Option to use Cloudflare's zlib for compression (#103)

Behind the nightly feature flag for now--stable ASM is coming in Rust 1.27

Cargo.lock

@@ -72,6 +72,11 @@ name = "byteorder"
 version = "1.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 
+[[package]]
+name = "cc"
+version = "1.0.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+
 [[package]]
 name = "cfg-if"
 version = "0.1.2"
@@ -121,6 +126,14 @@ dependencies = [
  "url 1.7.0 (registry+https://github.com/rust-lang/crates.io-index)",
 ]
 
+[[package]]
+name = "cloudflare-zlib-sys"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+dependencies = [
+ "cc 1.0.17 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
 [[package]]
 name = "crc"
 version = "1.8.1"
@@ -348,6 +361,7 @@ dependencies = [
  "byteorder 1.2.2 (registry+https://github.com/rust-lang/crates.io-index)",
  "clap 2.31.2 (registry+https://github.com/rust-lang/crates.io-index)",
  "clippy 0.0.195 (registry+https://github.com/rust-lang/crates.io-index)",
+ "cloudflare-zlib-sys 0.1.2 (registry+https://github.com/rust-lang/crates.io-index)",
  "crc 1.8.1 (registry+https://github.com/rust-lang/crates.io-index)",
  "glob 0.2.11 (registry+https://github.com/rust-lang/crates.io-index)",
  "image 0.19.0 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -717,10 +731,12 @@ dependencies = [
 "checksum build_const 0.2.1 (registry+https://github.com/rust-lang/crates.io-index)" = "39092a32794787acd8525ee150305ff051b0aa6cc2abaf193924f5ab05425f39"
 "checksum byteorder 0.5.3 (registry+https://github.com/rust-lang/crates.io-index)" = "0fc10e8cc6b2580fda3f36eb6dc5316657f812a3df879a44a66fc9f0fdbc4855"
 "checksum byteorder 1.2.2 (registry+https://github.com/rust-lang/crates.io-index)" = "73b5bdfe7ee3ad0b99c9801d58807a9dbc9e09196365b0203853b99889ab3c87"
+"checksum cc 1.0.17 (registry+https://github.com/rust-lang/crates.io-index)" = "49ec142f5768efb5b7622aebc3fdbdbb8950a4b9ba996393cb76ef7466e8747d"
 "checksum cfg-if 0.1.2 (registry+https://github.com/rust-lang/crates.io-index)" = "d4c819a1287eb618df47cc647173c5c4c66ba19d888a6e50d605672aed3140de"
 "checksum clap 2.31.2 (registry+https://github.com/rust-lang/crates.io-index)" = "f0f16b89cbb9ee36d87483dc939fe9f1e13c05898d56d7b230a0d4dff033a536"
 "checksum clippy 0.0.195 (registry+https://github.com/rust-lang/crates.io-index)" = "1d48c23f9c2f4cb3ac08fb972e9a892778f7829c7ae611106887b42cf1749ad2"
 "checksum clippy_lints 0.0.195 (registry+https://github.com/rust-lang/crates.io-index)" = "f490c98750106b5034c6b625b83b69539e42537783cb802220153f53f9a83ffa"
+"checksum cloudflare-zlib-sys 0.1.2 (registry+https://github.com/rust-lang/crates.io-index)" = "1a27574ebbf993003163c445f5f70d3fa9017f09013e5007889991f46aa85ea2"
 "checksum crc 1.8.1 (registry+https://github.com/rust-lang/crates.io-index)" = "d663548de7f5cca343f1e0a48d14dcfb0e9eb4e079ec58883b7251539fa10aeb"
 "checksum crossbeam-deque 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "f739f8c5363aca78cfb059edf753d8f0d36908c348f3d8d1503f03d8b75d9cf3"
 "checksum crossbeam-epoch 0.3.1 (registry+https://github.com/rust-lang/crates.io-index)" = "927121f5407de9956180ff5e936fe3cf4324279280001cd56b669d28ee7e9150"

Cargo.toml

@@ -45,6 +45,10 @@ version = "^2.10.0"
 optional = true
 version = ">=0.0.85"
 
+[dependencies.cloudflare-zlib-sys]
+optional = true
+version = "^0.1.2"
+
 [dependencies.image]
 default-features = false
 features = ["png_codec"]
@@ -60,11 +64,12 @@ binary = [
     "regex",
 ]
 default = ["binary"]
+cfzlib = ["cloudflare-zlib-sys"]
 dev = [
     "nightly-binary",
     "clippy",
 ]
-nightly = []
+nightly = ["cfzlib"]
 nightly-binary = [
     "binary",
     "nightly",

appveyor.yml

@@ -8,6 +8,9 @@ environment:
       CHANNEL: stable
     - TARGET: x86_64-pc-windows-msvc
       CHANNEL: stable
+    - TARGET: x86_64-pc-windows-msvc
+      CHANNEL: nightly
+      TESTFLAGS: --features=cfzlib
 
