add --version flag to output upkr version

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

@@ -172,7 +172,7 @@ checksum = "dcc811dc4066ac62f84f11307873c4850cb653bfa9b1719cee2bd2204a4bc5dd"
 
 [[package]]
 name = "upkr"
-version = "0.2.0-pre3"
+version = "0.2.0"
 dependencies = [
  "anyhow",
  "cdivsufsort",

Cargo.toml

@@ -1,9 +1,10 @@
 [package]
 name = "upkr"
-version = "0.2.0-pre3"
+version = "0.2.0"
 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"

README.md

@@ -2,13 +2,57 @@
 
 Upkr is a simple general purpose lz packer designed to be used in the [MicroW8](https://github.com/exoticorn/microw8) platform.
 The compressed format is losely based on [Shrinkler](https://github.com/askeksa/Shrinkler) with the main difference being that
-Upkr doesn't differnetiate between literals at odd or even addresses and that I went with rANS/rABS instead of a range coder.
+Upkr doesn't differentiate between literals at odd or even addresses (by default) and that I went with rANS/rABS instead of a range coder.
 
-At this point, Upkr should still be considered unstable - the compressed format is not very likely to change but I still want
+Compression rate is on par with Shrinkler.
-to keep that option open a little longer.
+
+The differences compare to Shrinkler also makes it interesting on 8bit platforms. The z80 unpacker included in the release
+is both about twice as fast and smaller than the Shrinkler unpacker.
 
 ## Inspirations:
 
 * Ferris' blog about his [C64 intro packer](https://yupferris.github.io/blog/2020/08/31/c64-4k-intro-packer-deep-dive.html)
 * [Shrinkler](https://github.com/askeksa/Shrinkler)
 * Ryg's [sample rANS implementation](https://github.com/rygorous/ryg_rans)
+
+## Unpackers
+
+The release includes a reference c unpacker, as well as some optimized asm unpackers (arm and riscv). The unpckers in
+c_unpacker and asm_unpackers unpack the default upkr compressed format. The z80_unpacker
+is based on some variations to the compressed format. (Use `upkr --z80` to select those variations.)
+An optimized x86 (DOS) unpacker is currently being worked on out of tree.
+
+## Usage
+
+```
+  upkr [-l level(0-9)] [config options] <infile> [<outfile>]
+  upkr -u [config options] <infile> [<outfile>]
+  upkr --margin [config options] <infile>
+
+ -l, --level N       compression level 0-9
+ -0, ..., -9         short form for setting compression level
+ -u, --unpack        unpack infile
+ --margin            calculate margin for overlapped unpacking of a packed file
+
+Config presets for specific unpackers:
+ --z80               --big-endian-bitstream --invert-bit-encoding --simplified-prob-update -9
+ --x86               --bitstream --invert-is-match-bit --invert-continue-value-bit --invert-new-offset-bit
+ --x86b              --bitstream --invert-continue-value-bit --no-repeated-offsets -9
+
+Config options (need to match when packing/unpacking):
+ -b, --bitstream     bitstream mode
+ -p, --parity N      use N (2/4) parity contexts
+ -r, --reverse       reverse input & output
+
+Config options to tailor output to specific optimized unpackers:
+ --invert-is-match-bit
+ --invert-new-offset-bit
+ --invert-continue-value-bit
+ --invert-bit-encoding
+ --simplified-prob-update
+ --big-endian-bitstream   (implies --bitstream)
+ --no-repeated-offsets
+ --eof-in-length
+ --max-offset N
+ --max-length N
+```

asm_unpackers/Makefile

@@ -39,6 +39,19 @@ build/unpack_armv6m.bin: unpack_armv6m.S
 	arm-none-eabi-gcc -march=armv6-m -c -o build/unpack_armv6m.o $?
 	arm-none-eabi-objcopy -O binary --only-section=.text build/unpack_armv6m.o $@
 	
+build/unpack_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 $@ $^
@@ -47,5 +60,5 @@ 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
+sizes: build/unpack_armv6m.bin build/unpack_riscv64.bin build/unpack_riscv32.bin build/unpack_arm32.bin
 	ls -l build/*.bin

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_riscv.S

@@ -1,13 +1,11 @@
 .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
+// x14 context index
 
