This implements runtime support for buffered write barriers on amd64.
The buffered write barrier has a fast path that simply enqueues
pointers in a per-P buffer. Unlike the current write barrier, this
fast path is *not* a normal Go call and does not require the compiler
to spill general-purpose registers or put arguments on the stack. When
the buffer fills up, the write barrier takes the slow path, which
spills all general purpose registers and flushes the buffer. We don't
allow safe-points or stack splits while this frame is active, so it
doesn't matter that we have no type information for the spilled
registers in this frame.
One minor complication is cgocheck=2 mode, which uses the write
barrier to detect Go pointers being written to non-Go memory. We
obviously can't buffer this, so instead we set the buffer to its
minimum size, forcing the write barrier into the slow path on every
call. For this specific case, we pass additional information as
arguments to the flush function. This also requires enabling the cgo
write barrier slightly later during runtime initialization, after Ps
(and the per-P write barrier buffers) have been initialized.
The code in this CL is not yet active. The next CL will modify the
compiler to generate calls to the new write barrier.
This reduces the average cost of the write barrier by roughly a factor
of 4, which will pay for the cost of having it enabled more of the
time after we make the GC pacer less aggressive. (Benchmarks will be
in the next CL.)
Updates #14951.
Updates #22460.
Change-Id: I396b5b0e2c5e5c4acfd761a3235fd15abadc6cb1
Reviewed-on: https://go-review.googlesource.com/73711
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Currently most of these are marked go:nowritebarrier as a hint, but
it's actually important that these not invoke write barriers
recursively. The danger is that some gcWork method would invoke the
write barrier while the gcWork is in an inconsistent state and that
the write barrier would in turn invoke some other gcWork method, which
would crash or permanently corrupt the gcWork. Simply marking the
write barrier itself as go:nowritebarrierrec isn't sufficient to
prevent this if the write barrier doesn't use the outer method.
Thankfully, this doesn't cause any build failures, so we were getting
this right. :)
For #22460.
Change-Id: I35a7292a584200eb35a49507cd3fe359ba2206f6
Reviewed-on: https://go-review.googlesource.com/72554
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
This extends the sweeper to free workbufs back to the heap between GC
cycles, allowing this memory to be reused for GC'd allocations or
eventually returned to the OS.
This helps for applications that have high peak heap usage relative to
their regular heap usage (for example, a high-memory initialization
phase). Workbuf memory is roughly proportional to heap size and since
we currently never free workbufs, it's proportional to *peak* heap
size. By freeing workbufs, we can release and reuse this memory for
other purposes when the heap shrinks.
This is somewhat complicated because this costs ~1–2 µs per workbuf
span, so for large heaps it's too expensive to just do synchronously
after mark termination between starting the world and dropping the
worldsema. Hence, we do it asynchronously in the sweeper. This adds a
list of "free" workbuf spans that can be returned to the heap. GC
moves all workbuf spans to this list after mark termination and the
background sweeper drains this list back to the heap. If the sweeper
doesn't finish, that's fine, since getempty can directly reuse any
remaining spans to allocate more workbufs.
Performance impact is negligible. On the x/benchmarks, this reduces
GC-bytes-from-system by 6–11%.
Fixes#19325.
Change-Id: Icb92da2196f0c39ee984faf92d52f29fd9ded7a8
Reviewed-on: https://go-review.googlesource.com/38582
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Currently the runtime allocates workbufs from persistent memory, which
means they can never be freed.
Switch to allocating them from manually-managed heap spans. This
doesn't free them yet, but it puts us in a position to do so.
For #19325.
Change-Id: I94b2512a2f2bbbb456cd9347761b9412e80d2da9
Reviewed-on: https://go-review.googlesource.com/38581
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
The lfstack API is still a C-style API: lfstacks all have unhelpful
type uint64 and the APIs are package-level functions. Make the code
more readable and Go-style by creating an lfstack type with methods
for push, pop, and empty.
Change-Id: I64685fa3be0e82ae2d1a782a452a50974440a827
Reviewed-on: https://go-review.googlesource.com/38290
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
The gcstats structure is no longer consumed by anything and no longer
tracks statistics that are particularly relevant to the concurrent
garbage collector. Remove it. (Having statistics is probably a good
idea, but these aren't the stats we need these days and we don't have
a way to get them out of the runtime.)
In preparation for #13613.
Change-Id: Ib63e2f9067850668f9dcbfd4ed89aab4a6622c3f
Reviewed-on: https://go-review.googlesource.com/34936
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>
Since workbuf is now marked go:notinheap, the write barrier-preventing
wrapper type wbufptr is no longer necessary. Remove it.
