go/src/runtime/slice.go
Carl Mastrangelo c1e267cc73 runtime: make append only clear uncopied memory
Also add a benchmark that shows off the new behavior.  The
existing benchmarks reuse the same slice, and thus don't ever have
to clear memory.  Running the Append|Grow benchmarks in runtime:

name                              old time/op  new time/op  delta
AppendSliceLarge/1024Bytes-12      265ns ± 1%   265ns ± 3%     ~     (p=0.524 n=17+20)
AppendSliceLarge/4096Bytes-12      807ns ± 3%   772ns ± 1%   -4.38%  (p=0.000 n=20+20)
AppendSliceLarge/16384Bytes-12    3.20µs ± 4%  2.82µs ± 4%  -11.93%  (p=0.000 n=19+20)
AppendSliceLarge/65536Bytes-12    13.0µs ± 4%  11.0µs ± 3%  -15.22%  (p=0.000 n=20+20)
AppendSliceLarge/262144Bytes-12   62.7µs ± 1%  51.6µs ± 1%  -17.67%  (p=0.000 n=19+20)
AppendSliceLarge/1048576Bytes-12   337µs ± 3%   289µs ± 3%  -14.36%  (p=0.000 n=20+20)
GrowSliceBytes-12                 31.2ns ± 4%  31.4ns ±11%     ~     (p=0.308 n=19+18)
GrowSliceInts-12                  53.4ns ±14%  45.0ns ± 6%  -15.74%  (p=0.000 n=20+19)
GrowSlicePtr-12                   87.0ns ± 3%  83.3ns ± 3%   -4.26%  (p=0.000 n=18+17)
GrowSliceStruct24Bytes-12         88.9ns ± 5%  77.8ns ± 2%  -12.45%  (p=0.000 n=20+19)
Append-12                         17.2ns ± 1%  17.3ns ± 2%     ~     (p=0.464 n=18+17)
AppendGrowByte-12                 2.28ms ± 1%  1.92ms ± 2%  -15.65%  (p=0.000 n=20+18)
AppendGrowString-12                255ms ± 3%   253ms ± 4%     ~     (p=0.065 n=19+19)
AppendSlice/1Bytes-12             3.13ns ± 0%  3.11ns ± 1%   -0.65%  (p=0.000 n=17+18)
AppendSlice/4Bytes-12             3.02ns ± 2%  3.11ns ± 1%   +3.27%  (p=0.000 n=18+17)
AppendSlice/7Bytes-12             4.14ns ± 3%  4.13ns ± 2%     ~     (p=0.380 n=19+18)
AppendSlice/8Bytes-12             3.74ns ± 3%  3.68ns ± 1%   -1.76%  (p=0.000 n=19+18)
AppendSlice/15Bytes-12            4.03ns ± 2%  4.04ns ± 2%     ~     (p=0.261 n=19+20)
AppendSlice/16Bytes-12            4.03ns ± 2%  4.03ns ± 0%     ~     (p=0.062 n=18+17)
AppendSlice/32Bytes-12            3.23ns ± 4%  3.43ns ± 1%   +6.10%  (p=0.000 n=17+18)
AppendStr/1Bytes-12               3.51ns ± 1%  3.52ns ± 1%     ~     (p=0.321 n=18+19)
AppendStr/4Bytes-12               3.46ns ± 1%  3.46ns ± 1%     ~     (p=0.977 n=18+20)
AppendStr/8Bytes-12               3.18ns ± 1%  3.19ns ± 1%     ~     (p=0.650 n=16+17)
AppendStr/16Bytes-12              6.08ns ±27%  5.52ns ± 3%   -9.16%  (p=0.002 n=18+19)
AppendStr/32Bytes-12              3.71ns ± 1%  3.53ns ± 1%   -4.73%  (p=0.000 n=20+19)
AppendSpecialCase-12              17.7ns ± 1%  17.8ns ± 3%   +0.86%  (p=0.045 n=17+18)
AppendInPlace/NoGrow/Byte-12       375ns ± 1%   376ns ± 1%   +0.35%  (p=0.021 n=20+18)
AppendInPlace/NoGrow/1Ptr-12      1.01µs ± 1%  1.10µs ± 1%   +9.28%  (p=0.000 n=18+20)
AppendInPlace/NoGrow/2Ptr-12      1.85µs ± 2%  1.71µs ± 1%   -7.51%  (p=0.000 n=19+18)
AppendInPlace/NoGrow/3Ptr-12      2.57µs ± 2%  2.44µs ± 1%   -5.08%  (p=0.000 n=19+19)
AppendInPlace/NoGrow/4Ptr-12      3.52µs ± 2%  3.35µs ± 2%   -4.70%  (p=0.000 n=20+19)
