39 Commits

Author SHA1 Message Date
31fb91c629 Merge branch 'ped7g-z80_ped7g' 2022-09-27 22:34:19 +02:00
e429f252a5 Merge branch 'z80' 2022-09-27 22:28:12 +02:00
f6642f07c9 more config options, unpack error handling, fuzzing 2022-09-27 17:16:05 +02:00
Peter Helcmanovsky (Ped)
8a32e1384c z80_unpacker: readme.txt and comment update 2022-09-19 15:19:39 +02:00
Peter Helcmanovsky (Ped)
9913dcf4bb z80_unpacker: comment with possible LUT variant of updating probs value
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.
2022-09-19 14:31:00 +02:00
Peter Helcmanovsky (Ped)
a8fd3dc573 z80_unpacker: optimisation: -1B in decode_number (fwd 170B / rev 167B)
slightly slower code, ROM unpack is back to ~22.6s
2022-09-19 13:20:44 +02:00
Peter Helcmanovsky (Ped)
e1f9fa143a z80_unpacker: comment with caller size optimisation tip 2022-09-19 11:58:32 +02:00
Peter Helcmanovsky (Ped)
db1c7d2d14 z80_unpacker: optimisation: -1B in decode_number (fwd 171B / rev 168B) 2022-09-19 11:49:53 +02:00
Peter Helcmanovsky (Ped)
c1ffd0e7ed z80_unpacker: attempt for faster decode_number (+6B, ~1% faster) => not good
archived in comments for future reference
2022-09-19 11:42:56 +02:00
Peter Helcmanovsky (Ped)
00d084105a z80_unpacker: optimisation: -2B in backward unpack (fwd 172B / rev 169B)
backward was already -1B, so now the total difference is -3B.
2022-09-19 01:31:22 +02:00
Peter Helcmanovsky (Ped)
8e5298caee z80_unpacker: optimisation: -1B in decode_number = 172B (but +4T per length) 2022-09-19 01:09:21 +02:00
Peter Helcmanovsky (Ped)
1fb29f3a1b z80_unpacker: optimisation: -1B and -1T in decode_bit = 173B 2022-09-18 23:44:18 +02:00
c8924456aa -r reverses both input and output 2022-09-18 23:38:41 +02:00
7b0e22f459 Merge pull request #3 from ped7g/z80_ped7g
backward unpacker + example extended
2022-09-18 23:24:28 +02:00
Peter Helcmanovsky (Ped)
165f593a11 z80_unpacker: (codestyle) whitespace + temporary label rename 2022-09-18 23:04:37 +02:00
Peter Helcmanovsky (Ped)
d4bce4bf7c z80_unpacker: optimisation: -3B and ~-10T in decode_bit = 174B
unpack zx48.rom is now ~22.6s (from 23.0s)
(performance version is now 199 bytes, zx48.rom unpack 19.4s -> 19.0s)
2022-09-18 22:54:10 +02:00
Peter Helcmanovsky (Ped)
b13fa05413 z80_unpacker: add backward variant of unpacker + example extended 2022-09-18 00:23:14 +02:00
Peter Helcmanovsky (Ped)
3c773aca8d z80_unpacker: add performance variant of depacker 2022-09-16 03:38:03 +02:00
a5406deb30 Merge pull request #2 from ped7g/z80_ped7g
Z80 ped7g - few more optimisations for current variant of packer
2022-09-16 00:26:55 +02:00
Peter Helcmanovsky (Ped)
9211544cb9 z80_unpacker: add resulting snapshot file to example 2022-09-15 18:37:06 +02:00
Peter Helcmanovsky (Ped)
3fa9e0fa12 z80_unpacker: optimisations: 0B, -13T in decode_bit (stays 177B) 2022-09-15 18:22:33 +02:00
Peter Helcmanovsky (Ped)
aa3fad4d80 z80_unpacker: optimisations: -3B and ~-24T in decode_bit = 177B 2022-09-15 18:22:32 +02:00
Peter Helcmanovsky (Ped)
6624940ed9 z80_unpacker: optimisations: -2B and -27T in decode_bit = 180B 2022-09-15 18:22:32 +02:00
Peter Helcmanovsky (Ped)
c3a9773e5c z80_unpacker: optimisations: -1B in unpack implementation = 182B 2022-09-15 18:22:31 +02:00
Peter Helcmanovsky (Ped)
a75a35efb2 z80_unpacker: probs context-size for offset/length numbers as EQU 2022-09-15 18:22:27 +02:00
540a91d1ba forgot to add back -l 9 2022-09-15 00:18:30 +02:00
e7aaf1491a add old-prob-update to compare script, add reverse option 2022-09-14 23:51:38 +02:00
a1dabaf7f9 add simple script to compare compression of variants 2022-09-14 23:41:14 +02:00
75e375fb1f Merge branch 'ped7g-z80_ped7g' into z80 2022-09-14 09:03:28 +02:00
Peter Helcmanovsky (Ped)
c7ea11bce3 z80_unpacker: optimisations: -2B in unpack implementation = 183B 2022-09-14 01:44:04 +02:00
Peter Helcmanovsky (Ped)
02d20867ee z80_unpacker: optimisations: -2B in unpack implementation = 185B 2022-09-14 01:01:56 +02:00
Peter Helcmanovsky (Ped)
511ddefc08 z80_unpacker: optimisations: -4T per offset/length bit decoded
making the 256-alignment of probs array even more baked-in, but there
was no real chance to get rid of that any way
2022-09-14 00:01:51 +02:00
Peter Helcmanovsky (Ped)
d30baaa91f z80_unpacker: optimisations: -1B by keeping write_ptr in DE' 2022-09-13 23:57:59 +02:00
Peter Helcmanovsky (Ped)
919a892ef0 z80_unpacker: optimisations: -1B by decode_length returning CF=0 2022-09-13 23:25:03 +02:00
Peter Helcmanovsky (Ped)
ea5c0b1b15 z80_unpacker: optimisations: shorter >>4 in probs update 2022-09-13 23:15:18 +02:00
Peter Helcmanovsky (Ped)
a19ec2abb7 z80_unpacker: optimisations: remove .offset init
first offset is mandatory in packed data
2022-09-13 22:53:15 +02:00
Peter Helcmanovsky (Ped)
7b051113e1 z80_unpacker: initial working version with screen-slideshow example 2022-09-13 22:12:03 +02:00
f1f1c64a76 implement simplified prob update, update unpack.c 2022-09-10 12:01:42 +02:00
36cb6d77b5 BE bitstream, flip bit encoding 2022-09-10 11:31:09 +02:00
31 changed files with 1042 additions and 62 deletions