Change-Id: I3e5b5803a1547d65de1c1a9c22458a38e08549b7
Reviewed-on: https://go-review.googlesource.com/35971
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
If the scheduler has no user work and there's no GC work visible, it
puts the P to sleep (or blocks on the network). However, if we later
enqueue more GC work, there's currently nothing that specifically
wakes up the scheduler to let it start an idle GC worker. As a result,
we can underutilize the CPU during GC if Ps have been put to sleep.
Fix this by making GC wake idle Ps when work buffers are put on the
full list. We already have a hook to do this, since we use this to
preempt a random P if we need more dedicated workers. We expand this
hook to instead wake an idle P if there is one. The logic we use for
this is identical to the logic used to wake an idle P when we ready a
goroutine.
To make this really sound, we also fix the scheduler to re-check the
idle GC worker condition after releasing its P. This closes a race
where 1) the scheduler checks for idle work and finds none, 2) new
work is enqueued but there are no idle Ps so none are woken, and 3)
the scheduler releases its P.
There is one subtlety here. Currently we call enlistWorker directly
from putfull, but the gcWork is in an inconsistent state in the places
that call putfull. This isn't a problem right now because nothing that
enlistWorker does touches the gcWork, but with the added call to
wakep, it's possible to get a recursive call into the gcWork
(specifically, while write barriers are disallowed, this can do an
allocation, which can dispose a gcWork, which can put a workbuf). To
handle this, we lift the enlistWorker calls up a layer and delay them
until the gcWork is in a consistent state.
Fixes#14179.
Change-Id: Ia2467a52e54c9688c3c1752e1fc00f5b37bbfeeb
Reviewed-on: https://go-review.googlesource.com/32434
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
This covers basically all sysAlloc'd, persistentalloc'd, and
fixalloc'd types.
Change-Id: I0487c887c2a0ade5e33d4c4c12d837e97468e66b
Reviewed-on: https://go-review.googlesource.com/30941
Reviewed-by: Rick Hudson <rlh@golang.org>
Currently the time spent in scanobject is proportional to the size of
the object being scanned. Since scanobject is non-preemptible, large
objects can cause significant goroutine (and even whole application)
delays through several means:
1. If a GC assist picks up a large object, the allocating goroutine is
blocked for the whole scan, even if that scan well exceeds that
goroutine's debt.
2. Since the scheduler does not run on the P performing a large object
scan, goroutines in that P's run queue do not run unless they are
stolen by another P (which can take some time). If there are a few
large objects, all of the Ps may get tied up so the scheduler
doesn't run anywhere.
3. Even if a large object is scanned by a background worker and other
Ps are still running the scheduler, the large object scan doesn't
flush background credit until the whole scan is done. This can
easily cause all allocations to block in assists, waiting for
credit, causing an effective STW.
Fix this by splitting large objects into 128 KB "oblets" and scanning
at most one oblet at a time. Since we can scan 1–2 MB/ms, this equates
to bounding scanobject at roughly 100 µs. This improves assist
behavior both because assists can no longer get "unlucky" and be stuck
scanning a large object, and because it causes the background worker
to flush credit and unblock assists more frequently when scanning
large objects. This also improves GC parallelism if the heap consists
primarily of a small number of very large objects by letting multiple
workers scan a large objects in parallel.
Fixes#10345. Fixes#16293.
This substantially improves goroutine latency in the benchmark from
issue #16293, which exercises several forms of very large objects:
name old max-latency new max-latency delta
SliceNoPointer-12 154µs ± 1% 155µs ± 2% ~ (p=0.087 n=13+12)
SlicePointer-12 314ms ± 1% 5.94ms ±138% -98.11% (p=0.000 n=19+20)
SliceLivePointer-12 1148ms ± 0% 4.72ms ±167% -99.59% (p=0.000 n=19+20)
MapNoPointer-12 72509µs ± 1% 408µs ±325% -99.44% (p=0.000 n=19+18)
ChanPointer-12 313ms ± 0% 4.74ms ±140% -98.49% (p=0.000 n=18+20)
ChanLivePointer-12 1147ms ± 0% 3.30ms ±149% -99.71% (p=0.000 n=19+20)
name old P99.9-latency new P99.9-latency delta
SliceNoPointer-12 113µs ±25% 107µs ±12% ~ (p=0.153 n=20+18)
SlicePointer-12 309450µs ± 0% 133µs ±23% -99.96% (p=0.000 n=20+20)
SliceLivePointer-12 961ms ± 0% 1.35ms ±27% -99.86% (p=0.000 n=20+20)
MapNoPointer-12 448µs ±288% 119µs ±18% -73.34% (p=0.000 n=18+20)
ChanPointer-12 309450µs ± 0% 134µs ±23% -99.96% (p=0.000 n=20+19)
ChanLivePointer-12 961ms ± 0% 1.35ms ±27% -99.86% (p=0.000 n=20+20)
This has negligible effect on all metrics from the garbage, JSON, and
HTTP x/benchmarks.
