jr t4
+
+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
+
+	addi x14, x14, 1
+	
+	sub t2, sp, x14
+	lbu x12, (t2)
+	
+	andi x8, x13, 255
+	sltu x15, x8, x12
+	beqz x15, 1f
+	xori x12, x12, 255
+	addi x12, x12, 1
+1:
+	srli x8, x13, 8
+	addi x8, x8, 1
+	sub x8, x8, x15
+	mul x8, x8, x12
+	sub x13, x13, x8
+	
+	addi x8, x12, 8
+	srli x8, x8, 4
+	sub x12, x12, x8
+	beqz x15, 1f
+	sub x12, x0, x12
+1:
+
+	sb x12, (t2)
+	
+	jalr ra
+
+// x14 context index
+// return: x9 negtive decoded number 
+upkr_decode_number:
+	mv t3, ra
+	mv t5, x14
+	li x9, 0
+	li t1, -1
+1:
+	jal upkr_decode_bit
+	beqz x15, 1f
+	jal upkr_decode_bit
+	beqz x15, 2f
+	add x9, x9, t1
+2:
+	add t1, t1, t1
+	j 1b
+1:
+	add x9, x9, t1
+	
+	mv x14, t5
+	jr t3

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

@@ -1,7 +1,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-int upkr_unpack(void* destination, void* compressed_data);
+void* upkr_unpack(void* destination, void* compressed_data);
 
 int main(int argn, char** argv) {
   void* input_buffer = malloc(1024*1024);
@@ -13,7 +13,8 @@ int main(int argn, char** argv) {
   
   printf("Compressed size: %d\n", in_size);
   
-  int out_size = upkr_unpack(output_buffer, input_buffer);
+  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);
   

c_unpacker/unpack.c

@@ -19,8 +19,8 @@ int upkr_decode_bit(int context_index) {
             upkr_current_byte = *upkr_data_ptr++;
             upkr_bits_left = 8;
         }
-        upkr_state = (upkr_state << 1) + (upkr_current_byte >> 7);
+        upkr_state = (upkr_state << 1) + (upkr_current_byte & 1);
-        upkr_current_byte <<= 1;
+        upkr_current_byte >>= 1;
         --upkr_bits_left;
     }
 #else
@@ -30,18 +30,16 @@ int upkr_decode_bit(int context_index) {
 #endif
    
     int prob = upkr_probs[context_index];
-    int bit = (upkr_state & 255) >= prob ? 1 : 0;
+    int bit = (upkr_state & 255) < prob ? 1 : 0;
     
-    int prob_offset = 16;
-    int state_offset = 0;
-    int state_scale = prob;
     if(bit) {
-        state_offset = -prob;
+        upkr_state = prob * (upkr_state >> 8) + (upkr_state & 255);
-        state_scale = 256 - prob;
+        prob += (256 - prob + 8) >> 4;
-        prob_offset = 0;
+    } else {
+        upkr_state = (256 - prob) * (upkr_state >> 8) + (upkr_state & 255) - prob;
+        prob -= (prob + 8) >> 4;
     }
-    upkr_state = state_offset + state_scale * (upkr_state >> 8) + (upkr_state & 255);
+    upkr_probs[context_index] = prob;
-    upkr_probs[context_index] = prob_offset + prob - ((prob + 8) >> 4); 
 
     return bit;
 }
@@ -56,7 +54,7 @@ int upkr_decode_length(int context_index) {
     return length | (1 << bit_pos);
 }
 
-int upkr_unpack(void* destination, void* compressed_data) {
+void* upkr_unpack(void* destination, void* compressed_data) {
     upkr_data_ptr = (u8*)compressed_data;
     upkr_state = 0;
 #ifdef UPKR_BITSTREAM
@@ -94,5 +92,5 @@ int upkr_unpack(void* destination, void* compressed_data) {
         }
     }
     
-    return write_ptr - (u8*)destination;
+    return write_ptr;
 }

compare-variants

@@ -1,50 +0,0 @@
-#!/bin/env ruby
-
-configs = [
-    [:master, '-b'],
-    [:z80, '-b'],
-    [:z80, ['-b', '-r']],
-    ['old-prob-update', '-b']
-]
-
-files = Dir[ARGV[0] + '/*'].select {|f| !(f =~ /\.txt$/) }
-short_names = files.map {|f| File.basename(f)[..16] }
-results = []
-
-def print_results(configs, names, results)
-    configs.each_with_index do |config, i|
-        printf "%d: %s\n", i + 1, config
-    end
-    
-    print '                '
-    configs.each_index do |i|
-        printf "  %-4d", i + 1
-    end
-    puts
-    names.each_with_index do |name, i|
-        printf "%16s", name
-        for res in results
-            res = res[i]
-            printf "  %-4s", res if res
-        end
-        puts
-    end
-end
-
-for config in configs
-    raise unless system('git', 'checkout', config[0].to_s)
-    config_results = []
-    results << config_results
-    for file in files
-        if system(*['cargo', 'run', '--release', 'pack', '-l', '9', config[1], file, '/tmp/out.upk'].flatten) &&
-            system(*['cargo', 'run', '--release', 'unpack', config[1], '/tmp/out.upk', '/tmp/out.bin'].flatten) &&
-            File.read(file) == File.read('/tmp/out.bin')
-            size = File.size('/tmp/out.upk')
-            config_results << size
-        else
-            config_results << 'ERR'
-        end
-        print_results(configs, short_names, results)
-    end
-end
-