58
Cargo.lock generated
View File

@@ -95,6 +95,24 @@ dependencies = [
"winapi", "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]] [[package]]
name = "sacabase" name = "sacabase"
version = "2.0.0" version = "2.0.0"
@@ -104,6 +122,37 @@ dependencies = [
"num-traits", "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]] [[package]]
name = "time" name = "time"
version = "0.1.44" version = "0.1.44"
@@ -115,14 +164,21 @@ dependencies = [
"winapi", "winapi",
] ]
[[package]]
name = "unicode-ident"
version = "1.0.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "dcc811dc4066ac62f84f11307873c4850cb653bfa9b1719cee2bd2204a4bc5dd"
[[package]] [[package]]
name = "upkr" name = "upkr"
version = "0.2.0-pre2" version = "0.2.0-pre3"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"cdivsufsort", "cdivsufsort",
"lexopt", "lexopt",
"pbr", "pbr",
"thiserror",
] ]
[[package]] [[package]]

View File

@@ -1,6 +1,6 @@
[package] [package]
name = "upkr" name = "upkr"
version = "0.2.0-pre2" version = "0.2.0-pre3"
edition = "2021" edition = "2021"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
@@ -9,4 +9,5 @@ edition = "2021"
cdivsufsort = "2" cdivsufsort = "2"
lexopt = "0.2.1" lexopt = "0.2.1"
anyhow = "1" anyhow = "1"
thiserror = "1.0.36"
pbr = "1" pbr = "1"

3
fuzz/.gitignore vendored Normal file
View File

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

247
fuzz/Cargo.lock generated Normal file
View File

@@ -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"

31
fuzz/Cargo.toml Normal file
View File

@@ -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

View File

@@ -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);
});

