@@ -7,10 +7,11 @@ use std::ptr;
77#[ cfg( test) ]
88mod tests;
99
10+ const ELEM_SIZE : usize = mem:: size_of :: < u64 > ( ) ;
1011const BUFFER_SIZE_ELEMS : usize = 8 ;
11- const BUFFER_SIZE_BYTES : usize = BUFFER_SIZE_ELEMS * mem :: size_of :: < u64 > ( ) ;
12+ const BUFFER_SIZE_BYTES : usize = BUFFER_SIZE_ELEMS * ELEM_SIZE ;
1213const BUFFER_SIZE_ELEMS_SPILL : usize = BUFFER_SIZE_ELEMS + 1 ;
13- const BUFFER_SIZE_BYTES_SPILL : usize = BUFFER_SIZE_ELEMS_SPILL * mem :: size_of :: < u64 > ( ) ;
14+ const BUFFER_SIZE_BYTES_SPILL : usize = BUFFER_SIZE_ELEMS_SPILL * ELEM_SIZE ;
1415const BUFFER_SPILL_INDEX : usize = BUFFER_SIZE_ELEMS_SPILL - 1 ;
1516
1617#[ derive( Debug , Clone ) ]
@@ -54,15 +55,16 @@ macro_rules! compress {
5455 } } ;
5556}
5657
57- // Copies up to 8 bytes from source to destination. This may be faster than
58- // calling `ptr::copy_nonoverlapping` with an arbitrary count, since all of
59- // the copies have fixed sizes and thus avoid calling memcpy.
58+ // Copies up to 8 bytes from source to destination. This performs better than
59+ // `ptr::copy_nonoverlapping` on microbenchmarks and may perform better on real
60+ // workloads since all of the copies have fixed sizes and avoid calling memcpy.
6061#[ inline]
6162unsafe fn copy_nonoverlapping_small ( src : * const u8 , dst : * mut u8 , count : usize ) {
62- debug_assert ! ( count <= 8 ) ;
63+ const COUNT_MAX : usize = 8 ;
64+ debug_assert ! ( count <= COUNT_MAX ) ;
6365
64- if count == 8 {
65- ptr:: copy_nonoverlapping ( src, dst, 8 ) ;
66+ if count == COUNT_MAX {
67+ ptr:: copy_nonoverlapping ( src, dst, COUNT_MAX ) ;
6668 return ;
6769 }
6870
@@ -85,7 +87,7 @@ unsafe fn copy_nonoverlapping_small(src: *const u8, dst: *mut u8, count: usize)
8587 debug_assert_eq ! ( i, count) ;
8688}
8789
88- // Implementation
90+ // # Implementation
8991//
9092// This implementation uses buffering to reduce the hashing cost for inputs
9193// consisting of many small integers. Buffering simplifies the integration of
@@ -99,10 +101,11 @@ unsafe fn copy_nonoverlapping_small(src: *const u8, dst: *mut u8, count: usize)
99101//
100102// When a write fills the buffer, a buffer processing function is invoked to
101103// hash all of the buffered input. The buffer processing functions are marked
102- // #[inline(never)] so that they aren't inlined into the append functions, which
103- // ensures the more frequently called append functions remain inlineable and
104- // don't include register pushing/popping that would only be made necessary by
105- // inclusion of the complex buffer processing path which uses those registers.
104+ // `#[inline(never)]` so that they aren't inlined into the append functions,
105+ // which ensures the more frequently called append functions remain inlineable
106+ // and don't include register pushing/popping that would only be made necessary
107+ // by inclusion of the complex buffer processing path which uses those
108+ // registers.
106109//
107110// The buffer includes a "spill"--an extra element at the end--which simplifies
108111// the integer write buffer processing path. The value that fills the buffer can
@@ -118,7 +121,7 @@ unsafe fn copy_nonoverlapping_small(src: *const u8, dst: *mut u8, count: usize)
118121// efficiently implemented with an uninitialized buffer. On the other hand, an
119122// uninitialized buffer may become more important should a larger one be used.
120123//
121- // Platform Dependence
124+ // # Platform Dependence
122125//
123126// The SipHash algorithm operates on byte sequences. It parses the input stream
124127// as 8-byte little-endian integers. Therefore, given the same byte sequence, it
@@ -131,14 +134,14 @@ unsafe fn copy_nonoverlapping_small(src: *const u8, dst: *mut u8, count: usize)
131134// native size), or independent (by converting to a common size), supposing the
132135// values can be represented in 32 bits.