fuzz/.gitignore

@@ -0,0 +1,3 @@
+target
+corpus
+artifacts

fuzz/Cargo.lock

@@ -0,0 +1,247 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 3
+
+[[package]]
+name = "anyhow"
+version = "1.0.65"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "98161a4e3e2184da77bb14f02184cdd111e83bbbcc9979dfee3c44b9a85f5602"
+
+[[package]]
+name = "arbitrary"
+version = "1.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f44124848854b941eafdb34f05b3bcf59472f643c7e151eba7c2b69daa469ed5"
+
+[[package]]
+name = "autocfg"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d468802bab17cbc0cc575e9b053f41e72aa36bfa6b7f55e3529ffa43161b97fa"
+
+[[package]]
+name = "cc"
+version = "1.0.73"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2fff2a6927b3bb87f9595d67196a70493f627687a71d87a0d692242c33f58c11"
+dependencies = [
+ "jobserver",
+]
+
+[[package]]
+name = "cdivsufsort"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "edefce019197609da416762da75bb000bbd2224b2d89a7e722c2296cbff79b8c"
+dependencies = [
+ "cc",
+ "sacabase",
+]
+
+[[package]]
+name = "cfg-if"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
+
+[[package]]
+name = "crossbeam-channel"
+version = "0.5.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c2dd04ddaf88237dc3b8d8f9a3c1004b506b54b3313403944054d23c0870c521"
+dependencies = [
+ "cfg-if",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-utils"
+version = "0.8.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "51887d4adc7b564537b15adcfb307936f8075dfcd5f00dde9a9f1d29383682bc"
+dependencies = [
+ "cfg-if",
+ "once_cell",
+]
+
+[[package]]
+name = "jobserver"
+version = "0.1.25"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "068b1ee6743e4d11fb9c6a1e6064b3693a1b600e7f5f5988047d98b3dc9fb90b"
+dependencies = [
+ "libc",
+]
+
+[[package]]
+name = "lexopt"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "478ee9e62aaeaf5b140bd4138753d1f109765488581444218d3ddda43234f3e8"
+
+[[package]]
+name = "libc"
+version = "0.2.133"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c0f80d65747a3e43d1596c7c5492d95d5edddaabd45a7fcdb02b95f644164966"
+
+[[package]]
+name = "libfuzzer-sys"
+version = "0.4.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae185684fe19814afd066da15a7cc41e126886c21282934225d9fc847582da58"
+dependencies = [
+ "arbitrary",
+ "cc",
+ "once_cell",
+]
+
+[[package]]
+name = "num-traits"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "578ede34cf02f8924ab9447f50c28075b4d3e5b269972345e7e0372b38c6cdcd"
+dependencies = [
+ "autocfg",
+]
+
+[[package]]
+name = "once_cell"
+version = "1.15.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e82dad04139b71a90c080c8463fe0dc7902db5192d939bd0950f074d014339e1"
+
+[[package]]
+name = "pbr"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ff5751d87f7c00ae6403eb1fcbba229b9c76c9a30de8c1cf87182177b168cea2"
+dependencies = [
+ "crossbeam-channel",
+ "libc",
+ "time",
+ "winapi",
+]
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7bd7356a8122b6c4a24a82b278680c73357984ca2fc79a0f9fa6dea7dced7c58"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbe448f377a7d6961e30f5955f9b8d106c3f5e449d493ee1b125c1d43c2b5179"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "sacabase"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9883fc3d6ce3d78bb54d908602f8bc1f7b5f983afe601dabe083009d86267a84"
+dependencies = [
+ "num-traits",
+]
+
+[[package]]
+name = "syn"
+version = "1.0.101"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e90cde112c4b9690b8cbe810cba9ddd8bc1d7472e2cae317b69e9438c1cba7d2"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "thiserror"
+version = "1.0.36"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0a99cb8c4b9a8ef0e7907cd3b617cc8dc04d571c4e73c8ae403d80ac160bb122"
+dependencies = [
+ "thiserror-impl",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "1.0.36"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a891860d3c8d66fec8e73ddb3765f90082374dbaaa833407b904a94f1a7eb43"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "time"
+version = "0.1.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6db9e6914ab8b1ae1c260a4ae7a49b6c5611b40328a735b21862567685e73255"
+dependencies = [
+ "libc",
+ "wasi",
+ "winapi",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dcc811dc4066ac62f84f11307873c4850cb653bfa9b1719cee2bd2204a4bc5dd"
+
+[[package]]
+name = "upkr"
+version = "0.2.0-pre3"
+dependencies = [
+ "anyhow",
+ "cdivsufsort",
+ "lexopt",
+ "pbr",
+ "thiserror",
+]
+
+[[package]]
+name = "upkr-fuzz"
+version = "0.0.0"
+dependencies = [
+ "libfuzzer-sys",
+ "upkr",
+]
+
+[[package]]
+name = "wasi"
+version = "0.10.0+wasi-snapshot-preview1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a143597ca7c7793eff794def352d41792a93c481eb1042423ff7ff72ba2c31f"
+
+[[package]]
+name = "winapi"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c839a674fcd7a98952e593242ea400abe93992746761e38641405d28b00f419"
+dependencies = [
+ "winapi-i686-pc-windows-gnu",
+ "winapi-x86_64-pc-windows-gnu",
+]
+
+[[package]]
+name = "winapi-i686-pc-windows-gnu"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ac3b87c63620426dd9b991e5ce0329eff545bccbbb34f3be09ff6fb6ab51b7b6"
+
+[[package]]
+name = "winapi-x86_64-pc-windows-gnu"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f"

fuzz/Cargo.toml

@@ -0,0 +1,31 @@
+[package]
+name = "upkr-fuzz"
+version = "0.0.0"
+authors = ["Automatically generated"]
+publish = false
+edition = "2018"
+
+[package.metadata]
+cargo-fuzz = true
+
+[dependencies]
+libfuzzer-sys = "0.4"
+
+[dependencies.upkr]
+path = ".."
+
+# Prevent this from interfering with workspaces
+[workspace]
+members = ["."]
+
+[[bin]]
+name = "all_configs"
+path = "fuzz_targets/all_configs.rs"
+test = false
+doc = false
+
+[[bin]]
+name = "unpack"
+path = "fuzz_targets/unpack.rs"
+test = false
+doc = false

fuzz/fuzz_targets/all_configs.rs

@@ -0,0 +1,29 @@
+#![no_main]
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(|data: &[u8]| {
+    let mut config = upkr::Config::default();
+    let mut level = 1;
+    let mut data = data;
+    if data.len() > 2 {
+        let flags1 = data[0];
+        let flags2 = data[1];
+        data = &data[2..];
+        config.use_bitstream = (flags1 & 1) != 0;
+        config.parity_contexts = if (flags1 & 2) == 0 { 1 } else { 2 };
+        config.invert_bit_encoding = (flags1 & 4) != 0;
+        config.is_match_bit = (flags1 & 8) != 0;
+        config.new_offset_bit = (flags1 & 16) != 0;
+        config.continue_value_bit = (flags1 & 32) != 0;
+        config.bitstream_is_big_endian = (flags1 & 64) != 0;
+        config.simplified_prob_update = (flags1 & 128) != 0;
+        config.no_repeated_offsets = (flags2 & 32) != 0;
+        config.eof_in_length = (flags2 & 1) != 0;
+        config.max_offset = if (flags2 & 2) == 0 { usize::MAX } else { 32 };
+        config.max_length = if (flags2 & 4) == 0 { usize::MAX } else { 5 };
+        level = (flags2 >> 3) & 3;
+    }
+    let packed = upkr::pack(data, level, &config, None);
+    let unpacked = upkr::unpack(&packed, &config, 1024 * 1024).unwrap();
+    assert!(unpacked == data);
+});

fuzz/fuzz_targets/unpack.rs

@@ -0,0 +1,6 @@
+#![no_main]
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(|data: &[u8]| {
+    let _ = upkr::unpack(data, &upkr::Config::default(), 64 * 1024);
+});

src/context_state.rs

@@ -1,12 +1,17 @@
-use crate::rans::{ONE_PROB, PROB_BITS};
+use crate::{
+    rans::{ONE_PROB, PROB_BITS},
+    Config,
+};
 
 const INIT_PROB: u16 = 1 << (PROB_BITS - 1);
-const UPDATE_RATE: i32 = 4;
+const UPDATE_RATE: u32 = 4;
-const UPDATE_ADD: i32 = 8;
+const UPDATE_ADD: u32 = 8;
 
 #[derive(Clone)]
 pub struct ContextState {
     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 as u8; size],
+            invert_bit_encoding: config.invert_bit_encoding,
+            simplified_prob_update: config.simplified_prob_update,
         }
     }
 
@@ -33,13 +40,21 @@ impl<'a> Context<'a> {
 
     pub fn update(&mut self, bit: bool) {
         let old = self.state.contexts[self.index];
-        let offset = if !bit {
-            ONE_PROB as i32 >> UPDATE_RATE
-        } else {
-            0
-        };
 
-        self.state.contexts[self.index] =
+        self.state.contexts[self.index] = if self.state.simplified_prob_update {
-            (offset + old as i32 - ((old as i32 + UPDATE_ADD) >> UPDATE_RATE)) as u8;
+            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 {
+            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

