better parity contexts integration for literal length

Cargo.lock

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

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "upkr"
-version = "0.2.0"
+version = "0.2.0-pre3"
 edition = "2021"
 
 [profile.release]

README.md

@@ -2,57 +2,13 @@
 
 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 differentiate between literals at odd or even addresses (by default) and that I went with rANS/rABS instead of a range coder.
+Upkr doesn't differnetiate between literals at odd or even addresses and that I went with rANS/rABS instead of a range coder.
 
-Compression rate is on par with Shrinkler.
-
-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.
+At this point, Upkr should still be considered unstable - the compressed format is not very likely to change but I still want
+to keep that option open a little longer.
 
 ## 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
-```
+* Ryg's [sample rANS implementation](https://github.com/rygorous/ryg_rans)

release/.gitignore

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

release/Makefile

@@ -1,35 +0,0 @@
-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/greedy_packer.rs

@@ -12,6 +12,8 @@ pub fn pack(
     let mut rans_coder = RansCoder::new(config);
     let mut state = lz::CoderState::new(config);
 
+    let mut literal = vec![];
+
     let mut pos = 0;
     while pos < data.len() {
         if let Some(ref mut cb) = progress_callback {
@@ -22,6 +24,10 @@ pub fn pack(
             let max_offset = config.max_offset.min(1 << (m.length * 3 - 1).min(31));
             let offset = pos - m.pos;
             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,
@@ -43,6 +49,10 @@ pub fn pack(
                     .count()
                     .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,
@@ -55,11 +65,14 @@ pub fn pack(
         }
 
         if !encoded_match {
-            lz::Op::Literal(data[pos]).encode(&mut rans_coder, &mut state, config);
+            literal.push(data[pos]);
             pos += 1;
         }
     }
 
+    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/lz.rs

@@ -3,36 +3,53 @@ 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, config: &Config) {
-        let literal_base = state.pos % state.parity_contexts * 256;
         match self {
-            &Op::Literal(lit) => {
-                encode_bit(coder, state, literal_base, !config.is_match_bit);
-                let mut context_index = 1;
-                for i in (0..8).rev() {
-                    let bit = (lit >> i) & 1 != 0;
-                    encode_bit(coder, state, literal_base + context_index, bit);
-                    context_index = (context_index << 1) | bit as usize;
+            &Op::Literal(ref lit) => {
+                assert!(state.prev_was_match);
+                encode_length(
+                    coder,
+                    state,
+                    256 + state.pos % state.parity_contexts * 320,
+                    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;
-                state.pos += 1;
             }
             &Op::Match { offset, len } => {
-                encode_bit(coder, state, literal_base, config.is_match_bit);
+                if state.prev_was_match {
+                    encode_length(
+                        coder,
+                        state,
+                        256 + state.pos % state.parity_contexts * 320,
+                        1,
+                        config,
+                    );
+                }
                 let mut new_offset = true;
                 if !state.prev_was_match && !config.no_repeated_offsets {
                     new_offset = offset != state.last_offset;
                     encode_bit(
                         coder,
                         state,
-                        256 * state.parity_contexts,
+                        320 * state.parity_contexts,
                         new_offset == config.new_offset_bit,
                     );
                 }
@@ -41,14 +58,14 @@ impl Op {
                     encode_length(
                         coder,
                         state,
-                        256 * state.parity_contexts + 1,
+                        320 * state.parity_contexts + 1,
                         offset + if config.eof_in_length { 0 } else { 1 },
                         config,
                     );
                     state.last_offset = offset;
                 }
                 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, 320 * state.parity_contexts + 65, len, config);
                 state.prev_was_match = true;
                 state.pos += len as usize;
             }
@@ -57,25 +74,28 @@ impl Op {
 }
 