View File

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

View File

@@ -19,15 +19,16 @@ pub fn pack(
} }
let mut encoded_match = false; let mut encoded_match = false;
if let Some(m) = match_finder.matches(pos).next() { 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; let offset = pos - m.pos;
if offset < max_offset && m.length >= config.min_length() { if offset < max_offset && m.length >= config.min_length() {
let length = m.length.min(config.max_length);
lz::Op::Match { lz::Op::Match {
offset: offset as u32, offset: offset as u32,
len: m.length as u32, len: length as u32,
} }
.encode(&mut rans_coder, &mut state, config); .encode(&mut rans_coder, &mut state, config);
pos += m.length; pos += length;
encoded_match = true; encoded_match = true;
} }
} }
@@ -39,7 +40,8 @@ pub fn pack(
.iter() .iter()
.zip(data[(pos - offset)..].iter()) .zip(data[(pos - offset)..].iter())
.take_while(|(a, b)| a == b) .take_while(|(a, b)| a == b)
.count(); .count()
.min(config.max_length);
if length >= config.min_length() { if length >= config.min_length() {
lz::Op::Match { lz::Op::Match {
offset: offset as u32, offset: offset as u32,

View File

@@ -5,10 +5,11 @@ mod match_finder;
mod parsing_packer; mod parsing_packer;
mod rans; mod rans;
pub use lz::{calculate_margin, unpack}; pub use lz::{calculate_margin, unpack, UnpackError};
pub type ProgressCallback<'a> = &'a mut dyn FnMut(usize); pub type ProgressCallback<'a> = &'a mut dyn FnMut(usize);
#[derive(Debug)]
pub struct Config { pub struct Config {
pub use_bitstream: bool, pub use_bitstream: bool,
pub parity_contexts: usize, pub parity_contexts: usize,
@@ -23,6 +24,9 @@ pub struct Config {
pub no_repeated_offsets: bool, pub no_repeated_offsets: bool,
pub eof_in_length: bool, pub eof_in_length: bool,
pub max_offset: usize,
pub max_length: usize,
} }
impl Default for Config { impl Default for Config {
@@ -41,6 +45,9 @@ impl Default for Config {
no_repeated_offsets: false, no_repeated_offsets: false,
eof_in_length: false, eof_in_length: false,
max_offset: usize::MAX,
max_length: usize::MAX,
} }
} }
} }
@@ -58,13 +65,13 @@ impl Config {
pub fn pack( pub fn pack(
data: &[u8], data: &[u8],
level: u8, level: u8,
config: Config, config: &Config,
progress_callback: Option<ProgressCallback>, progress_callback: Option<ProgressCallback>,
) -> Vec<u8> { ) -> Vec<u8> {
if level == 0 { if level == 0 {
greedy_packer::pack(data, &config, progress_callback) greedy_packer::pack(data, config, progress_callback)
} else { } else {
parsing_packer::pack(data, level, &config, progress_callback) parsing_packer::pack(data, level, config, progress_callback)
} }
} }