@@ -1,16 +1,18 @@
-use crate::lz;
 use crate::match_finder::MatchFinder;
 use crate::rans::RansCoder;
 use crate::ProgressCallback;
+use crate::{lz, Config};
 
 pub fn pack(
     data: &[u8],
-    use_bitstream: bool,
+    config: &Config,
     mut progress_callback: Option<ProgressCallback>,
 ) -> Vec<u8> {
     let mut match_finder = MatchFinder::new(data);
-    let mut rans_coder = RansCoder::new(use_bitstream);
+    let mut rans_coder = RansCoder::new(config);
-    let mut state = lz::CoderState::new();
+    let mut state = lz::CoderState::new(config);
+
+    let mut literal = vec![];
 
     let mut pos = 0;
     while pos < data.len() {
@@ -19,15 +21,20 @@ pub fn pack(
         }
         let mut encoded_match = false;
         if let Some(m) = match_finder.matches(pos).next() {
-            let max_offset = 1 << (m.length * 3 - 1).min(31);
+            let max_offset = config.max_offset.min(1 << (m.length * 3 - 1).min(31));
             let offset = pos - m.pos;
-            if offset < max_offset {
+            if offset < max_offset && m.length >= config.min_length() {
+                if !literal.is_empty() {
+                    lz::Op::Literal(literal).encode(&mut rans_coder, &mut state, config);
+                    literal = vec![];
+                }
+                let length = m.length.min(config.max_length);
                 lz::Op::Match {
                     offset: offset as u32,
-                    len: m.length as u32,
+                    len: length as u32,
                 }
-                .encode(&mut rans_coder, &mut state);
+                .encode(&mut rans_coder, &mut state, config);
-                pos += m.length;
+                pos += length;
                 encoded_match = true;
             }
         }
@@ -39,13 +46,18 @@ pub fn pack(
                     .iter()
                     .zip(data[(pos - offset)..].iter())
                     .take_while(|(a, b)| a == b)
-                    .count();
+                    .count()
-                if length > 0 {
+                    .min(config.max_length);
+                if length >= config.min_length() {
+                    if !literal.is_empty() {
+                        lz::Op::Literal(literal).encode(&mut rans_coder, &mut state, config);
+                        literal = vec![];
+                    }
                     lz::Op::Match {
                         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;
                 }
@@ -53,11 +65,14 @@ pub fn pack(
         }
 
         if !encoded_match {
-            lz::Op::Literal(data[pos]).encode(&mut rans_coder, &mut state);
+            literal.push(data[pos]);
             pos += 1;
         }
     }
 
-    lz::encode_eof(&mut rans_coder, &mut state);
+    if !literal.is_empty() {
+        lz::Op::Literal(literal).encode(&mut rans_coder, &mut state, config);
+    }
+    lz::encode_eof(&mut rans_coder, &mut state, config);
     rans_coder.finish()
 }

src/lib.rs

@@ -5,20 +5,73 @@ mod match_finder;
 mod parsing_packer;
 mod rans;
 
-pub use lz::unpack;
+pub use lz::{calculate_margin, unpack, UnpackError};
 
 pub type ProgressCallback<'a> = &'a mut dyn FnMut(usize);
 
+#[derive(Debug)]
+pub struct Config {
+    pub use_bitstream: bool,
+    pub parity_contexts: usize,
+
+    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,
+    pub eof_in_length: bool,
+
+    pub max_offset: usize,
+    pub max_length: usize,
+}
+
+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,
+            eof_in_length: false,
+
+            max_offset: usize::MAX,
+            max_length: usize::MAX,
+        }
+    }
+}
+
+impl Config {
+    pub fn min_length(&self) -> usize {
+        if self.eof_in_length {
+            2
+        } else {
+            1
+        }
+    }
+}
+
 pub fn pack(
     data: &[u8],
     level: u8,
-    use_bitstream: bool,
+    config: &Config,
     progress_callback: Option<ProgressCallback>,
 ) -> Vec<u8> {
     if level == 0 {
-        greedy_packer::pack(data, use_bitstream, progress_callback)
+        greedy_packer::pack(data, config, progress_callback)
     } else {
-        parsing_packer::pack(data, level, use_bitstream, progress_callback)
+        parsing_packer::pack(data, level, config, progress_callback)
     }
 }
 

src/lz.rs

@@ -1,49 +1,102 @@
 use crate::context_state::ContextState;
 use crate::rans::{EntropyCoder, RansDecoder};
+use crate::Config;
+use thiserror::Error;
 
-#[derive(Copy, Clone, Debug)]
+#[derive(Clone, Debug)]
 pub enum Op {
-    Literal(u8),
+    Literal(Vec<u8>),
     Match { offset: u32, len: u32 },
 }
 
 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) {
         match self {
-            &Op::Literal(lit) => {
+            &Op::Literal(ref lit) => {
-                encode_bit(coder, state, 0, false);
+                assert!(state.prev_was_match);
-                let mut context_index = 1;
+                encode_length(
-                for i in (0..8).rev() {
+                    coder,
-                    let bit = (lit >> i) & 1 != 0;
+                    state,
-                    encode_bit(coder, state, context_index, bit);
+                    256 + state.pos % state.parity_contexts * 320,
-                    context_index = (context_index << 1) | bit as usize;
+                    lit.len() as u32 + 1,
+                    config,
+                );
+                for lit in lit {
+                    let literal_base = state.pos % state.parity_contexts * 320;
+                    let mut context_index = 1;
+                    for i in (0..8).rev() {
+                        let bit = (lit >> i) & 1 != 0;
+                        encode_bit(coder, state, literal_base + context_index, bit);
+                        context_index = (context_index << 1) | bit as usize;
+                    }
+                    state.pos += 1;
                 }
                 state.prev_was_match = false;
             }
             &Op::Match { offset, len } => {
-                encode_bit(coder, state, 0, true);
+                if state.prev_was_match {
-                if !state.prev_was_match {
+                    encode_length(
-                    encode_bit(coder, state, 256, offset != state.last_offset);
+                        coder,
-                } else {
+                        state,
-                    assert!(offset != state.last_offset);
+                        256 + state.pos % state.parity_contexts * 320,
+                        1,
+                        config,
+                    );
                 }
-                if offset != state.last_offset {
+                let mut new_offset = true;
-                    encode_length(coder, state, 257, offset + 1);
+                if !state.prev_was_match && !config.no_repeated_offsets {
+                    new_offset = offset != state.last_offset;
+                    encode_bit(
+                        coder,
+                        state,
+                        320 * state.parity_contexts,
+                        new_offset == config.new_offset_bit,
+                    );
+                }
+                assert!(offset as usize <= config.max_offset);
+                if new_offset {
+                    encode_length(
+                        coder,
+                        state,
+                        320 * state.parity_contexts + 1,
+                        offset + if config.eof_in_length { 0 } else { 1 },
+                        config,
+                    );
                     state.last_offset = offset;
                 }
-                encode_length(coder, state, 257 + 64, len);
+                assert!(len as usize >= config.min_length() && len as usize <= config.max_length);
+                encode_length(coder, state, 320 * 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);
+    if state.prev_was_match {
-    if !state.prev_was_match {
+        encode_length(
-        encode_bit(coder, state, 256, true);
+            coder,
+            state,
+            256 + state.pos % state.parity_contexts * 320,
+            1,
+            config,
+        );
+    }
+    if !state.prev_was_match && !config.no_repeated_offsets {
+        encode_bit(
+            coder,
+            state,
+            320 * state.parity_contexts,
+            config.new_offset_bit ^ config.eof_in_length,
+        );
+    }
+    if !config.eof_in_length || state.prev_was_match || config.no_repeated_offsets {
+        encode_length(coder, state, 320 * state.parity_contexts + 1, 1, config);
+    }
+    if config.eof_in_length {
+        encode_length(coder, state, 320 * state.parity_contexts + 65, 1, config);
     }
-    encode_length(coder, state, 257, 1);
 }
 
 fn encode_bit(
@@ -60,17 +113,18 @@ 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)]
@@ -78,14 +132,18 @@ pub struct CoderState {
     contexts: ContextState,
     last_offset: u32,
     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((64 + 256) * config.parity_contexts + 1 + 64 + 64, config),
             last_offset: 0,
-            prev_was_match: false,
+            prev_was_match: true,
+            pos: 0,
+            parity_contexts: config.parity_contexts,
         }
     }
 