 .global upkr_unpack
 .type upkr_unpack, %function
@@ -15,11 +13,11 @@ upkr_unpack:
 	mv t4, ra
 	mv x17, x8
 	mv t6, x9
-	li x13, FRAME_SIZE
+	li x9, 256 + 128
-	li x9, 128
+	mv x13, x9
 1:
-	addi sp, sp, -1
+	sub x8, sp, x13
-	sb x9, 0(sp)
+	sb x9, 0(x8)
 	addi x13, x13, -1
 	bnez x13, 1b
 	
@@ -35,7 +33,7 @@ upkr_unpack:
 
 .Lfinished_offset:
 	addi x14, x14, 64
-	jal t3, upkr_decode_number
+	jalr ra // jal upkr_decode_number
 1:
 	add x14, x10, t0
 	lbu x14, (x14)
@@ -58,36 +56,14 @@ upkr_unpack:
 .Lread_offset_inc_x14:
 	addi x14, x14, 1
 .Lread_offset:
-	jal t3, upkr_decode_number
+	jalr ra // jal 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
@@ -104,39 +80,52 @@ upkr_decode_bit:
 	srli x15, x13, 12
 	beqz x15, upkr_load_byte
 
-	mv t1, x14
+	addi x14, x14, 1
-	mv t2, x10
 	
-	add x14, x14, sp
+	sub t2, sp, x14
-	lbu x12, 0(x14)
+	lbu x12, (t2)
 	
-	andi x10, x13, 255
+	andi x8, x13, 255
-	sltu x15, x10, x12
+	sltu x15, x8, x12
-	srli x13, x13, 8
+	beqz x15, 1f
-	beqz x15, .Lelse
+	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
 	
-	mul x13, x13, x12
+	addi x8, x12, 8
-	add x13, x13, x10
+	srli x8, x8, 4
-	li x10, 256 + 8
+	sub x12, x12, x8
-	sub x10, x10, x12
+	beqz x15, 1f
-	srli x10, x10, 4
+	sub x12, x0, x12
-	add x12, x12, x10
+1:
-	j .Lendif
 
-.Lelse:
+	sb x12, (t2)
-	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:
+	jalr ra
 
-	sb x12, 0(x14)
+// 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
 	
-	addi x14, t1, 1
+	mv x14, t5
-	mv x10, t2
+	jr t3
-	ret

release/.gitignore

@@ -0,0 +1,4 @@
+*.zip
+*.tgz
+upkr-linux/
+upkr-windows/

release/Makefile

@@ -0,0 +1,35 @@
+VERSION := $(shell cargo run --release -- --version)
+
+all: clean upkr-linux-$(VERSION).tgz upkr-windows-$(VERSION).zip
+
+clean:
+	rm -rf upkr-linux
+	rm -f upkr-linux*.tgz
+	rm -rf upkr-windows
+	rm -f upkr-windows*.zip
+
+upkr-linux-$(VERSION).tgz: upkr-linux/upkr PHONY
+	cp ../README.md upkr-linux
+	cd .. && git archive HEAD c_unpacker | tar -xC release/upkr-linux
+	cd .. && git archive HEAD z80_unpacker | tar -xC release/upkr-linux
+	cd .. && git archive HEAD asm_unpackers | tar -xC release/upkr-linux
+	tar czf $@ upkr-linux
+
+upkr-windows-$(VERSION).zip: upkr-windows/upkr.exe PHONY
+	cp ../README.md upkr-windows/
+	cd .. && git archive HEAD c_unpacker | tar -xC release/upkr-windows
+	cd .. && git archive HEAD z80_unpacker | tar -xC release/upkr-windows
+	cd .. && git archive HEAD asm_unpackers | tar -xC release/upkr-windows
+	zip -r -9 $@ upkr-windows
+
+upkr-linux/upkr:
+	cargo build --target x86_64-unknown-linux-musl --release
+	mkdir -p upkr-linux
+	cp ../target/x86_64-unknown-linux-musl/release/upkr upkr-linux/
+
+upkr-windows/upkr.exe:
+	cargo build --target x86_64-pc-windows-gnu --release
+	mkdir -p upkr-windows
+	cp ../target/x86_64-pc-windows-gnu/release/upkr.exe upkr-windows/
+	
+PHONY:

src/main.rs

@@ -49,12 +49,22 @@ fn main() -> Result<()> {
                 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;
+            }
 