View File

@@ -1,6 +1,7 @@
use crate::context_state::ContextState; use crate::context_state::ContextState;
use crate::rans::{EntropyCoder, RansDecoder}; use crate::rans::{EntropyCoder, RansDecoder};
use crate::Config; use crate::Config;
use thiserror::Error;
#[derive(Copy, Clone, Debug)] #[derive(Copy, Clone, Debug)]
pub enum Op { pub enum Op {
@@ -25,17 +26,18 @@ impl Op {
} }
&Op::Match { offset, len } => { &Op::Match { offset, len } => {
encode_bit(coder, state, literal_base, config.is_match_bit); encode_bit(coder, state, literal_base, config.is_match_bit);
let mut new_offset = true;
if !state.prev_was_match && !config.no_repeated_offsets { if !state.prev_was_match && !config.no_repeated_offsets {
new_offset = offset != state.last_offset;
encode_bit( encode_bit(
coder, coder,
state, state,
256 * state.parity_contexts, 256 * state.parity_contexts,
(offset != state.last_offset) == config.new_offset_bit, new_offset == config.new_offset_bit,
); );
} else {
assert!(offset != state.last_offset || config.no_repeated_offsets);
} }
if offset != state.last_offset || config.no_repeated_offsets { assert!(offset as usize <= config.max_offset);
if new_offset {
encode_length( encode_length(
coder, coder,
state, state,
@@ -45,7 +47,7 @@ impl Op {
); );
state.last_offset = offset; state.last_offset = offset;
} }
assert!(!config.eof_in_length || len > 1); assert!(len as usize >= config.min_length() && len as usize <= config.max_length);
encode_length(coder, state, 256 * state.parity_contexts + 65, len, config); encode_length(coder, state, 256 * state.parity_contexts + 65, len, config);
state.prev_was_match = true; state.prev_was_match = true;
state.pos += len as usize; state.pos += len as usize;
@@ -69,7 +71,7 @@ pub fn encode_eof(coder: &mut dyn EntropyCoder, state: &mut CoderState, config:
config.new_offset_bit ^ config.eof_in_length, config.new_offset_bit ^ config.eof_in_length,
); );
} }
if !config.eof_in_length || config.no_repeated_offsets { if !config.eof_in_length || state.prev_was_match || config.no_repeated_offsets {
encode_length(coder, state, 256 * state.parity_contexts + 1, 1, config); encode_length(coder, state, 256 * state.parity_contexts + 1, 1, config);
} }
if config.eof_in_length { if config.eof_in_length {
@@ -130,24 +132,44 @@ impl CoderState {
} }
} }
pub fn unpack(packed_data: &[u8], config: &Config) -> Vec<u8> { #[derive(Error, Debug)]
let mut result = vec![]; pub enum UnpackError {
let _ = unpack_internal(Some(&mut result), packed_data, config); #[error("match offset out of range: {offset} > {position}")]
result 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 calculate_margin(packed_data: &[u8], config: &Config) -> isize { pub fn unpack(
unpack_internal(None, packed_data, config) packed_data: &[u8],
config: &Config,
max_size: usize,
) -> Result<Vec<u8>, UnpackError> {
let mut result = vec![];
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( pub fn unpack_internal(
mut result: Option<&mut Vec<u8>>, mut result: Option<&mut Vec<u8>>,
packed_data: &[u8], packed_data: &[u8],
config: &Config, config: &Config,
) -> isize { max_size: usize,
) -> Result<isize, UnpackError> {
let mut decoder = RansDecoder::new(packed_data, &config); let mut decoder = RansDecoder::new(packed_data, &config);
let mut contexts = ContextState::new((1 + 255) * config.parity_contexts + 1 + 64 + 64, &config); let mut contexts = ContextState::new((1 + 255) * config.parity_contexts + 1 + 64 + 64, &config);
let mut offset = 0; let mut offset = usize::MAX;
let mut position = 0usize; let mut position = 0usize;
let mut prev_was_match = false; let mut prev_was_match = false;
let mut margin = 0isize; let mut margin = 0isize;