@@ -94,55 +152,146 @@ impl CoderState {
     }
 }
 
-pub fn unpack(packed_data: &[u8], use_bitstream: bool) -> Vec<u8> {
+#[derive(Error, Debug)]
-    let mut decoder = RansDecoder::new(packed_data, use_bitstream);
+pub enum UnpackError {
-    let mut contexts = ContextState::new(1 + 255 + 1 + 64 + 64);
+    #[error("match offset out of range: {offset} > {position}")]
+    OffsetOutOfRange { offset: usize, position: usize },
+    #[error("Unpacked data over size limit: {size} > {limit}")]
+    OverSize { size: usize, limit: usize },
+    #[error("Unexpected end of input data")]
+    UnexpectedEOF {
+        #[from]
+        source: crate::rans::UnexpectedEOF,
+    },
+    #[error("Overflow while reading value")]
+    ValueOverflow,
+}
+
+pub fn unpack(
+    packed_data: &[u8],
+    config: &Config,
+    max_size: usize,
+) -> Result<Vec<u8>, UnpackError> {
     let mut result = vec![];
-    let mut offset = 0;
+    let _ = unpack_internal(Some(&mut result), packed_data, config, max_size)?;
+    Ok(result)
+}
+
+pub fn calculate_margin(packed_data: &[u8], config: &Config) -> Result<isize, UnpackError> {
+    unpack_internal(None, packed_data, config, usize::MAX)
+}
+
+pub fn unpack_internal(
+    mut result: Option<&mut Vec<u8>>,
+    packed_data: &[u8],
+    config: &Config,
+    max_size: usize,
+) -> Result<isize, UnpackError> {
+    let mut decoder = RansDecoder::new(packed_data, &config);
+    let mut contexts =
+        ContextState::new((64 + 256) * config.parity_contexts + 1 + 64 + 64, &config);
+    let mut offset = usize::MAX;
+    let mut position = 0usize;
     let mut prev_was_match = false;
+    let mut margin = 0isize;
 
     fn decode_length(
         decoder: &mut RansDecoder,
         contexts: &mut ContextState,
         mut context_index: usize,
-    ) -> usize {
+        config: &Config,
+    ) -> Result<usize, UnpackError> {
         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))?
-            length |= (decoder.decode_with_context(&mut contexts.context_mut(context_index + 1))
+            == config.continue_value_bit
+        {
+            length |= (decoder.decode_with_context(&mut contexts.context_mut(context_index + 1))?
                 as usize)
                 << bit_pos;
             bit_pos += 1;
+            if bit_pos >= 32 {
+                return Err(UnpackError::ValueOverflow);
+            }
             context_index += 2;
         }
-        length | (1 << bit_pos)
+        Ok(length | (1 << bit_pos))
     }
 
     loop {
-        if decoder.decode_with_context(&mut contexts.context_mut(0)) {
+        margin = margin.max(position as isize - decoder.pos() as isize);
-            if prev_was_match || decoder.decode_with_context(&mut contexts.context_mut(256)) {
+        let literal_length = decode_length(
-                offset = decode_length(&mut decoder, &mut contexts, 257) - 1;
+            &mut decoder,
-                if offset == 0 {
+            &mut contexts,
-                    break;
+            256 + position % config.parity_contexts * 320,
-                }
+            config,
-            }
+        )? - 1;
-            let length = decode_length(&mut decoder, &mut contexts, 257 + 64);
+        for _ in 0..literal_length {
-            for _ in 0..length {
+            let literal_base = position % config.parity_contexts * 320;
-                result.push(result[result.len() - offset]);
-            }
-            prev_was_match = true;
-        } else {
             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;
             }
-            result.push(byte);
+            if let Some(ref mut result) = result {
+                if result.len() < max_size {
+                    result.push(byte);
+                }
+            }
+            position += 1;
             prev_was_match = false;
         }
+
+        if config.no_repeated_offsets
+            || prev_was_match
+            || decoder
+                .decode_with_context(&mut contexts.context_mut(320 * config.parity_contexts))?
+                == config.new_offset_bit
+        {
+            offset = decode_length(
+                &mut decoder,
+                &mut contexts,
+                320 * config.parity_contexts + 1,
+                &config,
+            )? - if config.eof_in_length { 0 } else { 1 };
+            if offset == 0 {
+                break;
+            }
+        }
+        let length = decode_length(
+            &mut decoder,
+            &mut contexts,
+            320 * config.parity_contexts + 65,
+            &config,
+        )?;
+        if config.eof_in_length && length == 1 {
+            break;
+        }
+        if offset > position {
+            return Err(UnpackError::OffsetOutOfRange { offset, position });
+        }
+        if let Some(ref mut result) = result {
+            for _ in 0..length {
+                if result.len() < max_size {
+                    result.push(result[result.len() - offset]);
+                } else {
+                    break;
+                }
+            }
+        }
+        position += length;
+        prev_was_match = true;
     }
 
-    result
+    if position > max_size {
+        return Err(UnpackError::OverSize {
+            size: position,
+            limit: max_size,
+        });
+    }
+
+    Ok(margin + decoder.pos() as isize - position as isize)
 }

src/main.rs

@@ -1,75 +1,184 @@
-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 calculate_margin = false;
+    let mut level = 2;
+    let mut infile: Option<PathBuf> = None;
+    let mut outfile: Option<PathBuf> = None;
+    let mut max_unpacked_size = 512 * 1024 * 1024;
 
-    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 {
-            let use_bitstream = args.contains(["-b", "--bitstream"]);
+            Short('b') | Long("bitstream") => config.use_bitstream = true,
-            let reverse = args.contains(["-r", "--reverse"]);
+            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,
+            Long("eof-in-length") => config.eof_in_length = true,
 
-            let infile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+            Long("max-offset") => config.max_offset = parser.value()?.parse()?,
-            let outfile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+            Long("max-length") => config.max_length = parser.value()?.parse()?,
 
-            let mut data = vec![];
+            Long("z80") => {
-            File::open(infile)?.read_to_end(&mut data)?;
+                config.use_bitstream = true;
-            if reverse {
+                config.bitstream_is_big_endian = true;
-                data.reverse();
+                config.invert_bit_encoding = true;
+                config.simplified_prob_update = true;
+                level = 9;
+            }
+            Long("x86") => {
+                config.use_bitstream = true;
+                config.continue_value_bit = false;
+                config.is_match_bit = false;
+                config.new_offset_bit = false;
+            }
+            Long("x86b") => {
+                config.use_bitstream = true;
+                config.continue_value_bit = false;
+                config.no_repeated_offsets = true;
+                level = 9;
             }
 
-            let mut pb = pbr::ProgressBar::new(data.len() as u64);
+            Short('u') | Long("unpack") => unpack = true,
-            pb.set_units(pbr::Units::Bytes);
+            Long("margin") => calculate_margin = true,
-            let packed_data = upkr::pack(
+            Short('l') | Long("level") => level = parser.value()?.parse()?,
-                &data,
+            Short(n) if n.is_ascii_digit() => level = n as u8 - b'0',
-                level,
+            Short('h') | Long("help") => print_help(0),
-                use_bitstream,
+            Long("max-unpacked-size") => max_unpacked_size = parser.value()?.parse()?,
-                Some(&mut |pos| {
+            Value(val) if infile.is_none() => infile = Some(val.try_into()?),
-                    pb.set(pos as u64);
+            Value(val) if outfile.is_none() => outfile = Some(val.try_into()?),
-                }),
+            _ => return Err(arg.unexpected().into()),
-            );
-            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 use_bitstream = args.contains(["-b", "--bitstream"]);
-            let reverse = args.contains(["-r", "--reverse"]);
 
