Simple module compilation cache (#203)

* Simple module compilation cache

* Fix base64 encoding bug

* Use warn! everywhere in cache system

* Remove unused import

* Temporary workaround for long path on Windows

* Remove unused import for non-windows builds

* Add command line argument to enable cache system + apply minor review feedback

wasmtime-environ/Cargo.toml

@@ -12,14 +12,22 @@ readme = "README.md"
 edition = "2018"
 
 [dependencies]
-cranelift-codegen = "0.33.0"
-cranelift-entity = "0.33.0"
-cranelift-wasm = "0.33.0"
+cranelift-codegen = { version = "0.36.0", features = ["enable-serde"] }
+cranelift-entity = { version = "0.36.0", features = ["enable-serde"] }
+cranelift-wasm = { version = "0.36.0", features = ["enable-serde"] }
 lightbeam = { path = "../lightbeam", optional = true }
 failure = { version = "0.1.3", default-features = false }
 failure_derive = { version = "0.1.3", default-features = false }
 indexmap = "1.0.2"
 rayon = "1.1"
+directories = "2.0.1"
+sha2 = "0.8.0"
+base64 = "0.10.1"
+serde = { version = "1.0.94", features = ["derive"] }
+bincode = "1.1.4"
+lazy_static = "1.3.0"
+spin = "0.5.0"
+log = { version = "0.4.6", default-features = false }
 
 [features]
 default = ["std"]

wasmtime-environ/src/address_map.rs

@@ -4,10 +4,11 @@
 use cranelift_codegen::ir;
 use cranelift_entity::PrimaryMap;
 use cranelift_wasm::DefinedFuncIndex;
+use serde::{Deserialize, Serialize};
 use std::vec::Vec;
 
 /// Single source location to generated address mapping.
-#[derive(Debug, Clone)]
+#[derive(Serialize, Deserialize, Debug, Clone)]
 pub struct InstructionAddressMap {
     /// Original source location.
     pub srcloc: ir::SourceLoc,
@@ -20,7 +21,7 @@ pub struct InstructionAddressMap {
 }
 
 /// Function and its instructions addresses mappings.
-#[derive(Debug, Clone)]
+#[derive(Serialize, Deserialize, Debug, Clone)]
 pub struct FunctionAddressMap {
     /// Instructions maps.
     /// The array is sorted by the InstructionAddressMap::code_offset field.

