diff options
Diffstat (limited to 'compiler/rustc_serialize/src/opaque.rs')
| -rw-r--r-- | compiler/rustc_serialize/src/opaque.rs | 318 |
1 files changed, 62 insertions, 256 deletions
diff --git a/compiler/rustc_serialize/src/opaque.rs b/compiler/rustc_serialize/src/opaque.rs index 0f6e4b329b8..0ffc537eee0 100644 --- a/compiler/rustc_serialize/src/opaque.rs +++ b/compiler/rustc_serialize/src/opaque.rs @@ -1,4 +1,4 @@ -use crate::leb128::{self, largest_max_leb128_len}; +use crate::leb128; use crate::serialize::{Decodable, Decoder, Encodable, Encoder}; use std::fs::File; use std::io::{self, Write}; @@ -12,118 +12,17 @@ use std::ptr; // Encoder // ----------------------------------------------------------------------------- -pub struct MemEncoder { - pub data: Vec<u8>, -} - -impl MemEncoder { - pub fn new() -> MemEncoder { - MemEncoder { data: vec![] } - } - - #[inline] - pub fn position(&self) -> usize { - self.data.len() - } - - pub fn finish(self) -> Vec<u8> { - self.data - } -} - -macro_rules! write_leb128 { - ($enc:expr, $value:expr, $int_ty:ty, $fun:ident) => {{ - const MAX_ENCODED_LEN: usize = $crate::leb128::max_leb128_len::<$int_ty>(); - let old_len = $enc.data.len(); - - if MAX_ENCODED_LEN > $enc.data.capacity() - old_len { - $enc.data.reserve(MAX_ENCODED_LEN); - } - - // SAFETY: The above check and `reserve` ensures that there is enough - // room to write the encoded value to the vector's internal buffer. - unsafe { - let buf = &mut *($enc.data.as_mut_ptr().add(old_len) - as *mut [MaybeUninit<u8>; MAX_ENCODED_LEN]); - let encoded = leb128::$fun(buf, $value); - $enc.data.set_len(old_len + encoded.len()); - } - }}; -} - -impl Encoder for MemEncoder { - #[inline] - fn emit_usize(&mut self, v: usize) { - write_leb128!(self, v, usize, write_usize_leb128) - } - - #[inline] - fn emit_u128(&mut self, v: u128) { - write_leb128!(self, v, u128, write_u128_leb128); - } - - #[inline] - fn emit_u64(&mut self, v: u64) { - write_leb128!(self, v, u64, write_u64_leb128); - } - - #[inline] - fn emit_u32(&mut self, v: u32) { - write_leb128!(self, v, u32, write_u32_leb128); - } - - #[inline] - fn emit_u16(&mut self, v: u16) { - self.data.extend_from_slice(&v.to_le_bytes()); - } - - #[inline] - fn emit_u8(&mut self, v: u8) { - self.data.push(v); - } - - #[inline] - fn emit_isize(&mut self, v: isize) { - write_leb128!(self, v, isize, write_isize_leb128) - } - - #[inline] - fn emit_i128(&mut self, v: i128) { - write_leb128!(self, v, i128, write_i128_leb128) - } - - #[inline] - fn emit_i64(&mut self, v: i64) { - write_leb128!(self, v, i64, write_i64_leb128) - } - - #[inline] - fn emit_i32(&mut self, v: i32) { - write_leb128!(self, v, i32, write_i32_leb128) - } - - #[inline] - fn emit_i16(&mut self, v: i16) { - self.data.extend_from_slice(&v.to_le_bytes()); - } - - #[inline] - fn emit_raw_bytes(&mut self, s: &[u8]) { - self.data.extend_from_slice(s); - } -} - pub type FileEncodeResult = Result<usize, io::Error>; +/// The size of the buffer in `FileEncoder`. +const BUF_SIZE: usize = 8192; + /// `FileEncoder` encodes data to file via fixed-size buffer. /// -/// When encoding large amounts of data to a file, using `FileEncoder` may be -/// preferred over using `MemEncoder` to encode to a `Vec`, and then writing the -/// `Vec` to file, as the latter uses as much memory as there is encoded data, -/// while the former uses the fixed amount of memory allocated to the buffer. -/// `FileEncoder` also has the advantage of not needing to reallocate as data -/// is appended to it, but the disadvantage of requiring more error handling, -/// which has some runtime overhead. +/// There used to be a `MemEncoder` type that encoded all the data into a +/// `Vec`. `FileEncoder` is better because its memory use is determined by the +/// size of the buffer, rather than the full length of the encoded data, and +/// because it doesn't need to reallocate memory along the way. pub struct FileEncoder { /// The input buffer. For adequate performance, we need more control over /// buffering than `BufWriter` offers. If `BufWriter` ever offers a raw @@ -139,26 +38,12 @@ pub struct FileEncoder { impl FileEncoder { pub fn new<P: AsRef<Path>>(path: P) -> io::Result<Self> { - const DEFAULT_BUF_SIZE: usize = 8192; - FileEncoder::with_capacity(path, DEFAULT_BUF_SIZE) - } - - pub fn with_capacity<P: AsRef<Path>>(path: P, capacity: usize) -> io::Result<Self> { - // Require capacity at least as large as the largest LEB128 encoding - // here, so that we don't have to check or handle this on every write. - assert!