1
0
mirror of https://github.com/golang/go synced 2024-11-06 15:46:19 -07:00
go/src/runtime/mem_js.go

86 lines
2.2 KiB
Go
Raw Normal View History

// Copyright 2018 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.
//go:build js && wasm
package runtime
import (
"unsafe"
)
// Don't split the stack as this function may be invoked without a valid G,
// which prevents us from allocating more stack.
//
//go:nosplit
func sysAllocOS(n uintptr) unsafe.Pointer {
p := sysReserveOS(nil, n)
sysMapOS(p, n)
return p
}
func sysUnusedOS(v unsafe.Pointer, n uintptr) {
}
func sysUsedOS(v unsafe.Pointer, n uintptr) {
}
func sysHugePageOS(v unsafe.Pointer, n uintptr) {
}
// Don't split the stack as this function may be invoked without a valid G,
// which prevents us from allocating more stack.
//
//go:nosplit
func sysFreeOS(v unsafe.Pointer, n uintptr) {
}
func sysFaultOS(v unsafe.Pointer, n uintptr) {
}
var reserveEnd uintptr
func sysReserveOS(v unsafe.Pointer, n uintptr) unsafe.Pointer {
// TODO(neelance): maybe unify with mem_plan9.go, depending on how https://github.com/WebAssembly/design/blob/master/FutureFeatures.md#finer-grained-control-over-memory turns out
if v != nil {
// The address space of WebAssembly's linear memory is contiguous,
// so requesting specific addresses is not supported. We could use
// a different address, but then mheap.sysAlloc discards the result
// right away and we don't reuse chunks passed to sysFree.
return nil
}
runtime: make sysReserve return page-aligned memory on js-wasm This change ensures js-wasm returns page-aligned memory. While today its lack of alignment doesn't cause problems, this is an invariant of sysAlloc which is documented in HACKING.md but isn't upheld by js-wasm. Any code that calls sysAlloc directly for small structures expects a certain alignment (e.g. debuglog, tracebufs) but this is not maintained by js-wasm's sysAlloc. Where sysReserve comes into play is that sysAlloc is implemented in terms of sysReserve on js-wasm. Also, the documentation of sysReserve says that the returned memory is "OS-aligned" which on most platforms means page-aligned, but the "OS-alignment" on js-wasm is effectively 1, which doesn't seem right either. The expected impact of this change is increased memory use on wasm, since there's no way to decommit memory, and any small structures allocated with sysAlloc won't be packed quite as tightly. However, any memory increase should be minimal. Most calls to sysReserve and sysAlloc already aligned their request to physPageSize before calling it; there are only a few circumstances where this is not true, and they involve allocating an amount of memory returned by unsafe.Sizeof where it's actually quite important that we get the alignment right. Updates #35112. Change-Id: I9ca171e507ff3bd186326ccf611b35b9ebea1bfe Reviewed-on: https://go-review.googlesource.com/c/go/+/205277 Run-TryBot: Michael Knyszek <mknyszek@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com> Reviewed-by: Richard Musiol <neelance@gmail.com>
2019-11-04 13:01:18 -07:00
// Round up the initial reserveEnd to 64 KiB so that
// reservations are always aligned to the page size.
