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>