133136//
134- // In order to make SipHasher128 consistent with SipHasher in libstd, we choose
135- // to do the integer to byte sequence conversion in the platform-dependent way.
136- // Clients can achieve (nearly) platform-independent hashing by widening `isize`
137- // and `usize` integers to 64 bits on 32-bit systems and byte-swapping integers
138- // on big-endian systems before passing them to the writing functions. This
139- // causes the input byte sequence to look identical on big- and little- endian
140- // systems (supposing `isize` and `usize` values can be represented in 32 bits),
141- // which ensures platform-independent results.
137+ // In order to make ` SipHasher128` consistent with ` SipHasher` in libstd, we
138+ // choose to do the integer to byte sequence conversion in the platform-
139+ // dependent way. Clients can achieve (nearly) platform-independent hashing by
140+ // widening `isize` and `usize` integers to 64 bits on 32-bit systems and
141+ // byte-swapping integers on big-endian systems before passing them to the
142+ // writing functions. This causes the input byte sequence to look identical on
143+ // big- and little- endian systems (supposing `isize` and `usize` values can be
144+ // represented in 32 bits), which ensures platform-independent results.
142145impl SipHasher128 {
143146 #[ inline]
144147 pub fn new_with_keys ( key0 : u64 , key1 : u64 ) -> SipHasher128 {
@@ -156,7 +159,7 @@ impl SipHasher128 {
156159 } ;
157160
158161 unsafe {
159- // Initialize spill because we read from it in short_write_process_buffer.
162+ // Initialize spill because we read from it in ` short_write_process_buffer` .
160163 * hasher. buf . get_unchecked_mut ( BUFFER_SPILL_INDEX ) = MaybeUninit :: zeroed ( ) ;
161164 }
162165
@@ -190,9 +193,9 @@ impl SipHasher128 {
190193 // A specialized write function for values with size <= 8 that should only
191194 // be called when the write would cause the buffer to fill.
192195 //
193- // SAFETY: the write of x into self.buf starting at byte offset self.nbuf
194- // must cause self.buf to become fully initialized (and not overflow) if it
195- // wasn't already.
196+ // SAFETY: the write of `x` into ` self.buf` starting at byte offset
197+ // `self.nbuf` must cause ` self.buf` to become fully initialized (and not
198+ // overflow) if it wasn't already.
196199 #[ inline( never) ]
197200 unsafe fn short_write_process_buffer < T > ( & mut self , x : T ) {
198201 let size = mem:: size_of :: < T > ( ) ;
@@ -223,7 +226,7 @@ impl SipHasher128 {
223226 ptr:: copy_nonoverlapping ( src, self . buf . as_mut_ptr ( ) as * mut u8 , size - 1 ) ;
224227
225228 // This function should only be called when the write fills the buffer.
226- // Therefore, when size == 1, the new self.nbuf must be zero. The size
229+ // Therefore, when size == 1, the new ` self.nbuf` must be zero. The size
227230 // is statically known, so the branch is optimized away.
228231 self . nbuf = if size == 1 { 0 } else { nbuf + size - BUFFER_SIZE_BYTES } ;
229232 self . processed += BUFFER_SIZE_BYTES ;
@@ -240,7 +243,7 @@ impl SipHasher128 {
240243 unsafe {
241244 let dst = ( self . buf . as_mut_ptr ( ) as * mut u8 ) . add ( nbuf) ;
242245
243- if length < 8 {
246+ if length <= 8 {
244247 copy_nonoverlapping_small ( msg. as_ptr ( ) , dst, length) ;
245248 } else {
246249 // This memcpy is *not* optimized away.
@@ -259,9 +262,9 @@ impl SipHasher128 {
259262 // A write function for byte slices that should only be called when the
260263 // write would cause the buffer to fill.
261264 //
262- // SAFETY: self.buf must be initialized up to the byte offset self.nbuf, and
263- // msg must contain enough bytes to initialize the rest of the element
264- // containing the byte offset self.nbuf.
265+ // SAFETY: ` self.buf` must be initialized up to the byte offset ` self.nbuf`,
266+ // and ` msg` must contain enough bytes to initialize the rest of the element
267+ // containing the byte offset ` self.nbuf` .
265268 #[ inline( never) ]
266269 unsafe fn slice_write_process_buffer ( & mut self , msg : & [ u8 ] ) {
267270 let length = msg. len ( ) ;
@@ -272,19 +275,20 @@ impl SipHasher128 {
272275 // Always copy first part of input into current element of buffer.