wasmtime-environ/src/cache.rs

@@ -0,0 +1,342 @@
+use crate::address_map::ModuleAddressMap;
+use crate::compilation::{CodeAndJTOffsets, Compilation, Relocations};
+use crate::module::Module;
+use cranelift_codegen::ir;
+use cranelift_codegen::isa;
+use directories::ProjectDirs;
+use lazy_static::lazy_static;
+use log::warn;
+use serde::de::{self, Deserialize, Deserializer, MapAccess, SeqAccess, Visitor};
+use serde::ser::{self, Serialize, SerializeSeq, SerializeStruct, Serializer};
+#[cfg(windows)]
+use std::ffi::OsString;
+use std::fmt;
+use std::fs;
+use std::io;
+#[cfg(windows)]
+use std::path::Path;
+use std::path::PathBuf;
+
+/// Module for configuring the cache system.
+pub mod conf {
+    use spin::Once;
+
+    // Private static, so only internal function can access it.
+    static CACHE_ENABLED: Once<bool> = Once::new();
+
+    /// Returns true if and only if the cache is enabled.
+    pub fn cache_enabled() -> bool {
+        // Not everyone knows about the cache system, i.e. the tests,
+        // so the default is false.
+        *CACHE_ENABLED.call_once(|| false)
+    }
+
+    /// Initializes the cache system. Should be called exactly once,
+    /// and before using the cache system. Otherwise it can panic.
+    pub fn init(enabled: bool) {
+        // init() should be called exactly once
+        assert!(CACHE_ENABLED.r#try().is_none());
+        let val = *CACHE_ENABLED.call_once(|| enabled);
+        // But multiple threads can pass the first assertion, so let's guarantee consistency:
+        assert!(val == enabled);
+    }
+}
+
+lazy_static! {
+    static ref CACHE_DIR: Option<PathBuf> =
+        match ProjectDirs::from("", "CraneStation", "wasmtime") {
+            Some(proj_dirs) => {
+                let cache_dir = proj_dirs.cache_dir();
+                // Temporary workaround for: https://github.com/rust-lang/rust/issues/32689
+                #[cfg(windows)]
+                let mut long_path = OsString::from("\\\\?\\");
+                #[cfg(windows)]
+                let cache_dir = {
+                    if cache_dir.starts_with("\\\\?\\") {
+                        cache_dir
+                    }
+                    else {
+                        long_path.push(cache_dir.as_os_str());
+                        Path::new(&long_path)
+                    }
+                };
+                match fs::create_dir_all(cache_dir) {
+                    Ok(()) => (),
+                    Err(err) => warn!("Unable to create cache directory, failed with: {}", err),
+                };
+                Some(cache_dir.to_path_buf())
+            }
+            None => {
+                warn!("Unable to find cache directory");
+                None
+            }
+        };
+}
+
+pub struct ModuleCacheEntry {
+    mod_cache_path: Option<PathBuf>,
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+pub struct ModuleCacheData {
+    compilation: Compilation,
+    relocations: Relocations,
+    address_transforms: ModuleAddressMap,
+}
+
+type ModuleCacheDataTupleType = (Compilation, Relocations, ModuleAddressMap);
+
+impl ModuleCacheEntry {
+    pub fn new(module: &Module, _isa: &dyn isa::TargetIsa, _generate_debug_info: bool) -> Self {
+        // TODO: cache directory hierarchy with isa name, compiler name & git revision, and files with flag if debug symbols are available
+        let mod_cache_path = if conf::cache_enabled() {
+            CACHE_DIR.clone().and_then(|p| {
+                module.hash.map(|hash| {
+                    p.join(format!(
+                        "mod-{}",
+                        base64::encode_config(&hash, base64::URL_SAFE_NO_PAD) // standard encoding uses '/' which can't be used for filename
+                    ))
+                })
+            })
+        } else {
+            None
+        };
+
+        ModuleCacheEntry { mod_cache_path }
+    }
+
+    pub fn get_data(&self) -> Option<ModuleCacheData> {
+        if let Some(p) = &self.mod_cache_path {
+            match fs::read(p) {
+                Ok(cache_bytes) => match bincode::deserialize(&cache_bytes[..]) {
+                    Ok(data) => Some(data),
+                    Err(err) => {
+                        warn!("Failed to deserialize cached code: {}", err);
+                        None
+                    }
+                },
+                Err(_) => None,
+            }
+        } else {
+            None
+        }
+    }
+
+    pub fn update_data(&self, data: &ModuleCacheData) {
+        if let Some(p) = &self.mod_cache_path {
+            let cache_buf = match bincode::serialize(&data) {
+                Ok(data) => data,
+                Err(err) => {
+                    warn!("Failed to serialize cached code: {}", err);
+                    return;
+                }
+            };
+            match fs::write(p, &cache_buf) {
+                Ok(()) => (),
+                Err(err) => {
+                    warn!(
+                        "Failed to write cached code to disk, path: {}, message: {}",
+                        p.display(),
+                        err
+                    );
+                    match fs::remove_file(p) {
+                        Ok(()) => (),
+                        Err(err) => {
+                            if err.kind() != io::ErrorKind::NotFound {
+                                warn!(
+                                    "Failed to cleanup invalid cache, path: {}, message: {}",
+                                    p.display(),
+                                    err
+                                );
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+}
+
+impl ModuleCacheData {
+    pub fn from_tuple(data: ModuleCacheDataTupleType) -> Self {
+        Self {
+            compilation: data.0,
+            relocations: data.1,
+            address_transforms: data.2,
+        }
+    }
+
+    pub fn to_tuple(self) -> ModuleCacheDataTupleType {
+        (self.compilation, self.relocations, self.address_transforms)
+    }
+}
+
+//-////////////////////////////////////////////////////////////////////
+// Serialization and deserialization of type containing SecondaryMap //
+//-////////////////////////////////////////////////////////////////////
+
+enum JtOffsetsWrapper<'a> {