(capacity >= largest_max_leb128_len()); - - // Require capacity small enough such that some capacity checks can be - // done using guaranteed non-overflowing add rather than sub, which - // shaves an instruction off those code paths (on x86 at least). - assert!(capacity <= usize::MAX - largest_max_leb128_len()); - // Create the file for reading and writing, because some encoders do both // (e.g. the metadata encoder when -Zmeta-stats is enabled) let file = File::options().read(true).write(true).create(true).truncate(true).open(path)?; Ok(FileEncoder { - buf: Box::new_uninit_slice(capacity), + buf: Box::new_uninit_slice(BUF_SIZE), buffered: 0, flushed: 0, file, @@ -264,18 +149,10 @@ impl FileEncoder { } #[inline] - fn capacity(&self) -> usize { - self.buf.len() - } - - #[inline] fn write_one(&mut self, value: u8) { - // We ensure this during `FileEncoder` construction. - debug_assert!(self.capacity() >= 1); - let mut buffered = self.buffered; - if std::intrinsics::unlikely(buffered >= self.capacity()) { + if std::intrinsics::unlikely(buffered + 1 > BUF_SIZE) { self.flush(); buffered = 0; } @@ -291,13 +168,12 @@ impl FileEncoder { #[inline] fn write_all(&mut self, buf: &[u8]) { - let capacity = self.capacity(); let buf_len = buf.len(); - if std::intrinsics::likely(buf_len <= capacity) { + if std::intrinsics::likely(buf_len <= BUF_SIZE) { let mut buffered = self.buffered; - if std::intrinsics::unlikely(buf_len > capacity - buffered) { + if std::intrinsics::unlikely(buffered + buf_len > BUF_SIZE) { self.flush(); buffered = 0; } @@ -369,52 +245,39 @@ impl Drop for FileEncoder { } } -macro_rules! file_encoder_write_leb128 { - ($enc:expr, $value:expr, $int_ty:ty, $fun:ident) => {{ - const MAX_ENCODED_LEN: usize = $crate::leb128::max_leb128_len::<$int_ty>(); - - // We ensure this during `FileEncoder` construction. - debug_assert!($enc.capacity() >= MAX_ENCODED_LEN); +macro_rules! write_leb128 { + ($this_fn:ident, $int_ty:ty, $write_leb_fn:ident) => { + #[inline] + fn $this_fn(&mut self, v: $int_ty) { + const MAX_ENCODED_LEN: usize = $crate::leb128::max_leb128_len::<$int_ty>(); - let mut buffered = $enc.buffered; + let mut buffered = self.buffered; - // This can't overflow. See assertion in `FileEncoder::with_capacity`. - if std::intrinsics::unlikely(buffered + MAX_ENCODED_LEN > $enc.capacity()) { - $enc.flush(); - buffered = 0; - } + // This can't overflow because BUF_SIZE and MAX_ENCODED_LEN are both + // quite small. + if std::intrinsics::unlikely(buffered + MAX_ENCODED_LEN > BUF_SIZE) { + self.flush(); + buffered = 0; + } - // SAFETY: The above check and flush ensures that there is enough - // room to write the encoded value to the buffer. - let buf = unsafe { - &mut *($enc.buf.as_mut_ptr().add(buffered) as *mut [MaybeUninit<u8>; MAX_ENCODED_LEN]) - }; + // SAFETY: The above check and flush ensures that there is enough + // room to write the encoded value to the buffer. + let buf = unsafe { + &mut *(self.buf.as_mut_ptr().add(buffered) + as *mut [MaybeUninit<u8>; MAX_ENCODED_LEN]) + }; - let encoded = leb128::$fun(buf, $value); - $enc.buffered = buffered + encoded.len(); - }}; + let encoded = leb128::$write_leb_fn(buf, v); + self.buffered = buffered + encoded.len(); + } + }; } impl Encoder for FileEncoder { - #[inline] - fn emit_usize(&mut self, v: usize) { - file_encoder_write_leb128!(self, v, usize, write_usize_leb128) - } - - #[inline] - fn emit_u128(&mut self, v: u128) { - file_encoder_write_leb128!(self, v, u128, write_u128_leb128) - } - - #[inline] - fn emit_u64(&mut self, v: u64) { - file_encoder_write_leb128!(self, v, u64, write_u64_leb128) - } - - #[inline] - fn emit_u32(&mut self, v: u32) { - file_encoder_write_leb128!(self, v, u32, write_u32_leb128) - } + write_leb128!(emit_usize, usize, write_usize_leb128); + write_leb128!(emit_u128, u128, write_u128_leb128); + write_leb128!(emit_u64, u64, write_u64_leb128); + write_leb128!