@@ -157,31 +179,34 @@ pub fn unpack_internal(
contexts: &mut ContextState, contexts: &mut ContextState,
mut context_index: usize, mut context_index: usize,
config: &Config, config: &Config,
) -> usize { ) -> Result<usize, UnpackError> {
let mut length = 0; let mut length = 0;
let mut bit_pos = 0; let mut bit_pos = 0;
while decoder.decode_with_context(&mut contexts.context_mut(context_index)) while decoder.decode_with_context(&mut contexts.context_mut(context_index))?
== config.continue_value_bit == config.continue_value_bit
{ {
length |= (decoder.decode_with_context(&mut contexts.context_mut(context_index + 1)) length |= (decoder.decode_with_context(&mut contexts.context_mut(context_index + 1))?
as usize) as usize)
<< bit_pos; << bit_pos;
bit_pos += 1; bit_pos += 1;
if bit_pos >= 32 {
return Err(UnpackError::ValueOverflow);
}
context_index += 2; context_index += 2;
} }
length | (1 << bit_pos) Ok(length | (1 << bit_pos))
} }
loop { loop {
margin = margin.max(position as isize - decoder.pos() as isize); margin = margin.max(position as isize - decoder.pos() as isize);
let literal_base = position % config.parity_contexts * 256; let literal_base = position % config.parity_contexts * 256;
if decoder.decode_with_context(&mut contexts.context_mut(literal_base)) if decoder.decode_with_context(&mut contexts.context_mut(literal_base))?
== config.is_match_bit == config.is_match_bit
{ {
if config.no_repeated_offsets if config.no_repeated_offsets
|| prev_was_match || prev_was_match
|| decoder || decoder
.decode_with_context(&mut contexts.context_mut(256 * config.parity_contexts)) .decode_with_context(&mut contexts.context_mut(256 * config.parity_contexts))?
== config.new_offset_bit == config.new_offset_bit
{ {
offset = decode_length( offset = decode_length(
@@ -189,7 +214,7 @@ pub fn unpack_internal(
&mut contexts, &mut contexts,
256 * config.parity_contexts + 1, 256 * config.parity_contexts + 1,
&config, &config,
) - if config.eof_in_length { 0 } else { 1 }; )? - if config.eof_in_length { 0 } else { 1 };
if offset == 0 { if offset == 0 {
break; break;
} }
@@ -199,13 +224,20 @@ pub fn unpack_internal(
&mut contexts, &mut contexts,
256 * config.parity_contexts + 65, 256 * config.parity_contexts + 65,
&config, &config,
); )?;
if config.eof_in_length && length == 1 { if config.eof_in_length && length == 1 {
break; break;
} }
if offset > position {
return Err(UnpackError::OffsetOutOfRange { offset, position });
}
if let Some(ref mut result) = result { if let Some(ref mut result) = result {
for _ in 0..length { for _ in 0..length {
if result.len() < max_size {
result.push(result[result.len() - offset]); result.push(result[result.len() - offset]);
} else {
break;
}
} }
} }
position += length; position += length;
@@ -215,17 +247,26 @@ pub fn unpack_internal(
let mut byte = 0; let mut byte = 0;
for i in (0..8).rev() { for i in (0..8).rev() {
let bit = decoder let bit = decoder
.decode_with_context(&mut contexts.context_mut(literal_base + context_index)); .decode_with_context(&mut contexts.context_mut(literal_base + context_index))?;
context_index = (context_index << 1) | bit as usize; context_index = (context_index << 1) | bit as usize;
byte |= (bit as u8) << i; byte |= (bit as u8) << i;
} }
if let Some(ref mut result) = result { if let Some(ref mut result) = result {
if result.len() < max_size {
result.push(byte); result.push(byte);
} }
}
position += 1; position += 1;
prev_was_match = false; prev_was_match = false;
} }
} }
margin + decoder.pos() as isize - position as isize if position > max_size {
return Err(UnpackError::OverSize {
size: position,
limit: max_size,
});
}
Ok(margin + decoder.pos() as isize - position as isize)
} }