-            let infile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+    let infile = infile.unwrap_or_else(|| print_help(1));
-            let outfile = args.free_from_os_str::<PathBuf, bool>(|s| Ok(s.into()))?;
+    let outfile = outfile.unwrap_or_else(|| {
+        let mut name = infile.clone();
+        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);
+        }
+        name
+    });
 
-            let mut data = vec![];
+    if config.parity_contexts != 1 && config.parity_contexts != 2 && config.parity_contexts != 4 {
-            File::open(infile)?.read_to_end(&mut data)?;
+        eprintln!("--parity has to be 1, 2, or 4");
-            let mut unpacked_data = upkr::unpack(&data, use_bitstream);
+        process::exit(1);
+    }
+
+    if !unpack && !calculate_margin {
+        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();
+        }
+        if unpack {
+            let mut unpacked_data = upkr::unpack(&data, &config, max_unpacked_size)?;
             if reverse {
                 unpacked_data.reverse();
             }
             File::create(outfile)?.write_all(&unpacked_data)?;
         }
-        Some(other) => {
+        if calculate_margin {
-            bail!("Unknown subcommand '{}'", other);
+            println!("{}", upkr::calculate_margin(&data, &config)?);
         }
     }
 
     Ok(())
 }
 
-fn print_help() {
+fn print_help(exit_code: i32) -> ! {
     eprintln!("Usage:");
-    eprintln!("  upkr pack [-b] [-l level(0-9)] <infile> <outfile>");
+    eprintln!("  upkr [-l level(0-9)] [config options] <infile> [<outfile>]");
-    eprintln!("  upkr unpack [-b] <infile> <outfile>");
+    eprintln!("  upkr -u [config options] <infile> [<outfile>]");
+    eprintln!("  upkr --margin [config options] <infile>");
     eprintln!();
-    eprintln!(" -b, --bitstream     bitstream mode");
     eprintln!(" -l, --level N       compression level 0-9");
-    std::process::exit(1);
+    eprintln!(" -0, ..., -9         short form for setting compression level");
+    eprintln!(" -u, --unpack        unpack infile");
+    eprintln!(" --margin            calculate margin for overlapped unpacking of a packed file");
+    eprintln!();
+    eprintln!("Config presets for specific unpackers:");
+    eprintln!(" --z80               --big-endian-bitstream --invert-bit-encoding --simplified-prob-update -9");
+    eprintln!(
+        " --x86               --bitstream --invert-is-match-bit --invert-continue-value-bit --invert-new-offset-bit"
+    );
+    eprintln!(
+        " --x86b              --bitstream --invert-continue-value-bit --no-repeated-offsets -9"
+    );
+    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");
+    eprintln!(" --eof-in-length");
+    eprintln!(" --max-offset N");
+    eprintln!(" --max-length N");
+    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, use_bitstream: bool, 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);
+        ops.push(link.op.clone());
         parse = link.prev.clone();
     }
-    let mut state = lz::CoderState::new();
+    let mut state = lz::CoderState::new(config);
-    let mut coder = RansCoder::new(use_bitstream);
+    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()
 }
 
@@ -27,9 +32,15 @@ struct Parse {
     op: lz::Op,
 }
 
+struct LiteralPrefix {
+    arrival: Arrival,
+    prefix: Vec<u8>,
+}
+
 struct Arrival {
     parse: Option<Rc<Parse>>,
     state: lz::CoderState,
+    literal_prefix: Option<Box<LiteralPrefix>>,
     cost: f64,
 }
 
@@ -38,6 +49,7 @@ 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)
@@ -99,17 +111,22 @@ fn parse(
         cost_counter: &mut CostCounter,
         pos: usize,
         offset: usize,
-        length: usize,
+        mut length: usize,
         arrival: &Arrival,
         max_arrivals: usize,
+        config: &crate::Config,
     ) {
+        if length < config.min_length() {
+            return;
+        }
+        length = length.min(config.max_length);
         cost_counter.reset();
         let mut state = arrival.state.clone();
         let op = lz::Op::Match {
             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,
@@ -119,6 +136,7 @@ fn parse(
                     op,
                 })),
                 state,
+                literal_prefix: None,
                 cost: arrival.cost + cost_counter.cost(),
             },
             max_arrivals,
@@ -129,13 +147,14 @@ fn parse(
         0,
         Arrival {
             parse: None,
-            state: lz::CoderState::new(),
+            state: lz::CoderState::new(encoding_config),
+            literal_prefix: None,
             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| {
@@ -176,18 +195,21 @@ fn parse(
             for m in match_finder.matches(pos) {
                 closest_match = Some(closest_match.unwrap_or(0).max(m.pos));
                 let offset = pos - m.pos;
-                found_last_offset |= offset as u32 == arrival.state.last_offset();
+                if offset <= encoding_config.max_offset {
-                add_match(
+                    found_last_offset |= offset as u32 == arrival.state.last_offset();
-                    &mut arrivals,
+                    add_match(
-                    cost_counter,
+                        &mut arrivals,
-                    pos,
+                        cost_counter,
-                    offset,
+                        pos,
-                    m.length,
+                        offset,
-                    &arrival,
+                        m.length,
-                    max_arrivals,
+                        &arrival,
-                );
+                        max_arrivals,
-                if m.length >= config.greedy_size {
+                        encoding_config,
-                    break 'arrival_loop;
+                    );
+                    if m.length >= config.greedy_size {
+                        break 'arrival_loop;
+                    }
                 }
             }
 
@@ -198,6 +220,9 @@ fn parse(
                 && closest_match.iter().all(|p| *p < match_pos)
             {
                 let offset = pos - match_pos;
+                if offset > encoding_config.max_offset {
+                    break;
+                }
                 let length = match_length(offset);
                 assert!(length > 0);
                 add_match(
@@ -208,6 +233,7 @@ fn parse(
                     length,
                     &arrival,
                     max_arrivals,
+                    encoding_config,
                 );
                 found_last_offset |= offset as u32 == arrival.state.last_offset();
                 if offset < near_matches.len() {
@@ -228,24 +254,32 @@ fn parse(
                         length,
                         &arrival,
                         max_arrivals,
+                        encoding_config,
                     );
                 }
             }
 
             cost_counter.reset();
-            let mut state = arrival.state;
+            let (arrival, mut prefix) = if let Some(prefix) = arrival.literal_prefix {
-            let op = lz::Op::Literal(data[pos]);
+                (prefix.arrival, prefix.prefix)
-            op.encode(cost_counter, &mut state);
+            } else {
+                (arrival, vec![])
+            };
+            let mut state = arrival.state.clone();
+            prefix.push(data[pos]);
+            let op = lz::Op::Literal(prefix.clone());
+            op.encode(cost_counter, &mut state, encoding_config);
             add_arrival(
                 &mut arrivals,
                 pos + 1,
                 Arrival {
                     parse: Some(Rc::new(Parse {
-                        prev: arrival.parse,
+                        prev: arrival.parse.clone(),
                         op,
                     })),
                     state,
                     cost: arrival.cost + cost_counter.cost(),
+                    literal_prefix: Some(Box::new(LiteralPrefix { arrival, prefix })),
                 },
                 max_arrivals,
             );

src/rans.rs

@@ -1,4 +1,5 @@
-use crate::context_state::Context;
+use crate::{context_state::Context, Config};
+use thiserror::Error;
 
 pub const PROB_BITS: u32 = 8;
 pub const ONE_PROB: u32 = 1 << PROB_BITS;
@@ -15,20 +16,25 @@ pub trait EntropyCoder {
 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.bits.push(prob | ((bit as u16) << 15));
+        self.bits
+            .push(prob | (((bit ^ self.invert_bit_encoding) as u16) << 15));
     }
 }
 
 impl RansCoder {
-    pub fn new(use_bitstream: bool) -> RansCoder {
+    pub fn new(config: &Config) -> RansCoder {
         RansCoder {
             bits: Vec::new(),
-            use_bitstream,
+            use_bitstream: config.use_bitstream,
+            bitstream_is_big_endian: config.bitstream_is_big_endian,
+            invert_bit_encoding: config.invert_bit_encoding,
         }
     }
 