It shows slight improvement on some of the go1 benchmarks,
particularly Revcomp, which uses some multi-megabyte buffers:
name old time/op new time/op delta
BinaryTree17-12 2.46s ± 1% 2.47s ± 1% +0.32% (p=0.012 n=20+20)
Fannkuch11-12 2.82s ± 0% 2.81s ± 0% -0.61% (p=0.000 n=17+20)
FmtFprintfEmpty-12 50.8ns ± 5% 50.5ns ± 2% ~ (p=0.197 n=17+19)
FmtFprintfString-12 131ns ± 1% 132ns ± 0% +0.57% (p=0.000 n=20+16)
FmtFprintfInt-12 117ns ± 0% 116ns ± 0% -0.47% (p=0.000 n=15+20)
FmtFprintfIntInt-12 180ns ± 0% 179ns ± 1% -0.78% (p=0.000 n=16+20)
FmtFprintfPrefixedInt-12 186ns ± 1% 185ns ± 1% -0.55% (p=0.000 n=19+20)
FmtFprintfFloat-12 263ns ± 1% 271ns ± 0% +2.84% (p=0.000 n=18+20)
FmtManyArgs-12 741ns ± 1% 742ns ± 1% ~ (p=0.190 n=19+19)
GobDecode-12 7.44ms ± 0% 7.35ms ± 1% -1.21% (p=0.000 n=20+20)
GobEncode-12 6.22ms ± 1% 6.21ms ± 1% ~ (p=0.336 n=20+19)
Gzip-12 220ms ± 1% 219ms ± 1% ~ (p=0.130 n=19+19)
Gunzip-12 37.9ms ± 0% 37.9ms ± 1% ~ (p=1.000 n=20+19)
HTTPClientServer-12 82.5µs ± 3% 82.6µs ± 3% ~ (p=0.776 n=20+19)
JSONEncode-12 16.4ms ± 1% 16.5ms ± 2% +0.49% (p=0.003 n=18+19)
JSONDecode-12 53.7ms ± 1% 54.1ms ± 1% +0.71% (p=0.000 n=19+18)
Mandelbrot200-12 4.19ms ± 1% 4.20ms ± 1% ~ (p=0.452 n=19+19)
GoParse-12 3.38ms ± 1% 3.37ms ± 1% ~ (p=0.123 n=19+19)
RegexpMatchEasy0_32-12 72.1ns ± 1% 71.8ns ± 1% ~ (p=0.397 n=19+17)
RegexpMatchEasy0_1K-12 242ns ± 0% 242ns ± 0% ~ (p=0.168 n=17+20)
RegexpMatchEasy1_32-12 72.1ns ± 1% 72.1ns ± 1% ~ (p=0.538 n=18+19)
RegexpMatchEasy1_1K-12 385ns ± 1% 384ns ± 1% ~ (p=0.388 n=20+20)
RegexpMatchMedium_32-12 112ns ± 1% 112ns ± 3% ~ (p=0.539 n=20+20)
RegexpMatchMedium_1K-12 34.4µs ± 2% 34.4µs ± 2% ~ (p=0.628 n=18+18)
RegexpMatchHard_32-12 1.80µs ± 1% 1.80µs ± 1% ~ (p=0.522 n=18+19)
RegexpMatchHard_1K-12 54.0µs ± 1% 54.1µs ± 1% ~ (p=0.647 n=20+19)
Revcomp-12 387ms ± 1% 369ms ± 5% -4.89% (p=0.000 n=17+19)
Template-12 62.3ms ± 1% 62.0ms ± 0% -0.48% (p=0.002 n=20+17)
TimeParse-12 314ns ± 1% 314ns ± 0% ~ (p=1.011 n=20+13)
TimeFormat-12 358ns ± 0% 354ns ± 0% -1.12% (p=0.000 n=17+20)
[Geo mean] 53.5µs 53.3µs -0.23%
Change-Id: I2a0a179d1d6bf7875dd054b7693dd12d2a340132
Reviewed-on: https://go-review.googlesource.com/23540
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
The complexity of the GC work buffers put and tryGet
prevented them from being inlined. This CL simplifies
the fast path thus enabling inlining. If the fast
path does not succeed the previous put and tryGet
functions are called.
Change-Id: I6da6495d0dadf42bd0377c110b502274cc01acf5
Reviewed-on: https://go-review.googlesource.com/20704
Reviewed-by: Austin Clements <austin@google.com>
The tree's pretty inconsistent about single space vs double space
after a period in documentation. Make it consistently a single space,
per earlier decisions. This means contributors won't be confused by
misleading precedence.
This CL doesn't use go/doc to parse. It only addresses // comments.