 # Install Rust and Cargo
 # (Based on from https://github.com/rust-lang/libc/blob/master/appveyor.yml)
@@ -26,7 +29,7 @@ build: false
 # Equivalent to Travis' `script` phase
 test_script:
   - cargo build
-  - cargo test
+  - cargo test %TESTFLAGS%
 
 before_deploy:
   # Generate artifacts for release

src/deflate/mod.rs

@@ -13,8 +13,70 @@ pub fn inflate(data: &[u8]) -> Result<Vec<u8>, PngError> {
 }
 
 /// Compress a data stream using the DEFLATE algorithm
-pub fn deflate(data: &[u8], zc: u8, zs: u8, zw: u8) -> Vec<u8> {
+pub fn deflate(data: &[u8], zc: u8, zs: u8, zw: u8) -> Result<Vec<u8>, PngError> {
-    miniz_stream::compress_to_vec_oxipng(data, zc, zw.into(), zs.into())
+    #[cfg(feature = "cfzlib")]
+    {
+        if is_cfzlib_supported() {
+            return cfzlib_deflate(data, zc, zs, zw)
+        }
+    }
+
+    Ok(miniz_stream::compress_to_vec_oxipng(data, zc, zw.into(), zs.into()))
+}
+
+#[cfg(feature = "cfzlib")]
+fn is_cfzlib_supported() -> bool {
+    #[cfg(target_arch = "x86_64")] {
+        if is_x86_feature_detected!("sse4.2") && is_x86_feature_detected!("pclmulqdq") {
+            return true;
+        }
+    }
+    #[cfg(target_arch = "aarch64")] {
+        if is_arm_feature_detected!("neon") && is_arm_feature_detected!("crc") {
+            return true;
+        }
+    }
+    false
+}
+
+#[cfg(feature = "cfzlib")]
+pub fn cfzlib_deflate(data: &[u8], level: u8, strategy: u8, window_bits: u8) -> Result<Vec<u8>, PngError> {
+    use std::mem;
+    use cloudflare_zlib_sys::*;
+
+    assert!(data.len() < u32::max_value() as usize);
+    unsafe {
+        let mut stream = mem::zeroed();
+        if Z_OK != deflateInit2(
+                    &mut stream,
+                    level.into(),
+                    Z_DEFLATED,
+                    window_bits.into(),
+                    MAX_MEM_LEVEL,
+                    strategy.into()) {
+            return Err(PngError::new("deflateInit2"));
+        }
+
+        let max_size = deflateBound(&mut stream, data.len() as uLong) as usize;
+        // it's important to have the capacity pre-allocated,
+        // as unsafe set_len is called later
+        let mut out = Vec::with_capacity(max_size);
+
+        stream.next_in = data.as_ptr() as *mut _;
+        stream.total_in = data.len() as uLong;
+        stream.avail_in = data.len() as uInt;
+        stream.next_out = out.as_mut_ptr();
+        stream.avail_out = out.capacity() as uInt;
+        if Z_STREAM_END != deflate(&mut stream, Z_FINISH) {
+            return Err(PngError::new("deflate"));
+        }
+        if Z_OK != deflateEnd(&mut stream) {
+            return Err(PngError::new("deflateEnd"));
+        }
+        debug_assert!(stream.total_out as usize <= out.capacity());
+        out.set_len(stream.total_out as usize);
+        return Ok(out);
+    }
 }
 
 pub fn zopfli_deflate(data: &[u8]) -> Result<Vec<u8>, PngError> {

src/lib.rs

@@ -10,6 +10,8 @@ extern crate miniz_oxide;
 extern crate num_cpus;
 extern crate rayon;
 extern crate zopfli;
+#[cfg(feature = "cfzlib")]
+extern crate cloudflare_zlib_sys;
 
 use image::{DynamicImage, GenericImage, ImageFormat, Pixel};
 use png::PngData;
@@ -499,7 +501,10 @@ fn optimize_png(
                 let new_idat = if opts.deflate == Deflaters::Zlib {
                     deflate::deflate(filtered, trial.compression, trial.strategy, opts.window)
                 } else {
-                    deflate::zopfli_deflate(filtered).unwrap()
+                    deflate::zopfli_deflate(filtered)
+                };
+                let new_idat = if let Ok(n) = new_idat {n} else {
+                    return None;
                 };
 
                 if opts.verbosity == Some(1) {

src/png/mod.rs

@@ -603,27 +603,28 @@ impl PngData {
         assert!(!alphas.is_empty());
         let best = alphas
             .iter()
-            .map(|alpha| {
+            .filter_map(|alpha| {
                 let mut image = self.clone();
                 image.reduce_alpha_channel(*alpha);
-                let size = STD_FILTERS
+                STD_FILTERS
                     .iter()
-                    .map(|f| {
+                    .filter_map(|f| {
                         deflate::deflate(
                             &image.filter_image(*f),
                             STD_COMPRESSION,
                             STD_STRATEGY,
                             STD_WINDOW,
-                        ).len()
+                        ).ok()
+                        .as_ref().map(|l| l.len())
                     })
                     .min()
-                    .unwrap();
+                    .map(|size| (size, image))
-                (size, image)
             })
-            .min_by_key(|&(size, _)| size)
+            .min_by_key(|&(size, _)| size);
-            .unwrap();
 
-        self.raw_data = best.1.raw_data;
+        if let Some(best) = best {
+            self.raw_data = best.1.raw_data;
+        }
     }
 
     pub fn reduce_alpha_channel(&mut self, optim: AlphaOptim) -> bool {