initReserveEnd := alignUp(lastmoduledatap.end, physPageSize)
if reserveEnd < initReserveEnd {
reserveEnd = initReserveEnd
}
v = unsafe.Pointer(reserveEnd)
runtime: make sysReserve return page-aligned memory on js-wasm This change ensures js-wasm returns page-aligned memory. While today its lack of alignment doesn't cause problems, this is an invariant of sysAlloc which is documented in HACKING.md but isn't upheld by js-wasm. Any code that calls sysAlloc directly for small structures expects a certain alignment (e.g. debuglog, tracebufs) but this is not maintained by js-wasm's sysAlloc. Where sysReserve comes into play is that sysAlloc is implemented in terms of sysReserve on js-wasm. Also, the documentation of sysReserve says that the returned memory is "OS-aligned" which on most platforms means page-aligned, but the "OS-alignment" on js-wasm is effectively 1, which doesn't seem right either. The expected impact of this change is increased memory use on wasm, since there's no way to decommit memory, and any small structures allocated with sysAlloc won't be packed quite as tightly. However, any memory increase should be minimal. Most calls to sysReserve and sysAlloc already aligned their request to physPageSize before calling it; there are only a few circumstances where this is not true, and they involve allocating an amount of memory returned by unsafe.Sizeof where it's actually quite important that we get the alignment right. Updates #35112. Change-Id: I9ca171e507ff3bd186326ccf611b35b9ebea1bfe Reviewed-on: https://go-review.googlesource.com/c/go/+/205277 Run-TryBot: Michael Knyszek <mknyszek@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com> Reviewed-by: Richard Musiol <neelance@gmail.com>
2019-11-04 13:01:18 -07:00
reserveEnd += alignUp(n, physPageSize)
current := currentMemory()
runtime: make sysReserve return page-aligned memory on js-wasm This change ensures js-wasm returns page-aligned memory. While today its lack of alignment doesn't cause problems, this is an invariant of sysAlloc which is documented in HACKING.md but isn't upheld by js-wasm. Any code that calls sysAlloc directly for small structures expects a certain alignment (e.g. debuglog, tracebufs) but this is not maintained by js-wasm's sysAlloc. Where sysReserve comes into play is that sysAlloc is implemented in terms of sysReserve on js-wasm. Also, the documentation of sysReserve says that the returned memory is "OS-aligned" which on most platforms means page-aligned, but the "OS-alignment" on js-wasm is effectively 1, which doesn't seem right either. The expected impact of this change is increased memory use on wasm, since there's no way to decommit memory, and any small structures allocated with sysAlloc won't be packed quite as tightly. However, any memory increase should be minimal. Most calls to sysReserve and sysAlloc already aligned their request to physPageSize before calling it; there are only a few circumstances where this is not true, and they involve allocating an amount of memory returned by unsafe.Sizeof where it's actually quite important that we get the alignment right. Updates #35112. Change-Id: I9ca171e507ff3bd186326ccf611b35b9ebea1bfe Reviewed-on: https://go-review.googlesource.com/c/go/+/205277 Run-TryBot: Michael Knyszek <mknyszek@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com> Reviewed-by: Richard Musiol <neelance@gmail.com>