             Short('u') | Long("unpack") => unpack = true,
             Long("margin") => calculate_margin = true,
             Short('l') | Long("level") => level = parser.value()?.parse()?,
             Short(n) if n.is_ascii_digit() => level = n as u8 - b'0',
             Short('h') | Long("help") => print_help(0),
+            Long("version") => {
+                println!("{}", env!("CARGO_PKG_VERSION"));
+                process::exit(0);
+            }
             Long("max-unpacked-size") => max_unpacked_size = parser.value()?.parse()?,
             Value(val) if infile.is_none() => infile = Some(val.try_into()?),
             Value(val) if outfile.is_none() => outfile = Some(val.try_into()?),
@@ -149,11 +159,16 @@ fn print_help(exit_code: i32) -> ! {
     eprintln!(" -u, --unpack        unpack infile");
     eprintln!(" --margin            calculate margin for overlapped unpacking of a packed file");
     eprintln!();
+    eprintln!("Version: {}", env!("CARGO_PKG_VERSION"));
+    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");

z80_unpacker/.gitignore

@@ -0,0 +1,3 @@
+*.bin
+*.tap
+*.lst

z80_unpacker/Makefile

@@ -0,0 +1,11 @@
+all: unpack.bin example/example.sna
+
+# binary is positioned from ORG 0, not usable, just assembling to verify the syntax
+unpack.bin: unpack.asm
+	sjasmplus --msg=war --lst --lstlab=sort --raw=unpack.bin unpack.asm
+
+example/example.sna: unpack.asm example/example.asm
+	cd example && sjasmplus --msg=war --lst --lstlab=sort example.asm
+
+clean:
+	$(RM) unpack.bin unpack.lst example/example.sna example/example.lst

z80_unpacker/example/example.asm

@@ -0,0 +1,100 @@
+;; Example using upkr depacker for screens slideshow
+    OPT --syntax=abf
+    DEVICE ZXSPECTRUM48,$8FFF
+
+    ORG     $9000
+  ;; 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
+    INCBIN  "screens/Schafft - Poison (2017).scr.upk"
+    DB      0
+    INCBIN  "screens/diver - Mercenary 4. The Heaven's Devil (2014) (Forever 2014 Olympic Edition, 1).scr.upk"
+    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
+
+  ;;; FORWARD packed/unpacked data demo
+    ld      ix,compressed_scr_files.fwd
+.slideshow_loop.fwd:
+  ; set BORDER for next image
+    ld      a,(ix)
+    inc     ix
+    out     (254),a
+  ; call unpack of next image directly into VRAM
+    ld      de,$4000                    ; target VRAM
+    exx
+  ; IX = packed data, DE' = destination ($4000)
+  ; returned IX will point right after the packed data
+    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
+    ret
+
+  ; 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
+
+    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

@@ -0,0 +1,32 @@
+Z80 asm implementation of C unpacker, code-size focused (not performance).
+
+**ONLY BITSTREAM** variant is currently supported, make sure to use "-b" in packer.
+
+The project is expected to further evolve, including possible changes to binary format, this is
+initial version of Z80 unpacker to explore if/how it works and how it can be improved further.
+
+(copy full packer+depacker source to your project if you plan to use it, as future revisions
+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 bigger corpus of test data to benchmark future changes in algorithm/format (example and zx48.rom was used to do initial tests)
+- maybe try to beat double-loop `decode_number` with different encoding format
+- (@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

