David Crawshaw [Sun, 21 Feb 2016 03:54:15 +0000 (22:54 -0500)]
cmd/compile: remove rtype *uncommonType field
Instead of a pointer on every rtype, use a bit flag to indicate that
the contents of uncommonType directly follows the rtype value when it
is needed.
This requires a bit of juggling in the compiler's rtype encoder. The
backing arrays for fields in the rtype are presently encoded directly
after the slice header. This packing requires separating the encoding
of the uncommonType slice headers from their backing arrays.
Reduces binary size of godoc by ~180KB (1.5%).
No measurable change in all.bash time.
For #6853.
David Chase [Tue, 8 Mar 2016 20:08:25 +0000 (15:08 -0500)]
cmd/compile: guard the &-to-<<>> opt against small constants
Converting an and-K into a pair of shifts for K that will
fit in a one-byte argument is probably not an optimization,
and it also interferes with other patterns that we want to
see fire, like (<< (AND K)) [for small K] and bounds check
elimination for masked indices.
Turns out that on Intel, even 32-bit signed immediates beat
the shift pair; the size reduction of tool binaries is 0.09%
vs 0.07% for only the 8-bit immediates.
RLH found this one working on the new/next GC.
Change-Id: I2414a8de1dd58d680d18587577fbadb7ff4f67d9
Reviewed-on: https://go-review.googlesource.com/20410 Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
Alexandru Moșoi [Mon, 7 Mar 2016 18:29:15 +0000 (19:29 +0100)]
cmd/compile/internal/ssa: simplify nil checks in opt.
* Simplify the nilcheck generated by
for _, e := range a {}
* No effect on the generated code because these nil checks
don't end up in the generated code.
* Useful for other analysis, e.g. it'll remove one dependecy
on the induction variable.
Change-Id: I6ee66ddfdc010ae22aea8dca48163303d93de7a9
Reviewed-on: https://go-review.googlesource.com/20307
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Matthew Dempsky [Tue, 8 Mar 2016 06:05:49 +0000 (22:05 -0800)]
cmd/compile: cleanup compile function
Make more idiomatic with a defer cleanup, which allows declaring
variables closer to their first use, rather than up front before the
first goto statement.
Also, split the legacy code generation code path into a separate
genlegacy function, analogous to the new genssa.
Brad Fitzpatrick [Tue, 8 Mar 2016 18:57:19 +0000 (18:57 +0000)]
cmd/compile: fix old comment references to go.y
Change-Id: Iec323998133ef20ff962f06a46d15dd342a82f5f
Reviewed-on: https://go-review.googlesource.com/20390 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Ross Light [Tue, 8 Mar 2016 00:21:30 +0000 (16:21 -0800)]
os/user: make OS-specific getgrouplist calls
getgrouplist is non-standard and has slightly different semantics on
each platform. Darwin defines the function in terms of ints instead of
gid_ts. Solaris only recently supported the call, so stubbing out for
now.
Fixes #14696
Fixes #14709
Change-Id: I5a44538d41594909efb6f3f9610c55d638c36757
Reviewed-on: https://go-review.googlesource.com/20348
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Thu, 28 Jan 2016 16:44:07 +0000 (11:44 -0500)]
cmd/go: ignore C files when CGO_ENABLED=0
Before, those C files might have been intended for the Plan 9 C compiler,
but that option was removed in Go 1.5. We can simplify the maintenance
of cgo packages now if we assume C files (and C++ and M and SWIG files)
should only be considered when cgo is enabled.
Also remove newly unnecessary build tags in runtime/cgo's C files.
Fixes #14123
Change-Id: Ia5a7fe62b9469965aa7c3547fe43c6c9292b8205
Reviewed-on: https://go-review.googlesource.com/19613 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Caio Marcelo de Oliveira Filho [Mon, 29 Feb 2016 20:46:48 +0000 (17:46 -0300)]
net/http/httptest: record trailing headers in ResponseRecorder
Trailers() returns the headers that were set by the handler after the
headers were written "to the wire" (in this case HeaderMap) and that
were also specified in a proper header called "Trailer".
Neither HeaderMap or trailerMap (used for Trailers()) are manipulated by
the handler code, instead a third stagingMap is given to the
handler. This avoid a reference kept by handler to affect the recorded
results.