(emit_u32, u32, write_u32_leb128); #[inline] fn emit_u16(&mut self, v: u16) { @@ -426,25 +289,10 @@ impl Encoder for FileEncoder { self.write_one(v); } - #[inline] - fn emit_isize(&mut self, v: isize) { - file_encoder_write_leb128!(self, v, isize, write_isize_leb128) - } - - #[inline] - fn emit_i128(&mut self, v: i128) { - file_encoder_write_leb128!(self, v, i128, write_i128_leb128) - } - - #[inline] - fn emit_i64(&mut self, v: i64) { - file_encoder_write_leb128!(self, v, i64, write_i64_leb128) - } - - #[inline] - fn emit_i32(&mut self, v: i32) { - file_encoder_write_leb128!(self, v, i32, write_i32_leb128) - } + write_leb128!(emit_isize, isize, write_isize_leb128); + write_leb128!(emit_i128, i128, write_i128_leb128); + write_leb128!(emit_i64, i64, write_i64_leb128); + write_leb128!(emit_i32, i32, write_i32_leb128); #[inline] fn emit_i16(&mut self, v: i16) { @@ -541,30 +389,19 @@ impl<'a> MemDecoder<'a> { } macro_rules! read_leb128 { - ($dec:expr, $fun:ident) => {{ leb128::$fun($dec) }}; + ($this_fn:ident, $int_ty:ty, $read_leb_fn:ident) => { + #[inline] + fn $this_fn(&mut self) -> $int_ty { + leb128::$read_leb_fn(self) + } + }; } impl<'a> Decoder for MemDecoder<'a> { - #[inline] - fn position(&self) -> usize { - // SAFETY: This type guarantees start <= current - unsafe { self.current.sub_ptr(self.start) } - } - - #[inline] - fn read_u128(&mut self) -> u128 { - read_leb128!(self, read_u128_leb128) - } - - #[inline] - fn read_u64(&mut self) -> u64 { - read_leb128!(self, read_u64_leb128) - } - - #[inline] - fn read_u32(&mut self) -> u32 { - read_leb128!(self, read_u32_leb128) - } + read_leb128!(read_usize, usize, read_usize_leb128); + read_leb128!(read_u128, u128, read_u128_leb128); + read_leb128!(read_u64, u64, read_u64_leb128); + read_leb128!(read_u32, u32, read_u32_leb128); #[inline] fn read_u16(&mut self) -> u16 { @@ -584,25 +421,10 @@ impl<'a> Decoder for MemDecoder<'a> { } } - #[inline] - fn read_usize(&mut self) -> usize { - read_leb128!(self, read_usize_leb128) - } - - #[inline] - fn read_i128(&mut self) -> i128 { - read_leb128!(self, read_i128_leb128) - } - - #[inline] - fn read_i64(&mut self) -> i64 { - read_leb128!(self, read_i64_leb128) - } - - #[inline] - fn read_i32(&mut self) -> i32 { - read_leb128!(self, read_i32_leb128) - } + read_leb128!(read_isize, isize, read_isize_leb128); + read_leb128!(read_i128, i128, read_i128_leb128); + read_leb128!(read_i64, i64, read_i64_leb128); + read_leb128!(read_i32, i32, read_i32_leb128); #[inline] fn read_i16(&mut self) -> i16 { @@ -610,11 +432,6 @@ impl<'a> Decoder for MemDecoder<'a> { } #[inline] - fn read_isize(&mut self) -> isize { - read_leb128!(self, read_isize_leb128) - } - - #[inline] fn read_raw_bytes(&mut self, bytes: usize) -> &'a [u8] { if bytes > self.remaining() { Self::decoder_exhausted(); @@ -636,6 +453,12 @@ impl<'a> Decoder for MemDecoder<'a> { // Since we just checked current == end, the current pointer must be inbounds. unsafe { *self.current } } + + #[inline] + fn position(&self) -> usize { + // SAFETY: This type guarantees start <= current + unsafe { self.current.sub_ptr(self.start) } + } } // Specializations for contiguous byte sequences follow. The default implementations for slices @@ -645,13 +468,6 @@ impl<'a> Decoder for MemDecoder<'a> { // Specialize encoding byte slices. This specialization also applies to encoding `Vec<u8>`s, etc., // since the default implementations call `encode` on their slices internally. -impl Encodable<MemEncoder> for [u8] { - fn encode(&self, e: &mut MemEncoder) { - Encoder::emit_usize(e, self.len()); - e.emit_raw_bytes(self); - } -} - impl Encodable<FileEncoder> for [u8] { fn encode(&self, e: &mut FileEncoder) { Encoder::emit_usize(e, self.len()); @@ -675,16 +491,6 @@ impl IntEncodedWithFixedSize { pub const ENCODED_SIZE: usize = 8; } -impl Encodable<MemEncoder> for IntEncodedWithFixedSize { - #[inline] - fn encode(&self, e: &mut MemEncoder) { - let _start_pos = e.position(); - e.emit_raw_bytes(&self.0.to_le_bytes()); - let _end_pos = e.position(); - debug_assert_eq!((_end_pos - _start_pos), IntEncodedWithFixedSize::ENCODED_SIZE); - } -} - impl Encodable<FileEncoder> for IntEncodedWithFixedSize { #[inline] fn encode(&self, e: &mut FileEncoder) { |