@@ -38,18 +44,31 @@ impl RansCoder {
         let mut state = 1 << l_bits;
 
         let mut byte = 0u8;
-        let mut bit = 0;
+        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 {
-            Box::new(|state: &mut u32| {
+            if self.bitstream_is_big_endian {
-                byte |= ((*state & 1) as u8) << bit;
+                Box::new(|state: &mut u32| {
-                bit += 1;
+                    byte |= ((*state & 1) as u8) << bit;
-                if bit == 8 {
+                    bit += 1;
-                    buffer.push(byte);
+                    if bit == 8 {
-                    byte = 0;
+                        buffer.push(byte);
-                    bit = 0;
+                        byte = 0;
-                }
+                        bit = 0;
-                *state >>= 1;
+                    }
-            })
+                    *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);
@@ -61,7 +80,7 @@ impl RansCoder {
         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 {
+            let (start, prob) = if step & 32768 != 0 {
                 (0, prob)
             } else {
                 (prob, ONE_PROB - prob)
@@ -91,10 +110,11 @@ 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;
@@ -104,6 +124,7 @@ impl CostCounter {
         CostCounter {
             cost: 0.0,
             log2_table,
+            invert_bit_encoding: config.invert_bit_encoding,
         }
     }
 
@@ -118,7 +139,7 @@ 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
@@ -129,60 +150,85 @@ impl EntropyCoder for CostCounter {
 
 pub struct RansDecoder<'a> {
     data: &'a [u8],
+    pos: usize,
     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;
 
+#[derive(Debug, Error)]
+#[error("Unexpected end of input")]
+pub struct UnexpectedEOF;
+
 impl<'a> RansDecoder<'a> {
-    pub fn new(data: &'a [u8], use_bitstream: bool) -> RansDecoder<'a> {
+    pub fn new(data: &'a [u8], config: &Config) -> RansDecoder<'a> {
         RansDecoder {
             data,
+            pos: 0,
             state: 0,
-            use_bitstream,
+            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 {
+    pub fn pos(&self) -> usize {
-        let bit = self.decode_bit(context.prob());
+        self.pos
-        context.update(bit);
-        bit
     }
 
-    pub fn decode_bit(&mut self, prob: u16) -> bool {
+    pub fn decode_with_context(&mut self, context: &mut Context) -> Result<bool, UnexpectedEOF> {
+        let bit = self.decode_bit(context.prob())?;
+        context.update(bit);
+        Ok(bit)
+    }
+
+    pub fn decode_bit(&mut self, prob: u16) -> Result<bool, UnexpectedEOF> {
         let prob = prob as u32;
         if self.use_bitstream {
             while self.state < 32768 {
                 if self.bits_left == 0 {
-                    self.byte = self.data[0];
+                    if self.pos >= self.data.len() {
-                    self.data = &self.data[1..];
+                        return Err(UnexpectedEOF);
+                    }
+                    self.byte = self.data[self.pos];
+                    self.pos += 1;
                     self.bits_left = 8;
                 }
-                self.state = (self.state << 1) | (self.byte >> 7) as u32;
+                if self.bitstream_is_big_endian {
-                self.byte <<= 1;
+                    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;
+                if self.pos >= self.data.len() {
-                self.data = &self.data[1..];
+                    return Err(UnexpectedEOF);
+                }
+                self.state = (self.state << 8) | self.data[self.pos] as u32;
+                self.pos += 1;
             }
         }
 
-        let bit = (self.state & PROB_MASK) >= prob;
+        let bit = (self.state & PROB_MASK) < prob;
 
         let (start, prob) = if bit {
-            (prob, ONE_PROB - prob)
-        } else {
             (0, prob)
+        } else {
+            (prob, ONE_PROB - prob)
         };
         self.state = prob * (self.state >> PROB_BITS) + (self.state & PROB_MASK) - start;
 
-        bit
+        Ok(bit ^ self.invert_bit_encoding)
     }
 }

z80_unpacker/example/example.asm

@@ -3,7 +3,8 @@
     DEVICE ZXSPECTRUM48,$8FFF
 
     ORG     $9000
-compressed_scr_files:       ; border color byte + upkr-packed .scr file
+  ;; forward example data
+compressed_scr_files.fwd:               ; border color byte + upkr-packed .scr file
     DB      1
     INCBIN  "screens/Grongy - ZX Spectrum (2022).scr.upk"
     DB      7
@@ -13,37 +14,87 @@ compressed_scr_files:       ; border color byte + upkr-packed .scr file
     DB      6
     INCBIN  "screens/diver - Back to Bjork (2015).scr.upk"
 .e:
+  ;; backward example data (unpacker goes from the end of the data!)
+compressed_scr_files.rwd.e: EQU $-1     ; the final IX will point one byte ahead of "$" here
+    INCBIN  "screens.reversed/diver - Back to Bjork (2015).scr.upk"
+    DB      6
+    INCBIN  "screens.reversed/diver - Mercenary 4. The Heaven's Devil (2014) (Forever 2014 Olympic Edition, 1).scr.upk"
+    DB      0
+    INCBIN  "screens.reversed/Schafft - Poison (2017).scr.upk"
+    DB      7
+    INCBIN  "screens.reversed/Grongy - ZX Spectrum (2022).scr.upk"
+compressed_scr_files.rwd:               ; border color byte + upkr-packed .scr file (backward)
+    DB      1
 
 start:
     di
 ;     OPT --zxnext
-;     nextreg 7,3                 ; ZX Next: switch to 28Mhz
+;     nextreg 7,3                       ; ZX Next: switch to 28Mhz
-    ld      ix,compressed_scr_files
+
-.slideshow_loop
+  ;;; FORWARD packed/unpacked data demo
+    ld      ix,compressed_scr_files.fwd
+.slideshow_loop.fwd:
   ; set BORDER for next image
-    ldi     a,(ix)              ; fake: ld a,(ix) : inc ix
+    ld      a,(ix)
+    inc     ix
     out     (254),a
   ; call unpack of next image directly into VRAM
-    ld      de,$4000            ; target VRAM
+    ld      de,$4000                    ; target VRAM
     exx
   ; IX = packed data, DE' = destination ($4000)
   ; returned IX will point right after the packed data
-    call    upkr.unpack
+    call    fwd.upkr.unpack
   ; do some busy loop with CPU to delay between images
+    call    delay
+  ; check if all images were displayed, loop around from first one then
+    ld      a,ixl
+    cp      low compressed_scr_files.fwd.e
+    jr      nz,.slideshow_loop.fwd
+
+  ;;; BACKWARD packed/unpacked data demo
+    ld      ix,compressed_scr_files.rwd
+.slideshow_loop.rwd:
+  ; set BORDER for next image
+    ld      a,(ix)
+    dec     ix
+    out     (254),a
+  ; call unpack of next image directly into VRAM
+    ld      de,$5AFF                    ; target VRAM
+    exx
+  ; IX = packed data, DE' = destination
+  ; returned IX will point right ahead of the packed data
+    call    rwd.upkr.unpack
+  ; do some busy loop with CPU to delay between images
+    call    delay
+  ; check if all images were displayed, loop around from first one then
+    ld      a,ixl
+    cp      low compressed_scr_files.rwd.e
+    jr      nz,.slideshow_loop.rwd
+
+    jr      start
+
+delay:
     ld      bc,$AA00
 .delay:
     .8 ex      (sp),ix
     dec     c
     jr      nz,.delay
     djnz    .delay
-  ; check if all images were displayed, loop around from first one then
+    ret
-    ld      a,ixl
-    cp      low compressed_scr_files.e
-    jr      z,start
-    jr      .slideshow_loop
 