273276 // This function should only be called when the write fills the buffer,
274277 // so we know that there is enough input to fill the current element.
275- let valid_in_elem = nbuf & 0x7 ;
276- let needed_in_elem = 8 - valid_in_elem;
278+ let valid_in_elem = nbuf % ELEM_SIZE ;
279+ let needed_in_elem = ELEM_SIZE - valid_in_elem;
277280
278281 let src = msg. as_ptr ( ) ;
279282 let dst = ( self . buf . as_mut_ptr ( ) as * mut u8 ) . add ( nbuf) ;
280283 copy_nonoverlapping_small ( src, dst, needed_in_elem) ;
281284
282285 // Process buffer.
283286
284- // Using nbuf / 8 + 1 rather than (nbuf + needed_in_elem) / 8 to show
285- // the compiler that this loop's upper bound is > 0. We know that is
286- // true, because last step ensured we have a full element in the buffer.
287- let last = nbuf / 8 + 1 ;
287+ // Using `nbuf / ELEM_SIZE + 1` rather than `(nbuf + needed_in_elem) /
288+ // ELEM_SIZE` to show the compiler that this loop's upper bound is > 0.
289+ // We know that is true, because last step ensured we have a full
290+ // element in the buffer.
291+ let last = nbuf / ELEM_SIZE + 1 ;
288292
289293 for i in 0 ..last {
290294 let elem = self . buf . get_unchecked ( i) . assume_init ( ) . to_le ( ) ;
@@ -293,26 +297,26 @@ impl SipHasher128 {
293297 self . state . v0 ^= elem;
294298 }
295299
296- // Process the remaining u64 -sized chunks of input.
300+ // Process the remaining element -sized chunks of input.
297301 let mut processed = needed_in_elem;
298302 let input_left = length - processed;
299- let u64s_left = input_left / 8 ;
300- let u8s_left = input_left & 0x7 ;
303+ let elems_left = input_left / ELEM_SIZE ;
304+ let extra_bytes_left = input_left % ELEM_SIZE ;
301305
302- for _ in 0 ..u64s_left {
306+ for _ in 0 ..elems_left {
303307 let elem = ( msg. as_ptr ( ) . add ( processed) as * const u64 ) . read_unaligned ( ) . to_le ( ) ;
304308 self . state . v3 ^= elem;
305309 Sip24Rounds :: c_rounds ( & mut self . state ) ;
306310 self . state . v0 ^= elem;
307- processed += 8 ;
311+ processed += ELEM_SIZE ;
308312 }
309313
310314 // Copy remaining input into start of buffer.
311315 let src = msg. as_ptr ( ) . add ( processed) ;
312316 let dst = self . buf . as_mut_ptr ( ) as * mut u8 ;
313- copy_nonoverlapping_small ( src, dst, u8s_left ) ;
317+ copy_nonoverlapping_small ( src, dst, extra_bytes_left ) ;
314318
315- self . nbuf = u8s_left ;
319+ self . nbuf = extra_bytes_left ;
316320 self . processed += nbuf + processed;
317321 }
318322
@@ -321,7 +325,7 @@ impl SipHasher128 {
321325 debug_assert ! ( self . nbuf < BUFFER_SIZE_BYTES ) ;
322326
323327 // Process full elements in buffer.
324- let last = self . nbuf / 8 ;
328+ let last = self . nbuf / ELEM_SIZE ;
325329
326330 // Since we're consuming self, avoid updating members for a potential
327331 // performance gain.
@@ -335,14 +339,14 @@ impl SipHasher128 {
335339 }
336340
337341 // Get remaining partial element.
338- let elem = if self . nbuf % 8 != 0 {
342+ let elem = if self . nbuf % ELEM_SIZE != 0 {
339343 unsafe {
340344 // Ensure element is initialized by writing zero bytes. At most
341- // seven are required given the above check. It's safe to write
342- // this many because we have the spill element and we maintain
343- // self.nbuf such that this write will start before the spill.
345+ // `ELEM_SIZE - 1` are required given the above check. It's safe
346+ // to write this many because we have the spill and we maintain
347+ // ` self.nbuf` such that this write will start before the spill.
344348 let dst = ( self . buf . as_mut_ptr ( ) as * mut u8 ) . add ( self . nbuf ) ;
345- ptr:: write_bytes ( dst, 0 , 7 ) ;
349+ ptr:: write_bytes ( dst, 0 , ELEM_SIZE - 1 ) ;
346350 self . buf . get_unchecked ( last) . assume_init ( ) . to_le ( )
347351 }
348352 } else {
0 commit comments