It was generated with:
$ perl -i -npe 's,^(\s*// .+[a-z]\.) +([A-Z]),$1 $2,' $(git grep -l -E '^\s*//(.+\.) +([A-Z])')
$ go test go/doc -update
Change-Id: Iccdb99c37c797ef1f804a94b22ba5ee4b500c4f7
Reviewed-on: https://go-review.googlesource.com/20022
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Dave Day <djd@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Currently most uses of gcWork use the per-P gcWork, but there are two
places that still use a stack-based gcWork. Simplify things by making
these instead use the per-P gcWork.
Change-Id: I712d012cce9dd5757c8541824e9641ac1c2a329c
Reviewed-on: https://go-review.googlesource.com/19636
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
When gcWork was first introduced, the compiler's escape analysis
wasn't good enough to detect that that method receiver didn't escape,
so we had to hack around this.
Now that the compiler can figure out this for itself, remove these
hacks.
Change-Id: I9f73fab721e272410b8b6905b564e7abc03c0dfe
Reviewed-on: https://go-review.googlesource.com/19634
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
runtime/internal/sys will hold system-, architecture- and config-
specific constants.
Updates #11647
Change-Id: I6db29c312556087a42e8d2bdd9af40d157c56b54
Reviewed-on: https://go-review.googlesource.com/16817
Reviewed-by: Russ Cox <rsc@golang.org>
This change breaks out most of the atomics functions in the runtime
into package runtime/internal/atomic. It adds some basic support
in the toolchain for runtime packages, and also modifies linux/arm
atomics to remove the dependency on the runtime's mutex. The mutexes
have been replaced with spinlocks.
all trybots are happy!
In addition to the trybots, I've tested on the darwin/arm64 builder,
on the darwin/arm builder, and on a ppc64le machine.
Change-Id: I6698c8e3cf3834f55ce5824059f44d00dc8e3c2f
Reviewed-on: https://go-review.googlesource.com/14204
Run-TryBot: Michael Matloob <matloob@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Currently the GC work buffers are only 256 bytes and hence can record
only 24 64-bit pointer. They were reduced from 4K in commits db7fd1c
and a15818f as a way to minimize the amount of work the per-P workbuf
caches could "hide" from the mark phase and carry in to the mark
termination phase. However, this approach wasn't very robust and we
later added a "mark 2" phase to address this problem head-on.
Because of mark 2, there's now no benefit to having very small work
buffers. But there are plenty of downsides: small work buffers
increase contention on the work lists, increase the frequency and
hence net overhead of acquiring and releasing work buffers, and
somewhat increase memory overhead of the GC.
This commit expands work buffers back to 4K (504 64-bit pointers).
This reduces the rate of writes to work.full in the garbage benchmark
from a peak of ~780,000 writes/sec to a peak of ~32,000 writes/sec.
This has negligible effect on the go1 benchmarks. It slightly slows
down the garbage benchmark.
name old time/op new time/op delta
XBenchGarbage-12 5.37ms ± 5% 5.60ms ± 2% +4.37% (p=0.000 n=20+20)
Change-Id: Ic9cc28e7a125d23d9faf4f5e690fb8aa9bcdfb28
Reviewed-on: https://go-review.googlesource.com/15893
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Currently the concurrent root scan is performed in its entirety by the
GC coordinator before entering concurrent mark (which enables GC
workers). This scan is done sequentially, which can prolong the scan
phase, delay the mark phase, and means that the scan phase does not
obey the 25% CPU goal. Furthermore, there's no need to complete the
root scan before starting marking (in fact, we already allow GC
assists to happen during the scan phase), so this acts as an
unnecessary barrier between root scanning and marking.
This change shifts the root scan work out of the GC coordinator and in
to the GC workers. The coordinator simply sets up the scan state and
enqueues the right number of root scan jobs. The GC workers then drain
the root scan jobs prior to draining heap scan jobs.
This parallelizes the root scan process, makes it obey the 25% CPU
goal, and effectively eliminates root scanning as an isolated phase,
allowing the system to smoothly transition from root scanning to heap
marking. This also eliminates a major non-STW responsibility of the GC
coordinator, which will make it easier to switch to a decentralized
state machine. Finally, it puts us in a good position to perform root
scanning in assists as well, which will help satisfy assists at the
beginning of the GC cycle.
This is mostly straightforward. One tricky aspect is that we have to
deal with preemption deadlock: where two non-preemptible gorountines
are trying to preempt each other to perform a stack scan. Given the
context where this happens, the only instance of this is two
background workers trying to scan each other. We avoid this by simply
not scanning the stacks of background workers during the concurrent
phase; this is safe because we'll scan them during mark termination
(and their stacks are *very* small and should not contain any new
pointers).
This change also switches the root marking during mark termination to
use the same gcDrain-based code path as concurrent mark. This
shouldn't affect performance because STW root marking was already
parallel and tasks switched to heap marking immediately when no more
root marking tasks were available. However, it simplifies the code and
unifies these code paths.