2019-11-04 13:01:18 -07:00
// reserveEnd is always at a page boundary.
needed := int32(reserveEnd / physPageSize)
if current < needed {
if growMemory(needed-current) == -1 {
return nil
}
runtime,syscall/js: reuse wasm memory DataView Currently, every call to mem() incurs a new DataView object. This was necessary because the wasm linear memory could grow at any time. Now, whenever the memory grows, we make a call to the front-end. This allows us to reuse the existing DataView object and create a new one only when the memory actually grows. This gives us a boost in performance during DOM operations, while incurring an extra trip to front-end when memory grows. However, since the GrowMemory calls are meant to decrease over the runtime of an application, this is a good tradeoff in the long run. The benchmarks have been tested inside a browser (Google Chrome 75.0.3770.90 (Official Build) (64-bit)). It is hard to get stable nos. for DOM operations since the jumps make the timing very unreliable. But overall, it shows a clear gain. name old time/op new time/op delta DOM 135µs ±26% 84µs ±10% -37.22% (p=0.000 n=10+9) Go1 benchmarks do not show any noticeable degradation: name old time/op new time/op delta BinaryTree17 22.5s ± 0% 22.5s ± 0% ~ (p=0.743 n=8+9) Fannkuch11 15.1s ± 0% 15.1s ± 0% +0.17% (p=0.000 n=9+9) FmtFprintfEmpty 324ns ± 1% 303ns ± 0% -6.64% (p=0.000 n=9+10) FmtFprintfString 535ns ± 1% 515ns ± 0% -3.85% (p=0.000 n=10+10) FmtFprintfInt 609ns ± 0% 589ns ± 0% -3.28% (p=0.000 n=10+10) FmtFprintfIntInt 938ns ± 0% 920ns ± 0% -1.92% (p=0.000 n=9+10) FmtFprintfPrefixedInt 950ns ± 0% 924ns ± 0% -2.72% (p=0.000 n=10+9) FmtFprintfFloat 1.41µs ± 1% 1.43µs ± 0% +1.01% (p=0.000 n=10+10) FmtManyArgs 3.66µs ± 1% 3.46µs ± 0% -5.43% (p=0.000 n=9+10) GobDecode 38.8ms ± 1% 37.8ms ± 0% -2.50% (p=0.000 n=10+8) GobEncode 26.3ms ± 1% 26.3ms ± 0% ~ (p=0.853 n=10+10) Gzip 1.16s ± 1% 1.16s ± 0% -0.37% (p=0.008 n=10+9) Gunzip 210ms ± 0% 208ms ± 1% -1.01% (p=0.000 n=10+10) JSONEncode 48.0ms ± 0% 48.1ms ± 1% +0.29% (p=0.019 n=9+9) JSONDecode 348ms ± 1% 326ms ± 1% -6.34% (p=0.000 n=10+10) Mandelbrot200 6.62ms ± 0% 6.64ms ± 0% +0.37% (p=0.000 n=7+9) GoParse 23.9ms ± 1% 24.7ms ± 1% +2.98% (p=0.000 n=9+9) RegexpMatchEasy0_32 555ns ± 0% 561ns ± 0% +1.10% (p=0.000 n=8+10) RegexpMatchEasy0_1K 3.94µs ± 1% 3.94µs ± 0% ~ (p=0.906 n=9+8) RegexpMatchEasy1_32 516ns ± 0% 524ns ± 0% +1.51% (p=0.000 n=9+10) RegexpMatchEasy1_1K 4.39µs ± 1% 4.40µs ± 1% ~ (p=0.171 n=10+10) RegexpMatchMedium_32 25.1ns ± 0% 25.5ns ± 0% +1.51% (p=0.000 n=9+8) RegexpMatchMedium_1K 196µs ± 0% 203µs ± 1% +3.23% (p=0.000 n=9+10) RegexpMatchHard_32 11.2µs ± 1% 11.6µs ± 1% +3.62% (p=0.000 n=10+10) RegexpMatchHard_1K 334µs ± 1% 348µs ± 1% +4.21% (p=0.000 n=9+10) Revcomp 2.39s ± 0% 2.41s ± 0% +0.78% (p=0.000 