View File

@@ -12,6 +12,7 @@ fn main() -> Result<()> {
let mut level = 2; let mut level = 2;
let mut infile: Option<PathBuf> = None; let mut infile: Option<PathBuf> = None;
let mut outfile: Option<PathBuf> = None; let mut outfile: Option<PathBuf> = None;
let mut max_unpacked_size = 512 * 1024 * 1024;
let mut parser = lexopt::Parser::from_env(); let mut parser = lexopt::Parser::from_env();
while let Some(arg) = parser.next()? { while let Some(arg) = parser.next()? {
@@ -32,6 +33,9 @@ fn main() -> Result<()> {
Long("no-repeated-offsets") => config.no_repeated_offsets = true, Long("no-repeated-offsets") => config.no_repeated_offsets = true,
Long("eof-in-length") => config.eof_in_length = true, Long("eof-in-length") => config.eof_in_length = true,
Long("max-offset") => config.max_offset = parser.value()?.parse()?,
Long("max-length") => config.max_length = parser.value()?.parse()?,
Long("z80") => { Long("z80") => {
config.use_bitstream = true; config.use_bitstream = true;
config.bitstream_is_big_endian = true; config.bitstream_is_big_endian = true;
@@ -43,12 +47,15 @@ fn main() -> Result<()> {
config.use_bitstream = true; config.use_bitstream = true;
config.continue_value_bit = false; config.continue_value_bit = false;
config.is_match_bit = false; config.is_match_bit = false;
config.new_offset_bit = false;
} }
Short('u') | Long("unpack") => unpack = true, Short('u') | Long("unpack") => unpack = true,
Long("margin") => calculate_margin = true, Long("margin") => calculate_margin = true,
Short('l') | Long("level") => level = parser.value()?.parse()?, 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), Short('h') | Long("help") => print_help(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 infile.is_none() => infile = Some(val.try_into()?),
Value(val) if outfile.is_none() => outfile = Some(val.try_into()?), Value(val) if outfile.is_none() => outfile = Some(val.try_into()?),
_ => return Err(arg.unexpected().into()), _ => return Err(arg.unexpected().into()),
@@ -92,7 +99,7 @@ fn main() -> Result<()> {
let mut packed_data = upkr::pack( let mut packed_data = upkr::pack(
&data, &data,
level, level,
config, &config,
Some(&mut |pos| { Some(&mut |pos| {
pb.set(pos as u64); pb.set(pos as u64);
}), }),
@@ -117,14 +124,14 @@ fn main() -> Result<()> {
data.reverse(); data.reverse();
} }
if unpack { if unpack {
let mut unpacked_data = upkr::unpack(&data, &config); let mut unpacked_data = upkr::unpack(&data, &config, max_unpacked_size)?;
if reverse { if reverse {
unpacked_data.reverse(); unpacked_data.reverse();
} }
File::create(outfile)?.write_all(&unpacked_data)?; File::create(outfile)?.write_all(&unpacked_data)?;
} }
if calculate_margin { if calculate_margin {
println!("{}", upkr::calculate_margin(&data, &config)); println!("{}", upkr::calculate_margin(&data, &config)?);
} }
} }
@@ -138,12 +145,15 @@ fn print_help(exit_code: i32) -> ! {
eprintln!(" upkr --margin [config options] <infile>"); eprintln!(" upkr --margin [config options] <infile>");
eprintln!(); eprintln!();
eprintln!(" -l, --level N compression level 0-9"); eprintln!(" -l, --level N compression level 0-9");
eprintln!(" -0, ..., -9 short form for setting compression level");
eprintln!(" -u, --unpack unpack infile"); eprintln!(" -u, --unpack unpack infile");
eprintln!(" --margin calculate margin for overlapped unpacking of a packed file"); eprintln!(" --margin calculate margin for overlapped unpacking of a packed file");
eprintln!(); eprintln!();
eprintln!("Config presets for specific unpackers:"); eprintln!("Config presets for specific unpackers:");
eprintln!(" --z80 --big-endian-bitstream --invert-bit-encoding --simplified-prob-update"); eprintln!(" --z80 --big-endian-bitstream --invert-bit-encoding --simplified-prob-update -9");
eprintln!(" --x86 --bitstream --invert-is-match-bit --invert-continue-value-bit"); eprintln!(
" --x86 --bitstream --invert-is-match-bit --invert-continue-value-bit --invert-new-offset-bit"
);
eprintln!(); eprintln!();
eprintln!("Config options (need to match when packing/unpacking):"); eprintln!("Config options (need to match when packing/unpacking):");
eprintln!(" -b, --bitstream bitstream mode"); eprintln!(" -b, --bitstream bitstream mode");
@@ -158,5 +168,8 @@ fn print_help(exit_code: i32) -> ! {
eprintln!(" --simplified-prob-update"); eprintln!(" --simplified-prob-update");
eprintln!(" --big-endian-bitstream (implies --bitstream)"); eprintln!(" --big-endian-bitstream (implies --bitstream)");
eprintln!(" --no-repeated-offsets"); eprintln!(" --no-repeated-offsets");
eprintln!(" --eof-in-length");
eprintln!(" --max-offset N");
eprintln!(" --max-length N");
process::exit(exit_code); process::exit(exit_code);
} }

