feat(max_perf): enable fast_lzma

Author: chyyran

README.md

@@ -126,6 +126,13 @@ As a proof of concept, chd-rs implements an *extremely* basic reimplementation o
 The results from rchdman should be identical from chdman. rchdman is intended to be basic and does not implement multithreading or other functions, so in general it is slower than chdman. There are
 no plans to implement write-operations into rchdman.
 
+## Performance
+By default, chd-rs uses pure Rust codecs but if maximum performance is needed, `max_perf` can be enabled. This enables the zlib-ng backend of [flate2](https://crates.io/crates/flate2)
+as well as using experimental APIs in a custom [lzma-rs fork](https://github.com/SnowflakePowered/lzma-rs/tree/feature-perf-experiments) for some improvements in
+performance at the expense of higher memory usage. Combined with `codegen-units=1` and [Profile Guided Optimization](https://github.com/vadimcn/cargo-pgo), chd-rs is within 1% of libchdr performance.
+
+Without `max_perf`, chd-rs is already within 15% of libchdr without needing to link with C libraries like zlib-ng.
+
 ## `libchdr` API
 ⚠️*The C API has not been heavily tested. Use at your own risk.* ⚠️
 

chd-rs/src/compression/lzma.rs

@@ -100,6 +100,7 @@ impl CompressionCodecType for LzmaCodec {
     }
 }
 
+#[cfg(not(feature = "fast_lzma"))]
 impl CodecImplementation for LzmaCodec {
     fn new(hunk_size: u32) -> Result<Self> {
         Ok(LzmaCodec {
@@ -129,3 +130,37 @@ impl CodecImplementation for LzmaCodec {
         Ok(DecompressResult::new(len, read.position() as usize))
     }
 }
+
+#[cfg(feature = "fast_lzma")]
+impl CodecImplementation for LzmaCodec {
+    fn new(hunk_size: u32) -> Result<Self> {
+        let dict_size = get_lzma_dict_size(9, hunk_size);
+        Ok(LzmaCodec {
+            engine: LzmaDecoder::new_with_buffer(
+                LzmaParams::new(
+                    LzmaProperties {
+                        lc: 3,
+                        lp: 0,
+                        pb: 2,
+                    },
+                    dict_size,
+                    None,
+                ),
+                None,
+                vec![0; dict_size as usize],
+            )
+            .map_err(|_| Error::DecompressionError)?,
+        })
+    }
+
+    fn decompress(&mut self, input: &[u8], mut output: &mut [u8]) -> Result<DecompressResult> {
+        use lzma_rs::decompress::raw::LzAccumBuffer;
+        let mut read = Cursor::new(input);
+        let len = output.len();
+        self.engine.reset(Some(Some(len as u64)));
+        self.engine
+            .decompress_with_buffer::<LzAccumBuffer<_>, _, _>(&mut read, &mut output)
+            .map_err(|_| Error::DecompressionError)?;
+        Ok(DecompressResult::new(len, read.position() as usize))
+    }
+}

rchdman/Cargo.toml

@@ -5,7 +5,7 @@ edition = "2021"
 publish = false
 
 [dependencies]
-chd = { path = "../chd-rs", features = ["unstable_lending_iterators"]}
+chd = { path = "../chd-rs", features = ["unstable_lending_iterators", "max_perf"]}
 clap = { version = "3", features = ["derive"] }
 anyhow = "1"
 thousands = "0.2.0"

rchdman/src/main.rs

@@ -1,7 +1,8 @@
 use anyhow::anyhow;
 use chd::header::{CodecType, Header};
 use chd::iter::LendingIterator;
-use chd::map::{CompressionTypeLegacy, MapEntry, CompressionTypeV5};
+use chd::map::{CompressionTypeLegacy, CompressionTypeV5, MapEntry};
+use chd::metadata::Metadata;
 use chd::Chd;
 use clap::{Parser, Subcommand};
 use num_traits::cast::FromPrimitive;
@@ -12,7 +13,6 @@ use std::io::{BufReader, BufWriter, Read, Seek, Write};
 use std::path::{Path, PathBuf};
 use std::time::Instant;
 use thousands::Separable;
-use chd::metadata::Metadata;
 
 fn validate_file_exists(s: &OsStr) -> Result<PathBuf, std::io::Error> {
     let path = PathBuf::from(s);