   ; include the depacker library, optionally putting probs array buffer near end of RAM
     DEFINE  UPKR_PROBS_ORIGIN $FA00   ; if not defined, array will be put after unpack code
-    INCLUDE "../unpack.asm"
+
+    MODULE fwd
+        INCLUDE "../unpack.asm"
+    ENDMODULE
+
+    MODULE rwd
+        DEFINE BACKWARDS_UNPACK         ; defined to build backwards unpack
+                ; initial IX points at last byte of compressed data
+                ; initial DE' points at last byte of unpacked data
+
+        INCLUDE "../unpack.asm"
+    ENDMODULE
 
     SAVESNA "example.sna",start

z80_unpacker/readme.txt

@@ -10,10 +10,23 @@ may be incompatible with files you will produce with current version)
 
 Asm syntax is z00m's sjasmplus: https://github.com/z00m128/sjasmplus
 
+Backward direction unpacker added as compile-time option, see example for both forward/backward
+depacker in action.
+
+The packed/unpacked data-overlap has to be tested per-case, in worst case the packed data
+may need even more than 7 bytes to unpack final byte, but usually 1-4 bytes may suffice.
+
 TODO:
-- build base corpus of test data to benchmark future changes in algorithm/format
+- build bigger corpus of test data to benchmark future changes in algorithm/format (example and zx48.rom was used to do initial tests)
-- review first implementation to identify weak spots where the implementation can be shorter+faster
+- maybe try to beat double-loop `decode_number` with different encoding format
-with acceptable small changes to the format
-- review non-bitstream variant, if it's feasible to try to implement it with Z80
 - (@ped7g) Z80N version of unpacker for ZX Next devs
 - (@exoticorn) add Z80 specific packer (to avoid confusion with original MicroW8 variant), and land it all to master branch, maybe in "z80" directory or something? (and overall decide how to organise+merge this upstream into main repo)
+- (@exoticorn) add to packer output with possible packed/unpacked region overlap
+
+DONE:
+* review non-bitstream variant, if it's feasible to try to implement it with Z80
+    - Ped7g: IMHO nope, the 12b x 8b MUL code would probably quickly cancel any gains from the simpler state update
+* review first implementation to identify weak spots where the implementation can be shorter+faster
+with acceptable small changes to the format
+    - Ped7g: the decode_bit settled down and now doesn't feel so confused and redundant, the code seems pretty on point to me, no obvious simplification from format change
+    - Ped7g: the decode_number double-loop is surprisingly resilient, especially in terms of code size I failed to beat it, speed wise only negligible gains

z80_unpacker/unpack.asm

@@ -4,7 +4,7 @@
 ;; initial version by Peter "Ped" Helcmanovsky (C) 2022, licensed same as upkr project ("unlicensed")
 ;; to assemble use z00m's sjasmplus: https://github.com/z00m128/sjasmplus
 ;;
-;; you can define UPKR_PROBS_ORIGIN to specific 256 byte aligned address for probs array (386 bytes),
+;; you can define UPKR_PROBS_ORIGIN to specific 256 byte aligned address for probs array (320 bytes),
 ;; otherwise it will be positioned after the unpacker code (256 aligned)
 ;;
 ;; public API:
@@ -15,6 +15,15 @@
 ;;         modifies: all registers except IY, requires 10 bytes of stack space
 ;;
 
+;     DEFINE BACKWARDS_UNPACK         ; uncomment to build backwards depacker (write_ptr--, upkr_data_ptr--)
+            ; initial IX points at last byte of compressed data
+            ; initial DE' points at last byte of unpacked data
+
+;     DEFINE UPKR_UNPACK_SPEED        ; uncomment to get larger but faster unpack routine
+
+; code size hint: if you put probs array just ahead of BASIC entry point, you will get BC
+; initialised to probs.e by BASIC `USR` command and you can remove it from unpack init (-3B)
+
     OPT push reset --syntax=abf
     MODULE upkr
 