This has negligible effect on the go1 benchmarks. It slightly slows
down the garbage benchmark, possibly by making GC run slightly more
frequently.
name old time/op new time/op delta
XBenchGarbage-12 5.10ms ± 1% 5.24ms ± 1% +2.87% (p=0.000 n=18+18)
name old time/op new time/op delta
BinaryTree17-12 3.25s ± 3% 3.20s ± 5% -1.57% (p=0.013 n=20+20)
Fannkuch11-12 2.45s ± 1% 2.46s ± 1% +0.38% (p=0.019 n=20+18)
FmtFprintfEmpty-12 49.7ns ± 3% 49.9ns ± 4% ~ (p=0.851 n=19+20)
FmtFprintfString-12 170ns ± 2% 170ns ± 1% ~ (p=0.775 n=20+19)
FmtFprintfInt-12 161ns ± 1% 160ns ± 1% -0.78% (p=0.000 n=19+18)
FmtFprintfIntInt-12 267ns ± 1% 270ns ± 1% +1.04% (p=0.000 n=19+19)
FmtFprintfPrefixedInt-12 238ns ± 2% 238ns ± 1% ~ (p=0.133 n=18+19)
FmtFprintfFloat-12 311ns ± 1% 310ns ± 2% -0.35% (p=0.023 n=20+19)
FmtManyArgs-12 1.08µs ± 1% 1.06µs ± 1% -2.31% (p=0.000 n=20+20)
GobDecode-12 8.65ms ± 1% 8.63ms ± 1% ~ (p=0.377 n=18+20)
GobEncode-12 6.49ms ± 1% 6.52ms ± 1% +0.37% (p=0.015 n=20+20)
Gzip-12 319ms ± 3% 318ms ± 1% ~ (p=0.975 n=19+17)
Gunzip-12 41.9ms ± 1% 42.1ms ± 2% +0.65% (p=0.004 n=19+20)
HTTPClientServer-12 61.7µs ± 1% 62.6µs ± 1% +1.40% (p=0.000 n=18+20)
JSONEncode-12 16.8ms ± 1% 16.9ms ± 1% ~ (p=0.239 n=20+18)
JSONDecode-12 58.4ms ± 1% 60.7ms ± 1% +3.85% (p=0.000 n=19+20)
Mandelbrot200-12 3.86ms ± 0% 3.86ms ± 1% ~ (p=0.092 n=18+19)
GoParse-12 3.75ms ± 2% 3.75ms ± 2% ~ (p=0.708 n=19+20)
RegexpMatchEasy0_32-12 100ns ± 1% 100ns ± 2% +0.60% (p=0.010 n=17+20)
RegexpMatchEasy0_1K-12 341ns ± 1% 342ns ± 2% ~ (p=0.203 n=20+19)
RegexpMatchEasy1_32-12 82.5ns ± 2% 83.2ns ± 2% +0.83% (p=0.007 n=19+19)
RegexpMatchEasy1_1K-12 495ns ± 1% 495ns ± 2% ~ (p=0.970 n=19+18)
RegexpMatchMedium_32-12 130ns ± 2% 130ns ± 2% +0.59% (p=0.039 n=19+20)
RegexpMatchMedium_1K-12 39.2µs ± 1% 39.3µs ± 1% ~ (p=0.214 n=18+18)
RegexpMatchHard_32-12 2.03µs ± 2% 2.02µs ± 1% ~ (p=0.166 n=18+19)
RegexpMatchHard_1K-12 61.0µs ± 1% 60.9µs ± 1% ~ (p=0.169 n=20+18)
Revcomp-12 533ms ± 1% 535ms ± 1% ~ (p=0.071 n=19+17)
Template-12 68.1ms ± 2% 73.0ms ± 1% +7.26% (p=0.000 n=19+20)
TimeParse-12 355ns ± 2% 356ns ± 2% ~ (p=0.530 n=19+20)
TimeFormat-12 357ns ± 2% 347ns ± 1% -2.59% (p=0.000 n=20+19)
[Geo mean] 62.1µs 62.3µs +0.31%
name old speed new speed delta
GobDecode-12 88.7MB/s ± 1% 88.9MB/s ± 1% ~ (p=0.377 n=18+20)
GobEncode-12 118MB/s ± 1% 118MB/s ± 1% -0.37% (p=0.015 n=20+20)
Gzip-12 60.9MB/s ± 3% 60.9MB/s ± 1% ~ (p=0.944 n=19+17)
Gunzip-12 464MB/s ± 1% 461MB/s ± 2% -0.64% (p=0.004 n=19+20)
JSONEncode-12 115MB/s ± 1% 115MB/s ± 1% ~ (p=0.236 n=20+18)
JSONDecode-12 33.2MB/s ± 1% 32.0MB/s ± 1% -3.71% (p=0.000 n=19+20)
GoParse-12 15.5MB/s ± 2% 15.5MB/s ± 2% ~ (p=0.702 n=19+20)
RegexpMatchEasy0_32-12 320MB/s ± 1% 318MB/s ± 2% ~ (p=0.094 n=18+20)
RegexpMatchEasy0_1K-12 3.00GB/s ± 1% 2.99GB/s ± 1% ~ (p=0.194 n=20+19)
RegexpMatchEasy1_32-12 388MB/s ± 2% 385MB/s ± 2% -0.83% (p=0.008 n=19+19)
RegexpMatchEasy1_1K-12 2.07GB/s ± 1% 2.07GB/s ± 1% ~ (p=0.964 n=19+18)
RegexpMatchMedium_32-12 7.68MB/s ± 1% 7.64MB/s ± 2% -0.57% (p=0.020 n=19+20)
RegexpMatchMedium_1K-12 26.1MB/s ± 1% 26.1MB/s ± 1% ~ (p=0.211 n=18+18)
RegexpMatchHard_32-12 15.8MB/s ± 1% 15.8MB/s ± 1% ~ (p=0.180 n=18+19)
RegexpMatchHard_1K-12 16.8MB/s ± 1% 16.8MB/s ± 2% ~ (p=0.236 n=20+19)
Revcomp-12 477MB/s ± 1% 475MB/s ± 1% ~ (p=0.071 n=19+17)
Template-12 28.5MB/s ± 2% 26.6MB/s ± 1% -6.77% (p=0.000 n=19+20)
[Geo mean] 100MB/s 99.0MB/s -0.82%
Change-Id: I875bf6ceb306d1ee2f470cabf88aa6ede27c47a0
Reviewed-on: https://go-review.googlesource.com/16059