@@ -0,0 +1,381 @@
+;; https://github.com/exoticorn/upkr/blob/z80/c_unpacker/unpack.c - original C implementation
+;; C source in comments ahead of asm - the C macros are removed to keep only bitstream variant
+;;
+;; 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 (320 bytes),
+;; otherwise it will be positioned after the unpacker code (256 aligned)
+;;
+;; public API:
+;;
+;;     upkr.unpack
+;;         IN: IX = packed data, DE' (shadow DE) = destination
+;;         OUT: IX = after packed data
+;;         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
+
+NUMBER_BITS     EQU     16+15       ; context-bits per offset/length (16+15 for 16bit offsets/pointers)
+    ; numbers (offsets/lengths) are encoded like: 1a1b1c1d1e0 = 0000'0000'001e'dbca
+
+/*
+u8* upkr_data_ptr;
+u8 upkr_probs[1 + 255 + 1 + 2*32 + 2*32];
+u16 upkr_state;
+u8 upkr_current_byte;
+int upkr_bits_left;
+
+int upkr_unpack(void* destination, void* compressed_data) {
+    upkr_data_ptr = (u8*)compressed_data;
+    upkr_state = 0;
+    upkr_bits_left = 0;
+    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 - (u8*)destination;
+}
+*/
+; IN: IX = compressed_data, DE' = destination
+unpack:
+  ; ** reset probs to 0x80, also reset HL (state) to zero, and set BC to probs+context 0
+    ld      hl,probs.c>>1
+    ld      bc,probs.e
+    ld      a,$80
+.reset_probs:
+    dec     bc
+    ld      (bc),a              ; will overwrite one extra byte after the array because of odd length
+    dec     bc
+    ld      (bc),a
+    dec     l
+    jr      nz,.reset_probs
+    exa
+    ; BC = probs (context_index 0), state HL = 0, A' = 0x80 (no source bits left in upkr_current_byte)
+
+  ; ** main loop to decompress data
+    ; D = prev_was_match = uninitialised, literal is expected first => will reset D to "false"
+    ; values for false/true of prev_was_match are: false = high(probs), true = 1 + high(probs)
+.decompress_data:
+    ld      c,0
+    call    decode_bit          ; if(upkr_decode_bit(0))
+    jr      c,.copy_chunk
+
+  ; * extract byte from compressed data (literal)
+    inc     c                   ; C = byte = 1 (and also context_index)
+.decode_byte:
+    call    decode_bit          ; bit = upkr_decode_bit(byte);
+    rl      c                   ; byte = (byte << 1) + bit;
+    jr      nc,.decode_byte     ; while(byte < 256)
+    ld      a,c
+    exx
+    ld      (de),a              ; *write_ptr++ = byte;
+    IFNDEF BACKWARDS_UNPACK : inc de : ELSE : dec de : ENDIF
+    exx
+    ld      d,b                 ; prev_was_match = false
+    jr      .decompress_data
+
+  ; * copy chunk of already decompressed data (match)
+.copy_chunk:
+    ld      a,b
+    inc     b                   ; context_index = 256
+        ;             if(prev_was_match || upkr_decode_bit(256)) {
+        ;                 offset = upkr_decode_length(257) - 1;
+        ;                 if (0 == offset) break;
+        ;             }
+    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
+    call    decode_number       ; context_index is already 257-1 as needed by decode_number
+    dec     de                  ; offset = upkr_decode_length(257) - 1;
+    ld      a,d
+    or      e
+    ret     z                   ; if(offset == 0) break
+    ld      (.offset),de
+.keep_offset:
+        ;             int length = upkr_decode_length(257 + 64);
+        ;             while(length--) {
+        ;                 *write_ptr = write_ptr[-offset];
+        ;                 ++write_ptr;
+        ;             }
+        ;             prev_was_match = 1;
+    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
+    IFNDEF BACKWARDS_UNPACK
+        ; forward unpack (write_ptr++, upkr_data_ptr++)
+        ld      h,d             ; DE = write_ptr
+        ld      l,e
+.offset+*:  ld  bc,0
+        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
+
+/*
+int upkr_decode_bit(int context_index) {
+    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 >> 7);
+        upkr_current_byte <<= 1;
+        --upkr_bits_left;
+    }
+
+    int prob = upkr_probs[context_index];
+    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;
+        state_scale = 256 - prob;
+        prob_offset = 0;
+    }
+    upkr_state = state_offset + state_scale * (upkr_state >> 8) + (upkr_state & 255);
