teddy: port teddy searcher to std::arch

This commit ports the Teddy searcher to use std::arch and moves off the
portable SIMD vector API. Performance remains the same, and it looks
like the codegen is identical, which is great!

This also makes the `simd-accel` feature a no-op and adds a new
`unstable` feature which will enable the Teddy optimization. The `-C
target-feature` or `-C target-cpu` settings are no longer necessary,
since this will now do runtime target feature detection.

We also add a new `unstable` feature to the regex crate, which will
enable this new use of std::arch. Once enabled, the Teddy optimizations
becomes available automatically without any additional compile time
flags.

Author: BurntSushi

Cargo.toml

@@ -31,8 +31,6 @@ memchr = "2.0.0"
 thread_local = "0.3.2"
 # For parsing regular expressions.
 regex-syntax = { path = "regex-syntax", version = "0.5.1" }
-# For accelerating text search.
-simd = { version = "0.2.1", optional = true }
 # For compiling UTF-8 decoding into automata.
 utf8-ranges = "1.0.0"
 
@@ -45,10 +43,20 @@ quickcheck = { version = "0.6", default-features = false }
 rand = "0.4"
 
 [features]
-# Enable to use the unstable pattern traits defined in std.
+# We don't enable any features by default currently, but if the compiler
+# supports a specific type of feature, then regex's build.rs might enable
+# some default features.
+default = []
+# A blanket feature that governs whether unstable features are enabled or not.
+# Unstable features are disabled by default, and typically rely on unstable
+# features in rustc itself.
+unstable = ["pattern"]
+# Enable to use the unstable pattern traits defined in std. This is enabled
+# by default if the unstable feature is enabled.
 pattern = []
 # Enable to use simd acceleration.
-simd-accel = ["simd"]
+# Note that this is deprecated and is a no-op.
+simd-accel = []
 
 [lib]
 # There are no benchmarks in the library code itself

bench/Cargo.toml

@@ -18,7 +18,7 @@ libc = "0.2"
 onig = { version = "3", optional = true }
 libpcre-sys = { version = "0.2", optional = true }
 memmap = "0.6"
-regex = { version = "0.2.0", path = "..", features = ["simd-accel"] }
+regex = { version = "0.2.0", path = "..", features = ["unstable"] }
 regex-syntax = { version = "0.5.0", path = "../regex-syntax" }
 serde = "1"
 serde_derive = "1"

build.rs

@@ -0,0 +1,26 @@
+use std::env;
+use std::ffi::OsString;
+use std::process::Command;
+
+fn main() {
+    let rustc = env::var_os("RUSTC").unwrap_or(OsString::from("rustc"));
+    let output = Command::new(&rustc)
+        .arg("--version")
+        .output()
+        .unwrap()
+        .stdout;
+    let version = String::from_utf8(output).unwrap();
+
+    // If we're using nightly Rust, then we can enable vector optimizations.
+    // Note that these aren't actually activated unless the `nightly` feature
+    // is enabled.
+    //
+    // We also don't activate these if we've explicitly disabled auto
+    // optimizations. Disabling auto optimizations is intended for use in
+    // tests, so that we can reliably test fallback implementations.
+    if env::var_os("CARGO_CFG_REGEX_DISABLE_AUTO_OPTIMIZATIONS").is_none() {
+        if version.contains("nightly") {
+            println!("cargo:rustc-cfg=regex_runtime_teddy_ssse3");
+        }
+    }
+}

src/exec.rs

@@ -23,7 +23,7 @@ use compile::Compiler;
 use dfa;
 use error::Error;
 use input::{ByteInput, CharInput};
-use literals::LiteralSearcher;
+use literal::LiteralSearcher;
 use pikevm;
 use prog::Program;
 use re_builder::RegexOptions;

src/input.rs

@@ -16,7 +16,7 @@ use std::u32;
 
 use syntax;
 
-use literals::LiteralSearcher;
+use literal::LiteralSearcher;
 use prog::InstEmptyLook;
 use utf8::{decode_utf8, decode_last_utf8};
 

src/lib.rs

@@ -520,14 +520,15 @@ another matching engine with fixed memory requirements.
 #![deny(missing_docs)]
 #![cfg_attr(test, deny(warnings))]
 #![cfg_attr(feature = "pattern", feature(pattern))]
-#![cfg_attr(feature = "simd-accel", feature(cfg_target_feature))]
+#![cfg_attr(feature = "unstable", feature(target_feature, stdsimd))]
 
 extern crate aho_corasick;
 extern crate memchr;
 extern crate thread_local;
-#[macro_use] #[cfg(test)] extern crate quickcheck;
+#[cfg(test)]
+#[macro_use]
+extern crate quickcheck;
 extern crate regex_syntax as syntax;
-#[cfg(feature = "simd-accel")] extern crate simd;
 extern crate utf8_ranges;
 
 pub use error::Error;
@@ -645,7 +646,7 @@ mod exec;
 mod expand;
 mod freqs;
 mod input;
-mod literals;
+mod literal;
 #[cfg(feature = "pattern")]
 mod pattern;
 mod pikevm;
@@ -655,12 +656,9 @@ mod re_bytes;
 mod re_set;
 mod re_trait;
 mod re_unicode;
-#[cfg(feature = "simd-accel")]
-mod simd_accel;
-#[cfg(not(feature = "simd-accel"))]
-#[path = "simd_fallback/mod.rs"]
-mod simd_accel;
 mod sparse;
+#[cfg(feature = "unstable")]
+mod vector;
 
 /// The `internal` module exists to support suspicious activity, such as
 /// testing different matching engines and supporting the `regex-debug` CLI
@@ -670,6 +668,6 @@ pub mod internal {
     pub use compile::Compiler;
     pub use exec::{Exec, ExecBuilder};
     pub use input::{Char, Input, CharInput, InputAt};
-    pub use literals::LiteralSearcher;
+    pub use literal::LiteralSearcher;
     pub use prog::{Program, Inst, EmptyLook, InstRanges};
 }

src/literals.rs renamed to src/literal/mod.rs

@@ -16,8 +16,9 @@ use memchr::{memchr, memchr2, memchr3};
 use syntax::hir::literal::{Literal, Literals};
 
 use freqs::BYTE_FREQUENCIES;
+use self::teddy_ssse3::Teddy;
 
-use simd_accel::teddy128::{Teddy, is_teddy_128_available};
+mod teddy_ssse3;
 
 /// A prefix extracted from a compiled regular expression.
 ///
@@ -219,7 +220,7 @@ impl Matcher {
             }
         }
         let is_aho_corasick_fast = sset.dense.len() == 1 && sset.all_ascii;
-        if is_teddy_128_available() && !is_aho_corasick_fast {
+        if Teddy::available() && !is_aho_corasick_fast {
             // Only try Teddy if Aho-Corasick can't use memchr on an ASCII
             // byte. Also, in its current form, Teddy doesn't scale well to
             // lots of literals.

src/simd_fallback/teddy128.rs renamed to src/literal/teddy_ssse3/fallback.rs

@@ -1,9 +1,5 @@
 use syntax::hir::literal::Literals;
 
-pub fn is_teddy_128_available() -> bool {
-    false
-}
-
 #[derive(Debug, Clone)]
 pub struct Teddy(());
 
@@ -15,6 +11,7 @@ pub struct Match {
 }
 
 impl Teddy {
+    pub fn available() -> bool { false }
     pub fn new(_pats: &Literals) -> Option<Teddy> { None }
     pub fn patterns(&self) -> &[Vec<u8>] { &[] }
     pub fn len(&self) -> usize { 0 }