If a handler just modify the header but doesn't call any Write or Flush
method from ResponseWriter (or Flusher) interface, HeaderMap will not be
updated. In this case, calling Flush in the recorder is enough to get
the HeaderMap filled.
Deleting the string merging pass makes the linker 30-35% faster
but makes jujud (using the github.com/davecheney/benchjuju snapshot) 2.5% larger.
Two optimizations bring the space overhead down to 0.6%.
First, change the default alignment for string data to 1 byte.
(It was previously defaulting to larger amounts, usually pointer width.)
Second, write out the type string for T (usually a bigger expression) as "*T"[1:],
so that the type strings for T and *T share storage.
Combined, these obtain the bulk of the benefit of string merging
at essentially no cost. The remaining benefit from string merging
is not worth the excessive cost, so delete it.
As penance for making the jujud binary 0.6% larger,
the next CL in this sequence trims the reflect functype
information enough to make the jujud binary overall 0.75% smaller
(that is, that CL has a net -1.35% effect).
Russ Cox [Tue, 8 Mar 2016 17:47:49 +0000 (12:47 -0500)]
cmd/newlink: delete
cmd/link is clearly the way forward.
The original rationale for cmd/newlink was that it would be a clean Go reimplementation.
But when push came to shove, cmd/link got converted from C instead,
and all the work on build modes and the like is in cmd/link now.
Cleaning up cmd/link is likely a much better plan.
This directory is something to delete from releases and the
testdata is something that breaks every time the .6 format changes.
Fix both problems by just deleting it outright.
Brad Fitzpatrick [Tue, 8 Mar 2016 15:58:20 +0000 (15:58 +0000)]
os/user: skip Current test on android
Also, add more failure output to debug why linux/mips64le and
linux/ppc64 are failing. They should be working. I suspect their
builder test envs are missing something.
Change-Id: I97273fe72c4e3009db400394636d0da1ef147485
Reviewed-on: https://go-review.googlesource.com/20358
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Klaus Post [Tue, 8 Mar 2016 14:57:12 +0000 (15:57 +0100)]
hash/crc32: use slicing by 8 for Castagnoli and smaller sizes
This adds "slicing by 8" optimization to Castagnoli tables which will
speed up CRC32 calculation on systems without asssembler,
which are all but AMD64.
In my tests, it is faster to use "slicing by 8" for sizes all down to
16 bytes, so the switchover point has been adjusted.
There are no benchmarks for small sizes, so I have added one for 40 bytes,
as well as one for bigger sizes (32KB).
Aaron Zinman [Thu, 4 Feb 2016 01:36:15 +0000 (17:36 -0800)]
darwin/amd64: fix text-relocation issues for c-archive libraries
The existing implementation deals with absolute relocations in __TEXT
for darwin/amd64 in build-mode c-shared, but it ignores c-archive.
This results in issues when trying to use a c-archive in an iOS
app on the 64-bit simulator. This patch adds c-archive to the
handling of this issue.
Fixes #14217
Change-Id: I2e4d5193caa531171ad22fd0cd420a8bfb4646a6
Reviewed-on: https://go-review.googlesource.com/19206 Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
David Chase [Fri, 4 Mar 2016 19:19:49 +0000 (14:19 -0500)]
cmd/compile: Tinkering with schedule for debug and regalloc
This adds a heap-based proper priority queue to the
scheduler which made a relatively easy to test quite a few
heuristics that "ought to work well". For go tools
themselves (which may not be representative) the heuristic
that works best is (1) in line-number-order, then (2) from
more to fewer args, then (3) in variable ID order. Trying
to improve this with information about use at end of
blocks turned out to be fruitless -- all of my naive
attempts at using that information turned out worse than
ignoring it. I can confirm that the stores-early heuristic
tends to help; removing it makes the results slightly worse.
My metric is code size reduction, which I take to mean fewer
spills from register allocation. It's not uniform.
Here's the endpoints for "vet" from one set of pretty-good
heuristics (this is representative at least).
Overall there appears to be an 0.1% decrease in text size.