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This work queue is no longer used (there are many reads of
work.partial, but the only write is in putpartial, which is never
called).
Fixes#11922.
Change-Id: I08b76c0c02a0867a9cdcb94783e1f7629d44249a
Reviewed-on: https://go-review.googlesource.com/15892
Reviewed-by: Rick Hudson <rlh@golang.org>
Currently, gcController.scanWork is updated as lazily as possible
since it is only read at the end of the GC cycle. We're about to read
it during the GC cycle to improve the assist ratio revisions, so
modify gcDrain* to regularly flush to gcController.scanWork in much
the same way as we regularly flush to gcController.bgScanCredit.
One consequence of this is that it's difficult to keep gcw.scanWork
monotonic, so we give up on that and simply return the amount of scan
work done by gcDrainN rather than calculating it in the caller.
Change-Id: I7b50acdc39602f843eed0b5c6d2dacd7e762b81d
Reviewed-on: https://go-review.googlesource.com/15407
Reviewed-by: Rick Hudson <rlh@golang.org>
The scheduler, work buffer's dispose, and write barriers
can conspire to hide the a pointer from the GC's concurent
mark phase. If this pointer is the only path to a large
amount of marking the STW mark termination phase may take
a lot of time.
Consider the following:
1) dispose places a work buffer on the partial queue
2) the GC is busy so it does not immediately remove and
process the work buffer
3) the scheduler runs a mutator whose write barrier dequeues the
work buffer from the partial queue so the GC won't see it
This repeats until the GC reaches the mark termination
phase where the GC finally discovers the pointer along
with a lot of work to do.
This CL fixes the problem by having the mutator
dispose of the buffer to the full queue instead of
the partial queue. Since the write buffer never asks for full
buffers the conspiracy described above is not possible.
Updates #11694.
Change-Id: I2ce832f9657a7570f800e8ce4459cd9e304ef43b
Reviewed-on: https://go-review.googlesource.com/12840
Reviewed-by: Austin Clements <austin@google.com>
Some latency regressions have crept into our system over the past few
weeks. This CL fixes those by having the mark phase more aggressively
blacken objects so that the mark termination phase, a STW phase, has less
work to do. Three approaches were taken when the mark phase believes
it has no more work to do, ie all the work buffers are empty.
If things have gone well the mark phase is correct and there is
in fact little or no work. In that case the following items will
take very little time. If the mark phase is wrong this CL will
ferret that work out and give the mark phase a chance to deal with
it concurrently before mark termination begins.
When the mark phase first appears to be out of work, it does three things:
1) It switches from allocating white to allocating black to reduce the
number of unmarked objects reachable only from stacks.
2) It flushes and disables per-P GC work caches so all work must be in
globally visible work buffers.
3) It rescans the global roots---the BSS and data segments---so there
are fewer objects to blacken during mark termination. We do not rescan
stacks at this point, though that could be done in a later CL.
After these steps, it again drains the global work buffers.
On a lightly loaded machine the garbage benchmark has reduced the
number of GC cycles with latency > 10 ms from 83 out of 4083 cycles
down to 2 out of 3995 cycles. Maximum latency was reduced from
60+ msecs down to 20 ms.