AppendInPlace/Grow/Byte-12         212ns ± 1%   217ns ± 8%   +2.57%  (p=0.000 n=20+20)
AppendInPlace/Grow/1Ptr-12         214ns ± 2%   217ns ± 3%   +1.23%  (p=0.001 n=18+19)
AppendInPlace/Grow/2Ptr-12         298ns ± 2%   300ns ± 2%   +0.55%  (p=0.038 n=19+20)
AppendInPlace/Grow/3Ptr-12         367ns ± 2%   366ns ± 2%     ~     (p=0.452 n=20+18)
AppendInPlace/Grow/4Ptr-12         416ns ± 2%   411ns ± 2%   -1.18%  (p=0.000 n=20+19)
StackGrowth-12                    43.4ns ± 1%  43.4ns ± 0%     ~     (p=1.000 n=16+16)
StackGrowthDeep-12                11.4µs ± 4%  10.3µs ± 4%   -9.65%  (p=0.000 n=20+19)

Change-Id: I69a8afbd942c787c591d95b9d9439bd6db4d1e49
Reviewed-on: https://go-review.googlesource.com/30192
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-04 22:40:20 +00:00

216 lines
6 KiB
Go

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package runtime
import (
"unsafe"
)
type slice struct {
array unsafe.Pointer
len int
cap int
}
// maxElems is a lookup table containing the maximum capacity for a slice.
// The index is the size of the slice element.
var maxElems = [...]uintptr{
^uintptr(0),
_MaxMem / 1, _MaxMem / 2, _MaxMem / 3, _MaxMem / 4,
_MaxMem / 5, _MaxMem / 6, _MaxMem / 7, _MaxMem / 8,
_MaxMem / 9, _MaxMem / 10, _MaxMem / 11, _MaxMem / 12,
_MaxMem / 13, _MaxMem / 14, _MaxMem / 15, _MaxMem / 16,
_MaxMem / 17, _MaxMem / 18, _MaxMem / 19, _MaxMem / 20,
_MaxMem / 21, _MaxMem / 22, _MaxMem / 23, _MaxMem / 24,
_MaxMem / 25, _MaxMem / 26, _MaxMem / 27, _MaxMem / 28,
_MaxMem / 29, _MaxMem / 30, _MaxMem / 31, _MaxMem / 32,
}
// maxSliceCap returns the maximum capacity for a slice.
func maxSliceCap(elemsize uintptr) uintptr {
if elemsize < uintptr(len(maxElems)) {
return maxElems[elemsize]
}
return _MaxMem / elemsize
}
func makeslice(et *_type, len, cap int) slice {
// NOTE: The len > maxElements check here is not strictly necessary,
// but it produces a 'len out of range' error instead of a 'cap out of range' error
// when someone does make([]T, bignumber). 'cap out of range' is true too,
// but since the cap is only being supplied implicitly, saying len is clearer.
// See issue 4085.
maxElements := maxSliceCap(et.size)
if len < 0 || uintptr(len) > maxElements {
panic(errorString("makeslice: len out of range"))
}
if cap < len || uintptr(cap) > maxElements {
panic(errorString("makeslice: cap out of range"))
}
p := mallocgc(et.size*uintptr(cap), et, true)
return slice{p, len, cap}
}
func makeslice64(et *_type, len64, cap64 int64) slice {
len := int(len64)
if int64(len) != len64 {
panic(errorString("makeslice: len out of range"))
}
cap := int(cap64)
if int64(cap) != cap64 {
panic(errorString("makeslice: cap out of range"))
}
return makeslice(et, len, cap)
}
// growslice handles slice growth during append.
// It is passed the slice element type, the old slice, and the desired new minimum capacity,
// and it returns a new slice with at least that capacity, with the old data
// copied into it.
// The new slice's length is set to the old slice's length,
// NOT to the new requested capacity.