No timings yet, and given the distribution of size reductions
it might make sense to wait on those.
addr2line text (code) = -4392 bytes (-0.156273%)
api text (code) = -5502 bytes (-0.147644%)
asm text (code) = -5254 bytes (-0.187810%)
cgo text (code) = -4886 bytes (-0.148846%)
compile text (code) = -1577 bytes (-0.019346%) * changed
cover text (code) = -5236 bytes (-0.137992%)
dist text (code) = -5015 bytes (-0.167829%)
doc text (code) = -5180 bytes (-0.182121%)
fix text (code) = -5000 bytes (-0.215148%)
link text (code) = -5092 bytes (-0.152712%)
newlink text (code) = -5204 bytes (-0.196986%)
nm text (code) = -4398 bytes (-0.156018%)
objdump text (code) = -4582 bytes (-0.155046%)
pack text (code) = -4503 bytes (-0.294287%)
pprof text (code) = -6314 bytes (-0.085177%)
trace text (code) = -5856 bytes (-0.097818%)
vet text (code) = -5696 bytes (-0.117334%)
yacc text (code) = -4971 bytes (-0.213817%)
This leaves me sorely tempted to look into a "real" scheduler
to try to do a better job, but I think it might make more
sense to look into getting loop information into the
register allocator instead.
Fixes #14577.
Change-Id: I5238b83284ce76dea1eb94084a8cd47277db6827
Reviewed-on: https://go-review.googlesource.com/20240
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Matthew Dempsky [Mon, 7 Mar 2016 23:15:57 +0000 (15:15 -0800)]
cmd/internal/obj: stop using as+ALAST as an opcode
Currently, package obj reserves a range of 1<<12 opcodes for each
target architecture. E.g., mips64 has [6<<12, 7<<12).
However, because mips.ABEQ and mips.ALAST are both within that range,
the expression mips.ABEQ+mips.ALAST in turn falls (far) outside that
range around 12<<12, meaning it could theoretically collide with
another arch's opcodes.
More practically, it's a problem because 12<<12 overflows an int16,
which hampers fixing #14692. (We could also just switch to uint16 to
avoid the overflow, but that still leaves the first problem.)
As a workaround, use Michael Hudson-Doyle's solution from
https://golang.org/cl/20182 and use negative values for these variant
instructions.
Passes toolstash -cmp for GOARCH=arm and GOARCH=mips64.
Caio Marcelo de Oliveira Filho [Mon, 29 Feb 2016 15:06:57 +0000 (12:06 -0300)]
net/http: TimeoutHandler should start timer when serving request
TimeoutHandler was starting the Timer when the handler was created,
instead of when serving a request. It also was sharing it between
multiple requests, which is incorrect, as the requests might start
at different times.
Store the timeout duration and create the Timer when ServeHTTP is
called. Different requests will have different timers.
The testing plumbing was simplified to store the channel used to
control when timeout happens. It overrides the regular timer.
It is only necessary in a few places, and this inlining will
simplify the transition away from NodeLists.
Passes toolstash -cmp.
Change-Id: I4ee9b4bf56ffa04df23e20a0a83b302d36b33510
Reviewed-on: https://go-review.googlesource.com/20290 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Konstantin Shaposhnikov [Tue, 1 Mar 2016 14:55:06 +0000 (22:55 +0800)]
cmd/vet: remove -test flag
-test flag is a testing only flag that enables all vet checks. It was needed
because there was no way to run all vet checks in a single command
invocation. However it is possible to do this now by combining -all and -shadow
flags.
Also a recently added -tests flag is similarly named, having both -test and
-tests can be confusing.
Change-Id: Ie5bacbe0bef5c8409eeace46f16141fa4e782c32
Reviewed-on: https://go-review.googlesource.com/20006 Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Brad Fitzpatrick [Mon, 7 Mar 2016 22:11:48 +0000 (22:11 +0000)]
os, syscall: skip tests when environment doesn't permit testing
Fixes #14693
Change-Id: Id0a6a80b4c37c0b0f1c2755667b7233ed8964e40
Reviewed-on: https://go-review.googlesource.com/20342 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Austin Clements [Mon, 7 Mar 2016 22:10:19 +0000 (17:10 -0500)]
runtime: fix checkmark scanning of finalizers
Currently work.finalizersDone is reset only at the beginning of
gcStart. As a result, it will be set when checkmark runs, so checkmark
will skip scanning finalizers. Hence, if there are any bugs that cause
the regular scan of finalizers to miss pointers, checkmark will also
miss them and fail to detect the missed pointer.