+    Ref(&'a ir::JumpTableOffsets), // for serialization
+    Data(ir::JumpTableOffsets),    // for deserialization
+}
+
+impl Serialize for CodeAndJTOffsets {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let mut cajto = serializer.serialize_struct("CodeAndJTOffsets", 2)?;
+        cajto.serialize_field("body", &self.body)?;
+        cajto.serialize_field("jt_offsets", &JtOffsetsWrapper::Ref(&self.jt_offsets))?;
+        cajto.end()
+    }
+}
+
+impl<'de> Deserialize<'de> for CodeAndJTOffsets {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        #[derive(serde::Deserialize)]
+        #[serde(field_identifier, rename_all = "lowercase")]
+        enum Field {
+            Body,
+            JtOffsets,
+        };
+
+        struct CodeAndJTOffsetsVisitor;
+
+        impl<'de> Visitor<'de> for CodeAndJTOffsetsVisitor {
+            type Value = CodeAndJTOffsets;
+
+            fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
+                formatter.write_str("struct CodeAndJTOffsets")
+            }
+
+            fn visit_seq<V>(self, mut seq: V) -> Result<Self::Value, V::Error>
+            where
+                V: SeqAccess<'de>,
+            {
+                let body = seq
+                    .next_element()?
+                    .ok_or_else(|| de::Error::invalid_length(0, &self))?;
+                let jt_offsets = seq
+                    .next_element()?
+                    .ok_or_else(|| de::Error::invalid_length(1, &self))?;
+                match jt_offsets {
+                    JtOffsetsWrapper::Data(jt_offsets) => Ok(CodeAndJTOffsets { body, jt_offsets }),
+                    JtOffsetsWrapper::Ref(_) => Err(de::Error::custom(
+                        "Received invalid variant of JtOffsetsWrapper",
+                    )),
+                }
+            }
+
+            fn visit_map<V>(self, mut map: V) -> Result<Self::Value, V::Error>
+            where
+                V: MapAccess<'de>,
+            {
+                let mut body = None;
+                let mut jt_offsets = None;
+                while let Some(key) = map.next_key()? {
+                    match key {
+                        Field::Body => {
+                            if body.is_some() {
+                                return Err(de::Error::duplicate_field("body"));
+                            }
+                            body = Some(map.next_value()?);
+                        }
+                        Field::JtOffsets => {
+                            if jt_offsets.is_some() {
+                                return Err(de::Error::duplicate_field("jt_offsets"));
+                            }
+                            jt_offsets = Some(map.next_value()?);
+                        }
+                    }
+                }
+
+                let body = body.ok_or_else(|| de::Error::missing_field("body"))?;
+                let jt_offsets =
+                    jt_offsets.ok_or_else(|| de::Error::missing_field("jt_offsets"))?;
+                match jt_offsets {
+                    JtOffsetsWrapper::Data(jt_offsets) => Ok(CodeAndJTOffsets { body, jt_offsets }),
+                    JtOffsetsWrapper::Ref(_) => Err(de::Error::custom(
+                        "Received invalid variant of JtOffsetsWrapper",
+                    )),
+                }
+            }
+        }
+
+        const FIELDS: &'static [&'static str] = &["body", "jt_offsets"];
+        deserializer.deserialize_struct("CodeAndJTOffsets", FIELDS, CodeAndJTOffsetsVisitor)
+    }
+}
+
+impl Serialize for JtOffsetsWrapper<'_> {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        match self {
+            JtOffsetsWrapper::Ref(data) => {
+                // TODO: bincode encodes option as "byte for Some/None" and then optionally the content
+                // TODO: we can actually optimize it by encoding manually bitmask, then elements
+                let default_val = data.get_default();
+                let mut seq = serializer.serialize_seq(Some(1 + data.len()))?;
+                seq.serialize_element(&Some(default_val))?;
+                for e in data.values() {
+                    let some_e = Some(e);
+                    seq.serialize_element(if e == default_val { &None } else { &some_e })?;
+                }
+                seq.end()
+            }
+            JtOffsetsWrapper::Data(_) => Err(ser::Error::custom(
+                "Received invalid variant of JtOffsetsWrapper",
+            )),
+        }
+    }
+}
+
+impl<'de> Deserialize<'de> for JtOffsetsWrapper<'_> {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        struct JtOffsetsWrapperVisitor;
+
+        impl<'de> Visitor<'de> for JtOffsetsWrapperVisitor {
+            type Value = JtOffsetsWrapper<'static>;
+
+            fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
+                formatter.write_str("struct JtOffsetsWrapper")
+            }
+
+            fn visit_seq<A>(self, mut seq: A) -> Result<Self::Value, A::Error>
+            where
+                A: SeqAccess<'de>,
+            {
+                match seq.next_element()? {
+                    Some(Some(default_val)) => {
+                        let mut m = cranelift_entity::SecondaryMap::with_default(default_val);
+                        let mut idx = 0;
+                        while let Some(val) = seq.next_element()? {
+                            let val: Option<_> = val; // compiler can't infer the type, and this line is needed
+                            match ir::JumpTable::with_number(idx) {
+                                Some(jt_idx) => m[jt_idx] = val.unwrap_or(default_val),
+                                None => {
+                                    return Err(serde::de::Error::custom(
+                                        "Invalid JumpTable reference",
+                                    ))
+                                }
+                            };
+                            idx += 1;
+                        }
+                        Ok(JtOffsetsWrapper::Data(m))
+                    }
+                    _ => Err(serde::de::Error::custom("Default value required")),
+                }
+            }
+        }
+
+        deserializer.deserialize_seq(JtOffsetsWrapperVisitor {})
+    }
+}