@@ -100,7 +109,7 @@ unpack:
     ld      a,c
     exx
     ld      (de),a              ; *write_ptr++ = byte;
-    inc     de
+    IFNDEF BACKWARDS_UNPACK : inc de : ELSE : dec de : ENDIF
     exx
     ld      d,b                 ; prev_was_match = false
     jr      .decompress_data
@@ -116,8 +125,7 @@ unpack:
     cp      d                   ; CF = prev_was_match
     call    nc,decode_bit       ; if not prev_was_match, then upkr_decode_bit(256)
     jr      nc,.keep_offset     ; if neither, keep old offset
-    inc     c                   ; context_index to first "number" set for offsets decoding (257)
+    call    decode_number       ; context_index is already 257-1 as needed by decode_number
-    call    decode_number
     dec     de                  ; offset = upkr_decode_length(257) - 1;
     ld      a,d
     or      e
@@ -130,16 +138,25 @@ unpack:
         ;                 ++write_ptr;
         ;             }
         ;             prev_was_match = 1;
-    ld      c,low(257 + NUMBER_BITS)    ; context_index to second "number" set for lengths decoding
+    ld      c,low(257 + NUMBER_BITS - 1)    ; context_index to second "number" set for lengths decoding
     call    decode_number       ; length = upkr_decode_length(257 + 64);
     push    de
     exx
-    ld      h,d                 ; DE = write_ptr
+    IFNDEF BACKWARDS_UNPACK
-    ld      l,e
+        ; forward unpack (write_ptr++, upkr_data_ptr++)
-.offset+*:  ld      bc,0
+        ld      h,d             ; DE = write_ptr
-    sbc     hl,bc               ; CF=0 from decode_number ; HL = write_ptr - offset
+        ld      l,e
-    pop     bc                  ; BC = length
+.offset+*:  ld  bc,0
-    ldir
+        sbc     hl,bc           ; CF=0 from decode_number ; HL = write_ptr - offset
+        pop     bc              ; BC = length
+        ldir
+    ELSE
+        ; backward unpack (write_ptr--, upkr_data_ptr--)
+.offset+*:  ld  hl,0
+        add     hl,de           ; HL = write_ptr + offset
+        pop     bc              ; BC = length
+        lddr
+    ENDIF
     exx
     ld      d,b                 ; prev_was_match = true
     djnz    .decompress_data    ; adjust context_index back to 0..255 range, go to main loop
@@ -173,6 +190,9 @@ int upkr_decode_bit(int context_index) {
     return bit;
 }
 */
+inc_c_decode_bit:
+  ; ++low(context_index) before decode_bit (to get -1B by two calls in decode_number)
+    inc     c
 decode_bit:
   ; HL = upkr_state
   ; IX = upkr_data_ptr
@@ -193,7 +213,7 @@ decode_bit:
     jr      nz,.has_bit             ; CF=data, ZF=0 -> some bits + stop bit still available
   ; CF=1 (by stop bit)
     ld      a,(ix)
-    inc     ix                      ; upkr_current_byte = *upkr_data_ptr++;
+    IFNDEF BACKWARDS_UNPACK : inc ix : ELSE : dec ix : ENDIF    ; upkr_current_byte = *upkr_data_ptr++;
     adc     a,a                     ; CF=data, b0=1 as new stop bit
 .has_bit:
     adc     hl,hl                   ; upkr_state = (upkr_state << 1) + (upkr_current_byte >> 7);
@@ -215,6 +235,10 @@ decode_bit:
     ld      d,0
     ld      e,a                     ; DE = state_scale ; prob || (256-prob)
     ld      l,d                     ; H:L = (upkr_state>>8) : 0
+
+  IFNDEF UPKR_UNPACK_SPEED
+
+    ;; looped MUL for minimum unpack size
     ld      b,8                     ; counter
 .mulLoop:
     add     hl,hl
@@ -222,28 +246,41 @@ decode_bit:
     add     hl,de
 .mul0:
     djnz    .mulLoop                ; until HL = state_scale * (upkr_state>>8), also BC becomes (upkr_state & 255)
+
+  ELSE
+
+    ;;; unrolled MUL for better performance, +25 bytes unpack size
+    ld      b,d
+    DUP     8
+        add     hl,hl
+        jr      nc,0_f
+        add     hl,de
+0:
+    EDUP
+
+  ENDIF
+
     add     hl,bc                   ; HL = state_scale * (upkr_state >> 8) + (upkr_state & 255)
-    pop     af
+    pop     af                      ; restore prob and CF=bit
-    ld      d,-16                   ; D = -prob_offset (-16 0xF0 when bit = 0)
     jr      nc,.bit_is_0_2
-    ld      d,b                     ; D = -prob_offset (0 when bit = 1) (also does fix following ADD)
+    dec     d                       ; DE = -prob (also D = bit ? $FF : $00)
-    dec     h
+    add     hl,de                   ; HL += -prob
-    add     hl,de                   ; HL += -prob (HL += (256 - prob) - 256)
+    ; ^ this always preserves CF=1, because (state>>8) >= 128, state_scale: 7..250, prob: 7..250,
-.bit_is_0_2:                        ; HL = state_offset + state_scale * (upkr_state >> 8) + (upkr_state & 255) ; new upkr_state
+    ; so 7*128 > 250 and thus edge case `ADD hl=(7*128+0),de=(-250)` => CF=1
+.bit_is_0_2:
  ; *** adjust probs[context_index]
-    ld      e,a                     ; D:E = -prob_offset:prob, A = prob
+    rra                             ; + (bit<<4) ; part of -prob_offset, needs another -16
-    and     $F8
+    and     $FC                     ; clear/keep correct bits to get desired (prob>>4) + extras, CF=0
     rra
     rra
-    rra
+    rra                             ; A = (bit<<4) + (prob>>4), CF=(prob & 8)
-    rra
+    adc     a,-16                   ; A = (bit<<4) - 16 + ((prob + 8)>>4) ; -prob_offset = (bit<<4) - 16
-    adc     a,d                     ; A = -prob_offset + ((prob + 8) >> 4)
+    ld      e,a
-    neg
-    add     a,e                     ; A = prob_offset + prob - ((prob + 8) >> 4)
     pop     bc
-    ld      (bc),a                  ; update probs[context_index]
+    ld      a,(bc)                  ; A = prob (cheaper + shorter to re-read again from memory)
-    add     a,d                     ; bit=0: A = 23..249, D = 240 -> CF=1 || bit=1: D=0 -> CF=0
+    sub     e                       ; A = 16 - (bit<<4) + prob - ((prob + 8)>>4) ; = prob_offset + prob - ((prob + 8)>>4)
-    ccf                             ; resulting CF = bit restored
+    ld      (bc),a                  ; probs[context_index] = prob_offset + prob - ((prob + 8) >> 4);
+    add     a,d                     ; restore CF = bit (D = bit ? $FF : $00 && A > 0)
     pop     de
     ret
 
@@ -261,19 +298,16 @@ int upkr_decode_length(int context_index) {
 decode_number:
   ; HL = upkr_state
   ; IX = upkr_data_ptr
-  ; BC = probs+context_index
+  ; BC = probs+context_index-1
   ; A' = upkr_current_byte (!!! init to 0x80 at start, not 0x00)
   ; return length in DE, CF=0
-    ld      de,$7FFF            ; length = 0 with positional-stop-bit
+    ld      de,$FFFF            ; length = 0 with positional-stop-bit
-    jr      .loop_entry
+    or      a                   ; CF=0 to skip getting data bit and use only `rr d : rr e` to fix init DE
 .loop:
-    inc     c                   ; context_index + 1
+    call    c,inc_c_decode_bit  ; get data bit, context_index + 1 / if CF=0 just add stop bit into DE init
-    call    decode_bit
     rr      d
     rr      e                   ; DE = length = (length >> 1) | (bit << 15);
-    inc     c                   ; context_index += 2
+    call    inc_c_decode_bit    ; context_index += 2
-.loop_entry:
-    call    decode_bit
     jr      c,.loop
 .fix_bit_pos:
     ccf                         ; NC will become this final `| (1 << bit_pos)` bit
@@ -287,15 +321,61 @@ decode_number:
     ; reserve space for probs array without emitting any machine code (using only EQU)
 
     IFDEF UPKR_PROBS_ORIGIN     ; if specific address is defined by user, move probs array there
-    ORG UPKR_PROBS_ORIGIN
+probs:      EQU ((UPKR_PROBS_ORIGIN) + 255) & -$100     ; probs array aligned to 256
+    ELSE
+probs:      EQU ($ + 255) & -$100                       ; probs array aligned to 256
     ENDIF
-
+.real_c:    EQU 1 + 255 + 1 + 2*NUMBER_BITS             ; real size of probs array
-probs:      EQU ($+255) & -$100                 ; probs array aligned to 256
+.c:         EQU (.real_c + 1) & -2                      ; padding to even size (required by init code)
-.real_c:    EQU 1 + 255 + 1 + 2*NUMBER_BITS     ; real size of probs array
-.c:         EQU (.real_c + 1) & -2              ; padding to even size (required by init code)
 .e:         EQU probs + .c
 
     DISPLAY "upkr.unpack probs array placed at: ",/A,probs,",\tsize: ",/A,probs.c
 
+/*
+ archived: negligibly faster but +6B longer decode_number variant using HL' and BC' to
+ do `number|=(1<<bit_pos);` type of logic in single loop.
+*/
+; decode_number:
+;     exx
+;     ld      bc,1
+;     ld      l,b
+;     ld      h,b                 ; HL = 0
+; .loop
+;     exx
+;     inc     c
+;     call    decode_bit
+;     jr      nc,.done
+;     inc     c
+;     call    decode_bit
+;     exx
+;     jr      nc,.b0
+;     add     hl,bc
+; .b0:
+;     sla     c
+;     rl      b
+;     jr      .loop
+; .done:
+;     exx
+;     add     hl,bc
+;     push    hl
+;     exx
+;     pop     de
+;     ret
+
+/*
+ archived: possible LUT variant of updating probs value, requires 512-aligned 512B table (not tested)
+*/
+; code is replacing decode_bit from "; *** adjust probs[context_index]", followed by `ld (bc),a : add a,d ...`
+;     ld      c,a
+;     ld      a,high(probs_update_table)/2    ; must be 512 aligned
+;     rla
+;     ld      b,a
+;     ld      a,(bc)
+;     pop     bc
+; -------------------------------------------
+; probs_update_table: EQU probs-512
+; -------------------------------------------
+; table generator is not obvious and probably not short either, 20+ bytes almost for sure, maybe even 30-40
+
     ENDMODULE
     OPT pop