// This is for codegen convenience. The old slice's length is used immediately
// to calculate where to write new values during an append.
// TODO: When the old backend is gone, reconsider this decision.
// The SSA backend might prefer the new length or to return only ptr/cap and save stack space.
func growslice(et *_type, old slice, cap int) slice {
if raceenabled {
callerpc := getcallerpc(unsafe.Pointer(&et))
racereadrangepc(old.array, uintptr(old.len*int(et.size)), callerpc, funcPC(growslice))
}
if msanenabled {
msanread(old.array, uintptr(old.len*int(et.size)))
}
if et.size == 0 {
if cap < old.cap {
panic(errorString("growslice: cap out of range"))
}
// append should not create a slice with nil pointer but non-zero len.
// We assume that append doesn't need to preserve old.array in this case.
return slice{unsafe.Pointer(&zerobase), old.len, cap}
}
newcap := old.cap
doublecap := newcap + newcap
if cap > doublecap {
newcap = cap
} else {
if old.len < 1024 {
newcap = doublecap
} else {
for newcap < cap {
newcap += newcap / 4
}
}
}
var lenmem, newlenmem, capmem uintptr
const ptrSize = unsafe.Sizeof((*byte)(nil))
switch et.size {
case 1:
lenmem = uintptr(old.len)
newlenmem = uintptr(cap)
capmem = roundupsize(uintptr(newcap))
newcap = int(capmem)
case ptrSize:
lenmem = uintptr(old.len) * ptrSize
newlenmem = uintptr(cap) * ptrSize
capmem = roundupsize(uintptr(newcap) * ptrSize)
newcap = int(capmem / ptrSize)
default:
lenmem = uintptr(old.len) * et.size
newlenmem = uintptr(cap) * et.size
capmem = roundupsize(uintptr(newcap) * et.size)
newcap = int(capmem / et.size)
}
if cap < old.cap || uintptr(newcap) > maxSliceCap(et.size) {
panic(errorString("growslice: cap out of range"))
}
var p unsafe.Pointer
if et.kind&kindNoPointers != 0 {
p = mallocgc(capmem, nil, false)
memmove(p, old.array, lenmem)
// The append() that calls growslice is going to overwrite from old.len to cap (which will be the new length).
// Only clear the part that will not be overwritten.
memclr(add(p, newlenmem), capmem-newlenmem)
} else {
// Note: can't use rawmem (which avoids zeroing of memory), because then GC can scan uninitialized memory.
p = mallocgc(capmem, et, true)
if !writeBarrier.enabled {
memmove(p, old.array, lenmem)
} else {
for i := uintptr(0); i < lenmem; i += et.size {
typedmemmove(et, add(p, i), add(old.array, i))
}
}
}
return slice{p, old.len, newcap}
}
func slicecopy(to, fm slice, width uintptr) int {
if fm.len == 0 || to.len == 0 {
return 0
}
n := fm.len
if to.len < n {
n = to.len
}
if width == 0 {
return n
}
if raceenabled {
callerpc := getcallerpc(unsafe.Pointer(&to))
pc := funcPC(slicecopy)
racewriterangepc(to.array, uintptr(n*int(width)), callerpc, pc)
racereadrangepc(fm.array, uintptr(n*int(width)), callerpc, pc)
}
if msanenabled {
msanwrite(to.array, uintptr(n*int(width)))
msanread(fm.array, uintptr(n*int(width)))
}
size := uintptr(n) * width
if size == 1 { // common case worth about 2x to do here
// TODO: is this still worth it with new memmove impl?
*(*byte)(to.array) = *(*byte)(fm.array) // known to be a byte pointer
} else {
memmove(to.array, fm.array, size)
}
return n
}
func slicestringcopy(to []byte, fm string) int {
if len(fm) == 0 || len(to) == 0 {
return 0
}
n := len(fm)
if len(to) < n {
n = len(to)
}
if raceenabled {
callerpc := getcallerpc(unsafe.Pointer(&to))
pc := funcPC(slicestringcopy)
racewriterangepc(unsafe.Pointer(&to[0]), uintptr(n), callerpc, pc)
}
if msanenabled {
msanwrite(unsafe.Pointer(&to[0]), uintptr(n))
}
memmove(unsafe.Pointer(&to[0]), stringStructOf(&fm).str, uintptr(n))
return n
}