wasmtime-environ/src/compilation.rs

@@ -7,6 +7,7 @@ use crate::module_environ::FunctionBodyData;
 use cranelift_codegen::{binemit, ir, isa, CodegenError};
 use cranelift_entity::PrimaryMap;
 use cranelift_wasm::{DefinedFuncIndex, FuncIndex, WasmError};
+use serde::{Deserialize, Serialize};
 use std::ops::Range;
 use std::vec::Vec;
 
@@ -23,7 +24,7 @@ pub struct CodeAndJTOffsets {
 type Functions = PrimaryMap<DefinedFuncIndex, CodeAndJTOffsets>;
 
 /// The result of compiling a WebAssembly module's functions.
-#[derive(Debug)]
+#[derive(Deserialize, Serialize, Debug)]
 pub struct Compilation {
     /// Compiled machine code for the function bodies.
     functions: Functions,
@@ -94,7 +95,7 @@ impl<'a> Iterator for Iter<'a> {
 }
 
 /// A record of a relocation to perform.
-#[derive(Debug, Clone)]
+#[derive(Serialize, Deserialize, Debug, Clone)]
 pub struct Relocation {
     /// The relocation code.
     pub reloc: binemit::Reloc,
@@ -107,7 +108,7 @@ pub struct Relocation {
 }
 
 /// Destination function. Can be either user function or some special one, like `memory.grow`.
-#[derive(Debug, Copy, Clone)]
+#[derive(Serialize, Deserialize, Debug, Copy, Clone)]
 pub enum RelocationTarget {
     /// The user function index.
     UserFunc(FuncIndex),

wasmtime-environ/src/cranelift.rs

@@ -1,6 +1,7 @@
 //! Support for compiling with Cranelift.
 