n=8+9) Template 385ms ± 1% 336ms ± 0% -12.61% (p=0.000 n=10+9) TimeParse 2.18µs ± 1% 2.18µs ± 1% ~ (p=0.424 n=10+10) TimeFormat 2.28µs ± 1% 2.22µs ± 1% -2.30% (p=0.000 n=10+10) name old speed new speed delta GobDecode 19.8MB/s ± 1% 20.3MB/s ± 0% +2.56% (p=0.000 n=10+8) GobEncode 29.1MB/s ± 1% 29.2MB/s ± 0% ~ (p=0.810 n=10+10) Gzip 16.7MB/s ± 1% 16.8MB/s ± 0% +0.37% (p=0.007 n=10+9) Gunzip 92.2MB/s ± 0% 93.2MB/s ± 1% +1.03% (p=0.000 n=10+10) JSONEncode 40.4MB/s ± 0% 40.3MB/s ± 1% -0.28% (p=0.025 n=9+9) JSONDecode 5.58MB/s ± 1% 5.96MB/s ± 1% +6.80% (p=0.000 n=10+10) GoParse 2.42MB/s ± 0% 2.35MB/s ± 1% -2.83% (p=0.000 n=8+9) RegexpMatchEasy0_32 57.7MB/s ± 0% 57.0MB/s ± 0% -1.09% (p=0.000 n=8+10) RegexpMatchEasy0_1K 260MB/s ± 1% 260MB/s ± 0% ~ (p=0.963 n=9+8) RegexpMatchEasy1_32 62.1MB/s ± 0% 61.1MB/s ± 0% -1.53% (p=0.000 n=10+10) RegexpMatchEasy1_1K 233MB/s ± 1% 233MB/s ± 1% ~ (p=0.190 n=10+10) RegexpMatchMedium_32 39.8MB/s ± 0% 39.1MB/s ± 1% -1.74% (p=0.000 n=9+10) RegexpMatchMedium_1K 5.21MB/s ± 0% 5.05MB/s ± 1% -3.09% (p=0.000 n=9+10) RegexpMatchHard_32 2.86MB/s ± 1% 2.76MB/s ± 1% -3.43% (p=0.000 n=10+10) RegexpMatchHard_1K 3.06MB/s ± 1% 2.94MB/s ± 1% -4.06% (p=0.000 n=9+10) Revcomp 106MB/s ± 0% 105MB/s ± 0% -0.77% (p=0.000 n=8+9) Template 5.04MB/s ± 1% 5.77MB/s ± 0% +14.48% (p=0.000 n=10+9) Updates #32591 Change-Id: Id567e14a788e359248b2129ef1cf0adc8cc4ab7f Reviewed-on: https://go-review.googlesource.com/c/go/+/183457 Run-TryBot: Agniva De Sarker <agniva.quicksilver@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Richard Musiol <neelance@gmail.com>
2019-06-22 01:37:57 -06:00
resetMemoryDataView()
}
return v
}
func currentMemory() int32
func growMemory(pages int32) int32
runtime,syscall/js: reuse wasm memory DataView Currently, every call to mem() incurs a new DataView object. This was necessary because the wasm linear memory could grow at any time. Now, whenever the memory grows, we make a call to the front-end. This allows us to reuse the existing DataView object and create a new one only when the memory actually grows. This gives us a boost in performance during DOM operations, while incurring an extra trip to front-end when memory grows. However, since the GrowMemory calls are meant to decrease over the runtime of an application, this is a good tradeoff in the long run. The benchmarks have been tested inside a browser (Google Chrome 75.0.3770.90 (Official Build) (64-bit)). It is hard to get stable nos. for DOM operations since the jumps make the timing very unreliable. But overall, it shows a clear gain. name old time/op new time/op delta DOM 135µs ±26% 84µs ±10% -37.22% (p=0.000 n=10+9) Go1 benchmarks do not show any noticeable degradation: name old time/op new time/op delta BinaryTree17 22.5s ± 0% 22.5s ± 0% ~ (p=0.743 n=8+9) Fannkuch11 15.1s ± 0% 15.1s ± 0% +0.17% (p=0.000 n=9+9) FmtFprintfEmpty 324ns ± 1% 303ns ± 0% -6.64% (p=0.000 n=9+10) FmtFprintfString 535ns ± 1% 515ns ± 0% -3.85% (p=0.000 n=10+10) FmtFprintfInt 609ns ± 0% 589ns ± 0% -3.28% (p=0.000 n=10+10) FmtFprintfIntInt 938ns ± 0% 920ns ± 0% -1.92% (p=0.000 