Fix this by resetting finalizersDone in gcResetMarkState. This way it
gets reset before any full mark, which is exactly what we want.
Caio Marcelo de Oliveira Filho [Sat, 5 Mar 2016 18:57:17 +0000 (15:57 -0300)]
go/types: don't emit conversion error in non-numeric increment/decrement
In increment and decrement statements, explicit check that the type
of operand is numeric. This avoids a related but less clear error
about converting "1" to be emitted.
So, when checking
package main
func main() {
var x bool
x++
}
instead of emitting the error
prog.go:5:2: cannot convert 1 (untyped int constant) to bool
emits
prog.go:5:2: invalid operation: x++ (non-numeric type bool).
Updates #12525.
Change-Id: I00aa6bd0bb23267a2fe10ea3f5a0b20bbf3552bc
Reviewed-on: https://go-review.googlesource.com/20244 Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Brad Fitzpatrick [Thu, 3 Mar 2016 20:54:24 +0000 (20:54 +0000)]
os/user: make Current work without cgo
Fixes #14626
Change-Id: I91c40407dc35355e5c5046f24111a126f99260d9
Reviewed-on: https://go-review.googlesource.com/20192 Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Minux Ma <minux@golang.org>
Ross Light [Thu, 4 Feb 2016 23:39:00 +0000 (15:39 -0800)]
os/user: add LookupGroup, LookupGroupId, and User.GroupIds functions
As part of local testing with a large group member list, I discovered
that the lookup functions don't resize their buffer if they receive
ERANGE. I fixed this as a side-effect of this CL.
Thanks to @andrenth for the original CL.
Fixes #2617
Change-Id: Ie6aae2fe0a89eae5cce85786869a8acaa665ffe9
Reviewed-on: https://go-review.googlesource.com/19235 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Brad Fitzpatrick [Mon, 7 Mar 2016 16:35:45 +0000 (08:35 -0800)]
mime/multipart: don't call Read on io.Reader after an error is seen
The io.Reader contract makes no promises about how a Reader should
behave after it returns its first error. Usually the errors are
sticky, but they don't have to be. A regression in zlib.Reader (bug
accidentally relied on sticky errors.
Minimal fix: wrap the user's provided Reader in a Reader which
guarantees stickiness. The minimal fix is less scary than touching
the multipart state machine.
Joe Tsai [Sun, 6 Mar 2016 23:51:57 +0000 (15:51 -0800)]
compress/zlib: make errors persistent
Ensure that all errors (including io.EOF) are persistent across method
calls on zlib.Reader. Furthermore, ensure that these persistent errors
are properly cleared when Reset is called.
Richard Miller [Wed, 27 Jan 2016 18:43:36 +0000 (18:43 +0000)]
runtime: signal handling support for plan9_arm
Plan 9 trap/signal handling differs on ARM from other architectures
because ARM has a link register. Also trap message syntax varies
between different architectures (historical accident?).
Revised 7 March to clarify a comment.
Caio Marcelo de Oliveira Filho [Sat, 5 Mar 2016 20:36:56 +0000 (17:36 -0300)]
cmd/compile: don't emit conversion error in non-numeric increment/decrement
In increment and decrement statements, explicit check that the type
of operand is numeric earlier. This avoids a related but less clear
error about converting "1" to be emitted.
So, when compiling
package main
func main() {
var x bool
x++
}
instead of emitting two errors
prog.go:5: cannot convert 1 to type bool
prog.go:5: invalid operation: x++ (non-numeric type bool)
just emits the second error.
Fixes #12525.
Change-Id: I6e81330703765bef0d6eb6c57098c1336af7c799
Reviewed-on: https://go-review.googlesource.com/20245 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Austin Clements [Wed, 2 Mar 2016 22:27:59 +0000 (17:27 -0500)]
runtime: merge {bgMark,assist}StartTime
We used to start background mark workers and assists at different
times, so we needed to keep track of these separately. They're now set
to exactly the same time, so clean things up by merging them in to one
value, markStartTime.