 use crate::address_map::{FunctionAddressMap, InstructionAddressMap, ModuleAddressMap};
+use crate::cache::{ModuleCacheData, ModuleCacheEntry};
 use crate::compilation::{
     CodeAndJTOffsets, Compilation, CompileError, Relocation, RelocationTarget, Relocations,
 };
@@ -123,73 +124,90 @@ impl crate::compilation::Compiler for Cranelift {
         isa: &dyn isa::TargetIsa,
         generate_debug_info: bool,
     ) -> Result<(Compilation, Relocations, ModuleAddressMap), CompileError> {
-        let mut functions = PrimaryMap::with_capacity(function_body_inputs.len());
-        let mut relocations = PrimaryMap::with_capacity(function_body_inputs.len());
-        let mut address_transforms = PrimaryMap::with_capacity(function_body_inputs.len());
+        let cache_entry = ModuleCacheEntry::new(module, isa, generate_debug_info);
 
-        function_body_inputs
-            .into_iter()
-            .collect::<Vec<(DefinedFuncIndex, &FunctionBodyData<'data>)>>()
-            .par_iter()
-            .map(|(i, input)| {
-                let func_index = module.func_index(*i);
-                let mut context = Context::new();
-                context.func.name = get_func_name(func_index);
-                context.func.signature = module.signatures[module.functions[func_index]].clone();
+        let data = match cache_entry.get_data() {
+            Some(data) => data,
+            None => {
+                let mut functions = PrimaryMap::with_capacity(function_body_inputs.len());
+                let mut relocations = PrimaryMap::with_capacity(function_body_inputs.len());
+                let mut address_transforms = PrimaryMap::with_capacity(function_body_inputs.len());
 
-                let mut trans = FuncTranslator::new();
-                trans
-                    .translate(
-                        input.data,
-                        input.module_offset,
-                        &mut context.func,
-                        &mut FuncEnvironment::new(isa.frontend_config(), module),
-                    )
-                    .map_err(CompileError::Wasm)?;
+                function_body_inputs
+                    .into_iter()
+                    .collect::<Vec<(DefinedFuncIndex, &FunctionBodyData<'data>)>>()
+                    .par_iter()
+                    .map(|(i, input)| {
+                        let func_index = module.func_index(*i);
+                        let mut context = Context::new();
+                        context.func.name = get_func_name(func_index);
+                        context.func.signature =
+                            module.signatures[module.functions[func_index]].clone();
 
-                let mut code_buf: Vec<u8> = Vec::new();
-                let mut reloc_sink = RelocSink::new(func_index);
-                let mut trap_sink = binemit::NullTrapSink {};
-                context
-                    .compile_and_emit(isa, &mut code_buf, &mut reloc_sink, &mut trap_sink)
-                    .map_err(CompileError::Codegen)?;
+                        let mut trans = FuncTranslator::new();
+                        trans
+                            .translate(
+                                input.data,
+                                input.module_offset,
+                                &mut context.func,
+                                &mut FuncEnvironment::new(isa.frontend_config(), module),
+                            )
+                            .map_err(CompileError::Wasm)?;
 
-                let jt_offsets = context.func.jt_offsets.clone();
+                        let mut code_buf: Vec<u8> = Vec::new();
+                        let mut reloc_sink = RelocSink::new(func_index);
+                        let mut trap_sink = binemit::NullTrapSink {};
+                        context
+                            .compile_and_emit(isa, &mut code_buf, &mut reloc_sink, &mut trap_sink)
+                            .map_err(CompileError::Codegen)?;
 
-                let address_transform = if generate_debug_info {
-                    let body_len = code_buf.len();
-                    let at = get_address_transform(&context, isa);
+                        let jt_offsets = context.func.jt_offsets.clone();
 