Change-Id: I152285b48a7e56c5083a02e8e4485dd39c990492
Reviewed-on: https://go-review.googlesource.com/10590
Reviewed-by: Austin Clements <austin@google.com>
These were found by grepping the comments from the go code and feeding
the output to aspell.
Change-Id: Id734d6c8d1938ec3c36bd94a4dbbad577e3ad395
Reviewed-on: https://go-review.googlesource.com/10941
Reviewed-by: Aamir Khan <syst3m.w0rm@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Prior to this CL whenever the GC marking was enabled and
a P was looking for work we supplied a G to help
the GC do its marking tasks. Once this G finished all
the marking available it would release the P to find another
available G. In the case where there was no work the P would drop
into findrunnable which would execute the mark helper G which would
immediately return and the P would drop into findrunnable again repeating
the process. Since the P was always given a G to run it never blocks.
This CL first checks if the GC mark helper G has available work and if
not the P immediately falls through to its blocking logic.
Fixes#10901
Change-Id: I94ac9646866ba64b7892af358888bc9950de23b5
Reviewed-on: https://go-review.googlesource.com/10189
Reviewed-by: Austin Clements <austin@google.com>
Currently, each M has a cache of the most recently used *workbuf. This
is used primarily by the write barrier so it doesn't have to access
the global workbuf lists on every write barrier. It's also used by
stack scanning because it's convenient.
This cache is important for write barrier performance, but this
particular approach has several downsides. It's faster than no cache,
but far from optimal (as the benchmarks below show). It's complex:
access to the cache is sprinkled through most of the workbuf list
operations and it requires special care to transform into and back out
of the gcWork cache that's actually used for scanning and marking. It
requires atomic exchanges to take ownership of the cached workbuf and
to return it to the M's cache even though it's almost always used by
only the current M. Since it's per-M, flushing these caches is O(# of
Ms), which may be high. And it has some significant subtleties: for
example, in general the cache shouldn't be used after the
harvestwbufs() in mark termination because it could hide work from
mark termination, but stack scanning can happen after this and *will*
use the cache (but it turns out this is okay because it will always be
followed by a getfull(), which drains the cache).
This change replaces this cache with a per-P gcWork object. This
gcWork cache can be used directly by scanning and marking (as long as
preemption is disabled, which is a general requirement of gcWork).
Since it's per-P, it doesn't require synchronization, which simplifies
things and means the only atomic operations in the write barrier are
occasionally fetching new work buffers and setting a mark bit if the
object isn't already marked. This cache can be flushed in O(# of Ps),
which is generally small. It follows a simple flushing rule: the cache
can be used during any phase, but during mark termination it must be
flushed before allowing preemption. This also makes the dispose during
mutator assist no longer necessary, which eliminates the vast majority
of gcWork dispose calls and reduces contention on the global workbuf
lists. And it's a lot faster on some benchmarks:
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 11963668673 11206112763 -6.33%
BenchmarkFannkuch11 2643217136 2649182499 +0.23%
BenchmarkFmtFprintfEmpty 70.4 70.2 -0.28%
BenchmarkFmtFprintfString 364 307 -15.66%
BenchmarkFmtFprintfInt 317 282 -11.04%
BenchmarkFmtFprintfIntInt 512 483 -5.66%
BenchmarkFmtFprintfPrefixedInt 404 380 -5.94%
BenchmarkFmtFprintfFloat 521 479 -8.06%
BenchmarkFmtManyArgs 2164 1894 -12.48%
BenchmarkGobDecode 30366146 22429593 -26.14%
BenchmarkGobEncode 29867472 26663152 -10.73%
BenchmarkGzip 391236616 396779490 +1.42%
BenchmarkGunzip 96639491 96297024 -0.35%
BenchmarkHTTPClientServer 100110 70763 -29.31%
BenchmarkJSONEncode 51866051 52511382 +1.24%
BenchmarkJSONDecode 103813138 86094963 -17.07%
BenchmarkMandelbrot200 4121834 4120886 -0.02%
BenchmarkGoParse 16472789 5879949 -64.31%
BenchmarkRegexpMatchEasy0_32 140 140 +0.00%
BenchmarkRegexpMatchEasy0_1K 394 394 +0.00%
BenchmarkRegexpMatchEasy1_32 120 120 +0.00%
BenchmarkRegexpMatchEasy1_1K 621 614 -1.13%
BenchmarkRegexpMatchMedium_32 209 202 -3.35%
BenchmarkRegexpMatchMedium_1K 54889 55175 +0.52%
BenchmarkRegexpMatchHard_32 2682 2675 -0.26%
BenchmarkRegexpMatchHard_1K 79383 79524 +0.18%
BenchmarkRevcomp 584116718 584595320 +0.08%
BenchmarkTemplate 125400565 109620196 -12.58%
BenchmarkTimeParse 386 387 +0.26%
BenchmarkTimeFormat 580 447 -22.93%
(Best out of 10 runs. The delta of averages is similar.)