With this change, the OffPtr collapsing opt
rule matches increase from 160k to 263k,
and the Load-after-Store opt rule matches
increase from 217 to 853.
Keith Randall [Fri, 4 Mar 2016 18:26:57 +0000 (10:26 -0800)]
cmd/compile: Combine smaller loads into a larger load
This only deals with the loads themselves. The bounds checks
are a separate issue. Also doesn't handle stores, those are
harder because we need to make sure intermediate memory states
aren't observed (which is hard to do with rewrite rules).
Use one byte shorter instructions for zero-extending loads.
Update #14267
Change-Id: I40af25ab5208488151ba7db32bf96081878fa7d9
Reviewed-on: https://go-review.googlesource.com/20218 Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Michael Hudson-Doyle [Sun, 6 Mar 2016 20:47:02 +0000 (09:47 +1300)]
cmd/compile, cmd/link, cmd/newlink: remove support for weak symbols
They were only used for rtype.ptrToThis which David Crawshaw removed a couple
of weeks ago. Removes two traversals of Ctxt.Allsym from the linker but it
doesn't seem to make much difference to performance.
The new code is a bit less efficient,
but it does not involve altering the structure
of any linked lists.
This will make it easier to replace NodeLists
with Node slices.
We can return to a more efficient algorithm
when NodeLists have been replaced.
Martin Möhrmann [Sat, 27 Feb 2016 13:37:10 +0000 (14:37 +0100)]
fmt: refactor and unify float and complex formatting
Removes specialized functions for each verb and float/complex size
and replaces them with generic variants fmtFloat and
fmtComplex similar to other generic fmt functions.
Simplifies the complex formatting by relying on fmtFloat
to handle the verb and default precision selection.
Complex imaginary formatting does not need to clear the f.space flag
because the set f.plus flag will force a sign instead of a space.
Sets default precision for %b to -1 (same as %g and %G)
since precision for %b has no affect in strconv.AppendFloat.
Add more tests and group them a bit better.
Use local copies of +Inf,-Inf and NaN instead
of math package functions for testing.
Saves around 8kb in the go binary.
name old time/op new time/op delta
SprintfFloat-2 200ns ± 4% 196ns ± 4% -1.55% (p=0.007 n=20+20)
SprintfComplex-2 569ns ± 4% 570ns ± 3% ~ (p=0.804 n=20+20)
Change-Id: I36d35dab6f835fc2bd2c042ac97705868eb2446f
Reviewed-on: https://go-review.googlesource.com/20252 Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Rob Pike <r@golang.org>
Martin Möhrmann [Thu, 3 Mar 2016 16:06:43 +0000 (17:06 +0100)]
fmt: optimize byte slice and array formatting for %v and %d
Instead of calling printArg in fmtBytes to format each byte call
the byte formatting functions directly since it is known each
element is of type byte.
Add more tests for byte slice and array formatting.
name old time/op new time/op delta
SprintfBytes-2 843ns ±16% 417ns ±11% -50.58% (p=0.000 n=20+20)
Change-Id: I5b907dbf52091e3de9710b09d67649c76f4c17e9
Reviewed-on: https://go-review.googlesource.com/20176
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rob Pike <r@golang.org>
Matthew Dempsky [Sat, 5 Mar 2016 00:39:07 +0000 (16:39 -0800)]
cmd/compile: stop leaking 'any' type into user package
The new check corresponds to the (etype != TANY || Debug['A'] != 0)
that was lost in golang.org/cl/19936.
Fixes #14652.
Change-Id: Iec3788ff02529b3b0f0d4dd92ec9f3ef20aec849
Reviewed-on: https://go-review.googlesource.com/20271 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Lynn Boger [Tue, 16 Feb 2016 22:05:56 +0000 (16:05 -0600)]
crypto/cipher: xorBytes performance ppc64le/ppc64
Update supportsUnaligned in xor.go to be true for
GOARCH values ppc64le and ppc64. This allows the
xor of long buffers to be done on double words
(8 bytes) instead of a single byte at a time, which
significantly improves performance.