n=9+10) FmtFprintfPrefixedInt 950ns ± 0% 924ns ± 0% -2.72% (p=0.000 n=10+9) FmtFprintfFloat 1.41µs ± 1% 1.43µs ± 0% +1.01% (p=0.000 n=10+10) FmtManyArgs 3.66µs ± 1% 3.46µs ± 0% -5.43% (p=0.000 n=9+10) GobDecode 38.8ms ± 1% 37.8ms ± 0% -2.50% (p=0.000 n=10+8) GobEncode 26.3ms ± 1% 26.3ms ± 0% ~ (p=0.853 n=10+10) Gzip 1.16s ± 1% 1.16s ± 0% -0.37% (p=0.008 n=10+9) Gunzip 210ms ± 0% 208ms ± 1% -1.01% (p=0.000 n=10+10) JSONEncode 48.0ms ± 0% 48.1ms ± 1% +0.29% (p=0.019 n=9+9) JSONDecode 348ms ± 1% 326ms ± 1% -6.34% (p=0.000 n=10+10) Mandelbrot200 6.62ms ± 0% 6.64ms ± 0% +0.37% (p=0.000 n=7+9) GoParse 23.9ms ± 1% 24.7ms ± 1% +2.98% (p=0.000 n=9+9) RegexpMatchEasy0_32 555ns ± 0% 561ns ± 0% +1.10% (p=0.000 n=8+10) RegexpMatchEasy0_1K 3.94µs ± 1% 3.94µs ± 0% ~ (p=0.906 n=9+8) RegexpMatchEasy1_32 516ns ± 0% 524ns ± 0% +1.51% (p=0.000 n=9+10) RegexpMatchEasy1_1K 4.39µs ± 1% 4.40µs ± 1% ~ (p=0.171 n=10+10) RegexpMatchMedium_32 25.1ns ± 0% 25.5ns ± 0% +1.51% (p=0.000 n=9+8) RegexpMatchMedium_1K 196µs ± 0% 203µs ± 1% +3.23% (p=0.000 n=9+10) RegexpMatchHard_32 11.2µs ± 1% 11.6µs ± 1% +3.62% (p=0.000 n=10+10) RegexpMatchHard_1K 334µs ± 1% 348µs ± 1% +4.21% (p=0.000 n=9+10) Revcomp 2.39s ± 0% 2.41s ± 0% +0.78% (p=0.000 n=8+9) Template 385ms ± 1% 336ms ± 0% -12.61% (p=0.000 n=10+9) TimeParse 2.18µs ± 1% 2.18µs ± 1% ~ (p=0.424 n=10+10) TimeFormat 2.28µs ± 1% 2.22µs ± 1% -2.30% (p=0.000 n=10+10) name old speed new speed delta GobDecode 19.8MB/s ± 1% 20.3MB/s ± 0% +2.56% (p=0.000 n=10+8) GobEncode 29.1MB/s ± 1% 29.2MB/s ± 0% ~ (p=0.810 n=10+10) Gzip 16.7MB/s ± 1% 16.8MB/s ± 0% +0.37% (p=0.007 n=10+9) Gunzip 92.2MB/s ± 0% 93.2MB/s ± 1% +1.03% (p=0.000 n=10+10) JSONEncode 40.4MB/s ± 0% 40.3MB/s ± 1% -0.28% (p=0.025 n=9+9) JSONDecode 5.58MB/s ± 1% 5.96MB/s ± 1% +6.80% (p=0.000 n=10+10) GoParse 2.42MB/s ± 0% 2.35MB/s ± 1% -2.83% (p=0.000 n=8+9) RegexpMatchEasy0_32 57.7MB/s ± 0% 57.0MB/s ± 0% -1.09% (p=0.000 n=8+10) RegexpMatchEasy0_1K 260MB/s ± 1% 260MB/s ± 0% ~ (p=0.963 n=9+8) RegexpMatchEasy1_32 62.1MB/s ± 0% 61.1MB/s ± 0% -1.53% (p=0.000 n=10+10) RegexpMatchEasy1_1K 233MB/s ± 1% 233MB/s ± 1% ~ (p=0.190 n=10+10) RegexpMatchMedium_32 39.8MB/s ± 0% 39.1MB/s ± 1% -1.74% (p=0.000 n=9+10) RegexpMatchMedium_1K 5.21MB/s ± 0% 5.05MB/s ± 1% -3.09% (p=0.000 n=9+10) RegexpMatchHard_32 2.86MB/s ± 1% 2.76MB/s ± 1% -3.43% (p=0.000 n=10+10) RegexpMatchHard_1K 3.06MB/s ± 1% 2.94MB/s ± 1% -4.06% (p=0.000 n=9+10) Revcomp 106MB/s ± 0% 105MB/s ± 0% -0.77% (p=0.000 n=8+9) Template 5.04MB/s ± 1% 5.77MB/s ± 0% +14.48% (p=0.000 n=10+9) Updates #32591 Change-Id: Id567e14a788e359248b2129ef1cf0adc8cc4ab7f Reviewed-on: https://go-review.googlesource.com/c/go/+/183457 Run-TryBot: Agniva De Sarker <agniva.quicksilver@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Richard Musiol <neelance@gmail.com>
2019-06-22 01:37:57 -06:00
// resetMemoryDataView signals the JS front-end that WebAssembly's memory.grow instruction has been used.
// This allows the front-end to replace the old DataView object with a new one.
func resetMemoryDataView()
func sysMapOS(v unsafe.Pointer, n uintptr) {
}