View File

@@ -105,7 +105,7 @@ fn parse(
cost_counter: &mut CostCounter, cost_counter: &mut CostCounter,
pos: usize, pos: usize,
offset: usize, offset: usize,
length: usize, mut length: usize,
arrival: &Arrival, arrival: &Arrival,
max_arrivals: usize, max_arrivals: usize,
config: &crate::Config, config: &crate::Config,
@@ -113,6 +113,7 @@ fn parse(
if length < config.min_length() { if length < config.min_length() {
return; return;
} }
length = length.min(config.max_length);
cost_counter.reset(); cost_counter.reset();
let mut state = arrival.state.clone(); let mut state = arrival.state.clone();
let op = lz::Op::Match { let op = lz::Op::Match {
@@ -186,6 +187,7 @@ fn parse(
for m in match_finder.matches(pos) { for m in match_finder.matches(pos) {
closest_match = Some(closest_match.unwrap_or(0).max(m.pos)); closest_match = Some(closest_match.unwrap_or(0).max(m.pos));
let offset = pos - m.pos; let offset = pos - m.pos;
if offset <= encoding_config.max_offset {
found_last_offset |= offset as u32 == arrival.state.last_offset(); found_last_offset |= offset as u32 == arrival.state.last_offset();
add_match( add_match(
&mut arrivals, &mut arrivals,
@@ -201,6 +203,7 @@ fn parse(
break 'arrival_loop; break 'arrival_loop;
} }
} }
}
let mut near_matches_left = config.num_near_matches; let mut near_matches_left = config.num_near_matches;
let mut match_pos = last_seen[data[pos] as usize]; let mut match_pos = last_seen[data[pos] as usize];
@@ -209,6 +212,9 @@ fn parse(
&& closest_match.iter().all(|p| *p < match_pos) && closest_match.iter().all(|p| *p < match_pos)
{ {
let offset = pos - match_pos; let offset = pos - match_pos;
if offset > encoding_config.max_offset {
break;
}
let length = match_length(offset); let length = match_length(offset);
assert!(length > 0); assert!(length > 0);
add_match( add_match(

View File

@@ -1,4 +1,5 @@
use crate::{context_state::Context, Config}; use crate::{context_state::Context, Config};
use thiserror::Error;
pub const PROB_BITS: u32 = 8; pub const PROB_BITS: u32 = 8;
pub const ONE_PROB: u32 = 1 << PROB_BITS; pub const ONE_PROB: u32 = 1 << PROB_BITS;
@@ -160,6 +161,10 @@ pub struct RansDecoder<'a> {
const PROB_MASK: u32 = ONE_PROB - 1; const PROB_MASK: u32 = ONE_PROB - 1;
#[derive(Debug, Error)]
#[error("Unexpected end of input")]
pub struct UnexpectedEOF;
impl<'a> RansDecoder<'a> { impl<'a> RansDecoder<'a> {
pub fn new(data: &'a [u8], config: &Config) -> RansDecoder<'a> { pub fn new(data: &'a [u8], config: &Config) -> RansDecoder<'a> {
RansDecoder { RansDecoder {
@@ -178,17 +183,20 @@ impl<'a> RansDecoder<'a> {
self.pos self.pos
} }
pub fn decode_with_context(&mut self, context: &mut Context) -> bool { pub fn decode_with_context(&mut self, context: &mut Context) -> Result<bool, UnexpectedEOF> {
let bit = self.decode_bit(context.prob()); let bit = self.decode_bit(context.prob())?;
context.update(bit); context.update(bit);
bit Ok(bit)
} }
pub fn decode_bit(&mut self, prob: u16) -> bool { pub fn decode_bit(&mut self, prob: u16) -> Result<bool, UnexpectedEOF> {
let prob = prob as u32; let prob = prob as u32;
if self.use_bitstream { if self.use_bitstream {
while self.state < 32768 { while self.state < 32768 {
if self.bits_left == 0 { if self.bits_left == 0 {
if self.pos >= self.data.len() {
return Err(UnexpectedEOF);
}
self.byte = self.data[self.pos]; self.byte = self.data[self.pos];
self.pos += 1; self.pos += 1;
self.bits_left = 8; self.bits_left = 8;
@@ -204,8 +212,11 @@ impl<'a> RansDecoder<'a> {
} }
} else { } else {
while self.state < 4096 { 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;
} }
} }
@@ -218,6 +229,6 @@ impl<'a> RansDecoder<'a> {
}; };
self.state = prob * (self.state >> PROB_BITS) + (self.state & PROB_MASK) - start; self.state = prob * (self.state >> PROB_BITS) + (self.state & PROB_MASK) - start;
bit ^ self.invert_bit_encoding Ok(bit ^ self.invert_bit_encoding)
} }
} }

3
z80_unpacker/.gitignore vendored Normal file
View File

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

11
z80_unpacker/Makefile Normal file
View File

@@ -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

View File

@@ -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

Binary file not shown.

32
z80_unpacker/readme.txt Normal file
View File

@@ -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

381
z80_unpacker/unpack.asm Normal file
View File

@@ -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