Fixes #14350
Change-Id: Iccc6b9d3df2e604a55f4c1e4890bdd3bb0d77ab0
Reviewed-on: https://go-review.googlesource.com/19519 Reviewed-by: Minux Ma <minux@golang.org>
Rhys Hiltner [Fri, 4 Mar 2016 06:37:14 +0000 (22:37 -0800)]
cmd/link/internal/ld: don't panic on short buildid
When the linker was written in C, command line arguments were passed
around as null-terminated byte arrays which encouraged checking
characters one at a time. In Go, that can easily lead to
out-of-bounds panics.
Use the more idiomatic strings.HasPrefix when checking cmd/link's -B
argument to avoid the panic, and replace the manual hex decode with
use of the encoding/hex package.
Austin Clements [Wed, 2 Mar 2016 02:30:26 +0000 (21:30 -0500)]
runtime: disable gcMarkRootCheck debugging check during STW
gcMarkRootCheck takes ~10ns per goroutine. This is just a debugging
check, so disable it (plus, if something is going to go wrong, it's
more likely to go wrong during concurrent mark).
We may be able to re-enable this later, or move it to after we've
started the world again. (But not for 1.6.x.)
For 1.6.x.
Fixes #14419.
name / 95%ile-time/markTerm old new delta
500kIdleGs-12 24.0ms ± 0% 18.9ms ± 6% -21.46% (p=0.000 n=15+20)
Austin Clements [Tue, 1 Mar 2016 20:09:24 +0000 (15:09 -0500)]
runtime: reset mark state before stopping the world
Currently we reset the mark state during STW sweep termination. This
involves looping over all of the goroutines. Each iteration of this
loop takes ~25ns, so at around 400k goroutines, we'll exceed our 10ms
pause goal.
However, it's safe to do this before we stop the world for sweep
termination because nothing is consuming this state yet. Hence, move
the reset to just before STW.
This isn't perfect: a long reset can still delay allocating goroutines
that block on GC starting. But it's certainly better to block some
things eventually than to block everything immediately.
For 1.6.x.
Fixes #14420.
name \ 95%ile-time/sweepTerm old new delta
500kIdleGs-12 11312µs ± 6% 18.9µs ± 6% -99.83% (p=0.000 n=16+20)
David Crawshaw [Sat, 27 Feb 2016 17:33:14 +0000 (08:33 -0900)]
cmd/link: deduplicate read-only string data
Many read-only strings in Go binaries are substrings of other read-only
strings. A common source is the text form of type information, which
will include both "struct { X int }" and "*struct { X int }" or
"*bytes.Reader" and "func(*bytes.Reader)" in the same binary.
Because this character data is referred to by separate string headers,
we can skip writing the smaller string and modify the pointer
relocation to point to the larger string. This CL does this
deduplication in the linker after the reachable set of strings has
been determined.
This removes 765KB from juju (1.4% without DWARF).
Link time goes at tip goes form 4.6s to 6.3s, but note that this CL
is part of a series that recently reduced link time from 9.6s.
For #6853.
Change-Id: Ib2087cf627c9f1e9a1181f9b4c8f81d1a3f42191
Reviewed-on: https://go-review.googlesource.com/19987 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Fri, 4 Mar 2016 15:55:39 +0000 (07:55 -0800)]
cmd/compile: small cleanup to syslook calls
Passing copy==1 to syslook is only necessary to support subsequent
calls to substArgTypes. typ2Itab and concatstring* don't have "any"
parameters, so no point in deep copying their function signatures at
every call site.
For a couple other syslook calls (makemap and conv[IET]2[IET]), move
them closer to their corresponding substArgTypes calls so it's easier
to see that all syslook(fn, 1) calls are necessary.
Change-Id: I4a0588ab2b8b5b8ce7a0a44b24c8cf8fda489af6
Reviewed-on: https://go-review.googlesource.com/20215
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Fri, 4 Mar 2016 01:39:55 +0000 (17:39 -0800)]
math/big: use correct precision in Float.Float32/64 for denormals
When a big.Float is converted to a denormal float32/64, the rounding
precision depends on the size of the denormal. Rounding may round up
and thus change the size (exponent) of the denormal. Recompute the
correct precision again for correct placement of the mantissa.
Fixes #14553.
Change-Id: Iedab5810a2d2a405cc5da28c6de7be34cb035b86
Reviewed-on: https://go-review.googlesource.com/20198 Reviewed-by: Alan Donovan <adonovan@google.com>