-                    Some(FunctionAddressMap {
-                        instructions: at,
-                        body_offset: 0,
-                        body_len,
+                        let address_transform = if generate_debug_info {
+                            let body_len = code_buf.len();
+                            let at = get_address_transform(&context, isa);
+
+                            Some(FunctionAddressMap {
+                                instructions: at,
+                                body_offset: 0,
+                                body_len,
+                            })
+                        } else {
+                            None
+                        };
+
+                        Ok((
+                            code_buf,
+                            jt_offsets,
+                            reloc_sink.func_relocs,
+                            address_transform,
+                        ))
                     })
-                } else {
-                    None
-                };
+                    .collect::<Result<Vec<_>, CompileError>>()?
+                    .into_iter()
+                    .for_each(|(function, func_jt_offsets, relocs, address_transform)| {
+                        functions.push(CodeAndJTOffsets {
+                            body: function,
+                            jt_offsets: func_jt_offsets,
+                        });
+                        relocations.push(relocs);
+                        if let Some(address_transform) = address_transform {
+                            address_transforms.push(address_transform);
+                        }
+                    });
 
-                Ok((
-                    code_buf,
-                    jt_offsets,
-                    reloc_sink.func_relocs,
-                    address_transform,
-                ))
-            })
-            .collect::<Result<Vec<_>, CompileError>>()?
-            .into_iter()
-            .for_each(|(function, func_jt_offsets, relocs, address_transform)| {
-                functions.push(CodeAndJTOffsets {
-                    body: function,
-                    jt_offsets: func_jt_offsets,
-                });
-                relocations.push(relocs);
-                if let Some(address_transform) = address_transform {
-                    address_transforms.push(address_transform);
-                }
-            });
+                // TODO: Reorganize where we create the Vec for the resolved imports.
 
-        // TODO: Reorganize where we create the Vec for the resolved imports.
-        Ok((Compilation::new(functions), relocations, address_transforms))
+                let data = ModuleCacheData::from_tuple((
+                    Compilation::new(functions),
+                    relocations,
+                    address_transforms,
+                ));
+                cache_entry.update_data(&data);
+                data
+            }
+        };
+
+        Ok(data.to_tuple())
     }
 }

wasmtime-environ/src/lib.rs

@@ -45,11 +45,14 @@ mod module_environ;
 mod tunables;
 mod vmoffsets;
 
+mod cache;
+
 pub mod cranelift;
 #[cfg(feature = "lightbeam")]
 pub mod lightbeam;
 
 pub use crate::address_map::{FunctionAddressMap, InstructionAddressMap, ModuleAddressMap};
+pub use crate::cache::conf as cache_conf;
 pub use crate::compilation::{
     Compilation, CompileError, Compiler, Relocation, RelocationTarget, Relocations,
 };

wasmtime-environ/src/module.rs

@@ -170,6 +170,10 @@ pub struct Module {
 
     /// WebAssembly table initializers.
     pub table_elements: Vec<TableElements>,
+
+    /// Hash of the source wasm code if this module is not synthesized.
+    /// TODO: this is temporary workaround. Will be replaced with derive macro.
+    pub hash: Option<[u8; 32]>,
 }
 
 impl Module {
@@ -188,6 +192,7 @@ impl Module {
             exports: IndexMap::new(),
             start_func: None,
             table_elements: Vec::new(),
+            hash: None,
         }
     }
 

wasmtime-environ/src/module_environ.rs

@@ -10,6 +10,7 @@ use cranelift_wasm::{
     self, translate_module, DefinedFuncIndex, FuncIndex, Global, GlobalIndex, Memory, MemoryIndex,
     SignatureIndex, Table, TableIndex, WasmResult,
 };
+use sha2::{Digest, Sha256};
 use std::boxed::Box;
 use std::string::String;
 use std::vec::Vec;
@@ -79,6 +80,11 @@ impl<'data> ModuleEnvironment<'data> {
     pub fn translate(mut self, data: &'data [u8]) -> WasmResult<ModuleTranslation<'data>> {
         translate_module(data, &mut self)?;
 
+        // TODO: this is temporary workaround and will be replaced with derive macro.
+        let mut hasher = Sha256::new();
+        hasher.input(data);
+        self.result.module.hash = Some(hasher.result().into());
+
         Ok(self.result)
     }
 }