 pub fn encode_eof(coder: &mut dyn EntropyCoder, state: &mut CoderState, config: &Config) {
-    encode_bit(
-        coder,
-        state,
-        state.pos % state.parity_contexts * 256,
-        config.is_match_bit,
-    );
+    if state.prev_was_match {
+        encode_length(
+            coder,
+            state,
+            256 + state.pos % state.parity_contexts * 320,
+            1,
+            config,
+        );
+    }
     if !state.prev_was_match && !config.no_repeated_offsets {
         encode_bit(
             coder,
             state,
-            256 * state.parity_contexts,
+            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, 256 * state.parity_contexts + 1, 1, config);
+        encode_length(coder, state, 320 * state.parity_contexts + 1, 1, config);
     }
     if config.eof_in_length {
-        encode_length(coder, state, 256 * state.parity_contexts + 65, 1, config);
+        encode_length(coder, state, 320 * state.parity_contexts + 65, 1, config);
     }
 }
 
@@ -119,9 +139,9 @@ pub struct CoderState {
 impl CoderState {
     pub fn new(config: &Config) -> CoderState {
         CoderState {
-            contexts: ContextState::new((1 + 255) * config.parity_contexts + 1 + 64 + 64, config),
+            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,
         }
@@ -168,7 +188,8 @@ pub fn unpack_internal(
     max_size: usize,
 ) -> Result<isize, UnpackError> {
     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((64 + 256) * config.parity_contexts + 1 + 64 + 64, &config);
     let mut offset = usize::MAX;
     let mut position = 0usize;
     let mut prev_was_match = false;
@@ -199,50 +220,14 @@ pub fn unpack_internal(
 
     loop {
         margin = margin.max(position as isize - decoder.pos() as isize);
-        let literal_base = position % config.parity_contexts * 256;
-        if decoder.decode_with_context(&mut contexts.context_mut(literal_base))?
-            == config.is_match_bit
-        {
-            if config.no_repeated_offsets
-                || prev_was_match
-                || decoder
-                    .decode_with_context(&mut contexts.context_mut(256 * config.parity_contexts))?
-                    == config.new_offset_bit
-            {
-                offset = decode_length(
-                    &mut decoder,
-                    &mut contexts,
-                    256 * config.parity_contexts + 1,
-                    &config,
-                )? - if config.eof_in_length { 0 } else { 1 };
-                if offset == 0 {
-                    break;
-                }
-            }
-            let length = decode_length(
-                &mut decoder,
-                &mut contexts,
-                256 * 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;
-        } else {
+        let literal_length = decode_length(
+            &mut decoder,
+            &mut contexts,
+            256 + position % config.parity_contexts * 320,
+            config,
+        )? - 1;
+        for _ in 0..literal_length {
+            let literal_base = position % config.parity_contexts * 320;
             let mut context_index = 1;
             let mut byte = 0;
             for i in (0..8).rev() {
@@ -259,6 +244,46 @@ pub fn unpack_internal(
             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;
     }
 
     if position > max_size {

src/main.rs

@@ -61,10 +61,6 @@ fn main() -> Result<()> {
             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()?),
@@ -159,8 +155,6 @@ 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!(

src/parsing_packer.rs

@@ -15,7 +15,7 @@ pub fn pack(
     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(config);
@@ -32,9 +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,
 }
 
@@ -130,6 +136,7 @@ fn parse(
                     op,
                 })),
                 state,
+                literal_prefix: None,
                 cost: arrival.cost + cost_counter.cost(),
             },
             max_arrivals,
@@ -141,6 +148,7 @@ fn parse(
         Arrival {
             parse: None,
             state: lz::CoderState::new(encoding_config),
+            literal_prefix: None,
             cost: 0.0,
         },
         max_arrivals,
@@ -252,19 +260,26 @@ fn parse(
             }
 
             cost_counter.reset();
-            let mut state = arrival.state;
-            let op = lz::Op::Literal(data[pos]);
+            let (arrival, mut prefix) = if let Some(prefix) = arrival.literal_prefix {
+                (prefix.arrival, prefix.prefix)
+            } 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,
             );