This also puts us in a good position to flush these caches when
nearing the end of concurrent marking, which will let us increase the
size of the work buffers while still controlling mark termination
pause time.
Change-Id: I2dd94c8517a19297a98ec280203cccaa58792522
Reviewed-on: https://go-review.googlesource.com/9178
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
This tracks the amount of scan work in terms of scanned pointers
during the concurrent mark phase. We'll use this information to
estimate scan work for the next cycle.
Currently this aggregates the work counter in gcWork and dispose
atomically aggregates this into a global work counter. dispose happens
relatively infrequently, so the contention on the global counter
should be low. If this turns out to be an issue, we can reduce the
number of disposes, and if it's still a problem, we can switch to
per-P counters.
Change-Id: Iac0364c466ee35fab781dbbbe7970a5f3c4e1fc1
Reviewed-on: https://go-review.googlesource.com/8832
Reviewed-by: Rick Hudson <rlh@golang.org>
This tracks the number of heap bytes marked by a GC cycle. We'll use
this information to precisely trigger the next GC cycle.
Currently this aggregates the work counter in gcWork and dispose
atomically aggregates this into a global work counter. dispose happens
relatively infrequently, so the contention on the global counter
should be low. If this turns out to be an issue, we can reduce the
number of disposes, and if it's still a problem, we can switch to
per-P counters.
Change-Id: I1bc377cb2e802ef61c2968602b63146d52e7f5db
Reviewed-on: https://go-review.googlesource.com/8388
Reviewed-by: Russ Cox <rsc@golang.org>
Currently, gcDrainN is documented saying that it must be run on the
system stack. In fact, the problem and solution here are somewhat
subtler. First, it doesn't have to happen on the system stack, it just
has to be non-stoppable (that is, non-preemptible). Second, this isn't
specific to gcDrainN (though gcDrainN is perhaps the most surprising
instance); it's general to anything that uses the gcWork structure.
Move the comment to gcWork and generalize it.
Change-Id: I5277b5abb070e47f8d783bc15a310b379c6adc22
Reviewed-on: https://go-review.googlesource.com/8247
Reviewed-by: Rick Hudson <rlh@golang.org>
Currently, we only exit the getfull barrier if there is work on the
full list, even though the exit path will take work from either the
full or partial list. Change this to exit the barrier if there is work
on either the full or partial lists.
I believe it's currently safe to check only the full list, since
during mark termination there is no reason to put a workbuf on a
partial list. However, checking both is more robust.
Change-Id: Icf095b0945c7cad326a87ff2f1dc49b7699df373
Reviewed-on: https://go-review.googlesource.com/7840
Reviewed-by: Rick Hudson <rlh@golang.org>
The distinction between gcWorkProducer and gcWork (producer and
consumer) is not serving us as originally intended, so merge these
into just gcWork.
The original intent was to replace the currentwbuf cache with a
gcWorkProducer. However, with gchelpwork (aka mutator assists),
mutators can both produce and consume work, so it will make more sense
to cache a whole gcWork.
Change-Id: I6e633e96db7cb23a64fbadbfc4607e3ad32bcfb3
Reviewed-on: https://go-review.googlesource.com/7733
Reviewed-by: Rick Hudson <rlh@golang.org>
Until recently, struct workbuf had only lfnode and uintptr fields
before the obj array to make it convenient to compute the size of the
obj array. It slowly grew more fields until this became inconvenient
enough that it was restructured to make the size computation easy.
Now the size computation doesn't care what the field types are, so
switch to more natural types.
Change-Id: I966140ba7ebb4aeb41d5c66d9d2a3bdc17dd4bcf
Reviewed-on: https://go-review.googlesource.com/5262
Reviewed-by: Russ Cox <rsc@golang.org>
This introduces a producer/consumer abstraction for GC work pointers
that internally handles the details of filling, draining, and
shuffling work buffers.
In addition to simplifying the GC code, this should make it easy for
us to change how we use work buffers, including cleaning up how we use
the work.partial queue, reintroducing a FIFO lookahead cache, adding
prefetching, and using dual buffers to avoid flapping.
This commit doesn't change any existing code. The following commit
will switch the garbage collector from explicit workbuf manipulation
to gcWork.
Change-Id: Ifbfe5fff45bf0362d6d7c3cecb061f0c9874077d
Reviewed-on: https://go-review.googlesource.com/5231
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
All of the other memory-related source files start with "m". Keep up
the tradition.
Change-Id: Idd88fdbf2a1453374fa12109b949b1c4d149a4f8
Reviewed-on: https://go-review.googlesource.com/4853
Reviewed-by: Minux Ma <minux@golang.org>