+    upkr_probs[context_index] = prob_offset + prob - ((prob + 8) >> 4);
+
+    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
+  ; BC = probs+context_index
+  ; A' = upkr_current_byte (!!! init to 0x80 at start, not 0x00)
+  ; preserves DE
+  ; ** while (state < 32768) - initial check
+    push    de
+    bit     7,h
+    jr      nz,.state_b15_set
+    exa
+  ; ** while body
+.state_b15_zero:
+  ; HL = upkr_state
+  ; IX = upkr_data_ptr
+  ; A = upkr_current_byte (init to 0x80 at start, not 0x00)
+    add     a,a                     ; upkr_current_byte <<= 1; // and testing if(upkr_bits_left == 0)
+    jr      nz,.has_bit             ; CF=data, ZF=0 -> some bits + stop bit still available
+  ; CF=1 (by stop bit)
+    ld      a,(ix)
+    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);
+    jp      p,.state_b15_zero       ; while (state < 32768)
+    exa
+  ; ** set "bit"
+.state_b15_set:
+    ld      a,(bc)                  ; A = upkr_probs[context_index]
+    dec     a                       ; prob is in ~7..249 range, never zero, safe to -1
+    cp      l                       ; CF = bit = prob-1 < (upkr_state & 255) <=> prob <= (upkr_state & 255)
+    inc     a
+  ; ** adjust state
+    push    bc
+    ld      c,l                     ; C = (upkr_state & 255); (preserving the value)
+    push    af
+    jr      nc,.bit_is_0
+    neg                             ; A = -prob == (256-prob), CF=1 preserved
+.bit_is_0:
+    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
+    jr      nc,.mul0
+    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                      ; restore prob and CF=bit
+    jr      nc,.bit_is_0_2
+    dec     d                       ; DE = -prob (also D = bit ? $FF : $00)
+    add     hl,de                   ; HL += -prob
+    ; ^ this always preserves CF=1, because (state>>8) >= 128, state_scale: 7..250, prob: 7..250,
+    ; 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]
+    rra                             ; + (bit<<4) ; part of -prob_offset, needs another -16
+    and     $FC                     ; clear/keep correct bits to get desired (prob>>4) + extras, CF=0
+    rra
+    rra
+    rra                             ; A = (bit<<4) + (prob>>4), CF=(prob & 8)
+    adc     a,-16                   ; A = (bit<<4) - 16 + ((prob + 8)>>4) ; -prob_offset = (bit<<4) - 16
+    ld      e,a
+    pop     bc
+    ld      a,(bc)                  ; A = prob (cheaper + shorter to re-read again from memory)
+    sub     e                       ; A = 16 - (bit<<4) + prob - ((prob + 8)>>4) ; = prob_offset + prob - ((prob + 8)>>4)
+    ld      (bc),a                  ; probs[context_index] = prob_offset + prob - ((prob + 8) >> 4);
+    add     a,d                     ; restore CF = bit (D = bit ? $FF : $00 && A > 0)
+    pop     de
+    ret
+
+/*
+int upkr_decode_length(int context_index) {
+    int length = 0;
+    int bit_pos = 0;
+    while(upkr_decode_bit(context_index)) {
+        length |= upkr_decode_bit(context_index + 1) << bit_pos++;
+        context_index += 2;
+    }
+    return length | (1 << bit_pos);
+}
+*/
+decode_number:
+  ; HL = upkr_state
+  ; IX = upkr_data_ptr
+  ; BC = probs+context_index-1
+  ; A' = upkr_current_byte (!!! init to 0x80 at start, not 0x00)
+  ; return length in DE, CF=0
+    ld      de,$FFFF            ; length = 0 with positional-stop-bit
+    or      a                   ; CF=0 to skip getting data bit and use only `rr d : rr e` to fix init DE
+.loop:
+    call    c,inc_c_decode_bit  ; get data bit, context_index + 1 / if CF=0 just add stop bit into DE init
+    rr      d
+    rr      e                   ; DE = length = (length >> 1) | (bit << 15);
+    call    inc_c_decode_bit    ; context_index += 2
+    jr      c,.loop
+.fix_bit_pos:
+    ccf                         ; NC will become this final `| (1 << bit_pos)` bit
+    rr      d
+    rr      e
+    jr      c,.fix_bit_pos      ; until stop bit is reached (all bits did land to correct position)
+    ret                         ; return with CF=0 (important for unpack routine)
+
+    DISPLAY "upkr.unpack total size: ",/D,$-unpack
+
+    ; 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
+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
+.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