]> Cypherpunks repositories - gostls13.git/log
gostls13.git
8 years agocmd/compile: make setting and accessing of node slice elements more uniform
Martin Möhrmann [Sun, 19 Feb 2017 14:57:58 +0000 (15:57 +0100)]
cmd/compile: make setting and accessing of node slice elements more uniform

Add Set3 function to complement existing Set1 and Set2 functions.
Consistently use Set1, Set2 and Set3 for []*Node instead of Set where applicable.

Add SetFirst and SetSecond for setting elements of []*Node to mirror
First and Second for accessing elements in []*Node.

Replace uses of Index by First and Second and
SetIndex with SetFirst and SetSecond where applicable.

Passes toolstash -cmp.

Change-Id: I8255aae768cf245c8f93eec2e9efa05b8112b4e5
Reviewed-on: https://go-review.googlesource.com/37430
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: speed up TestAssembly
Lorenzo Masini [Mon, 20 Feb 2017 16:17:28 +0000 (17:17 +0100)]
cmd/compile: speed up TestAssembly

TestAssembly was very slow, leading to it being skipped by default.
This is not surprising, it separately invoked the compiler and
parsed the result many times.

Now the test assembles one source file for arch/os combination,
containing the relevant functions.

Tests for each arch/os run in parallel.

Now the test runs approximately 10x faster on my Intel(R) Core(TM)
i5-6600 CPU @ 3.30GHz.

Fixes #18966

Change-Id: I45ab97630b627a32e17900c109f790eb4c0e90d9
Reviewed-on: https://go-review.googlesource.com/37270
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agotest: deflake locklinear a little
Russ Cox [Fri, 24 Feb 2017 21:01:50 +0000 (16:01 -0500)]
test: deflake locklinear a little

This should help on the openbsd systems where the test mostly passes.

I don't expect it to help on s390x where the test reliably fails.
But it should give more information when it does fail.

For #19276.

Change-Id: I496c291f2b4b0c747b8dd4315477d87d03010059
Reviewed-on: https://go-review.googlesource.com/37348
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/internal/browser: fix typo
Kevin Burke [Fri, 24 Feb 2017 20:07:25 +0000 (12:07 -0800)]
cmd/internal/browser: fix typo

Change-Id: I3c31f10c1082c7bc57aac18856014c55f79e0fed
Reviewed-on: https://go-review.googlesource.com/37409
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: check that pprof accepts but doesn't need executable
Russ Cox [Fri, 17 Feb 2017 20:27:12 +0000 (15:27 -0500)]
runtime: check that pprof accepts but doesn't need executable

The profiles are self-contained now.
Check that they work by themselves in the tests that invoke pprof,
but also keep checking that the old command lines work.

Change-Id: I24c74b5456f0b50473883c3640625c6612f72309
Reviewed-on: https://go-review.googlesource.com/37166
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
8 years agoruntime/pprof/internal/profile: move internal/pprof/profile here
Russ Cox [Fri, 17 Feb 2017 05:17:26 +0000 (00:17 -0500)]
runtime/pprof/internal/profile: move internal/pprof/profile here

Nothing needs internal/pprof anymore except the runtime/pprof tests.
Move the package here to prevent new dependencies.

Change-Id: Ia119af91cc2b980e0fa03a15f46f69d7f71d2926
Reviewed-on: https://go-review.googlesource.com/37165
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
8 years agoruntime/pprof: add streaming protobuf encoder
Russ Cox [Fri, 17 Feb 2017 05:10:39 +0000 (00:10 -0500)]
runtime/pprof: add streaming protobuf encoder

The existing code builds a full profile in memory.
Then it translates that profile into a data structure (in memory).
Then it marshals that data structure into a protocol buffer (in memory).
Then it gzips that marshaled form into the underlying writer.
So there are three copies of the full profile data in memory
at the same time before we're done. This is obviously dumb.

This CL implements a fully streaming conversion from
the original in-memory profile to the underlying writer.
There is now only one copy of the profile in memory.

For the non-CPU profiles, this is optimal, since we have to
have a full copy in memory to start with.

For the CPU profiles, we could still try to bound the profile
size stored in memory and stream fragments out during
the actual profiling, as Go 1.7 did (with a simpler format),
but so far that hasn't been necessary.

Change-Id: Ic36141021857791bf0cd1fce84178fb5e744b989
Reviewed-on: https://go-review.googlesource.com/37164
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
8 years agomath/big: use math/bits where appropriate
Robert Griesemer [Thu, 16 Feb 2017 23:04:33 +0000 (15:04 -0800)]
math/big: use math/bits where appropriate

This change adds math/bits as a new dependency of math/big.

- use bits.LeadingZeroes instead of local implementation
  (they are identical, so there's no performance loss here)

- leave other functionality local (ntz, bitLen) since there's
  faster implementations in math/big at the moment

Change-Id: I1218aa8a1df0cc9783583b090a4bb5a8a145c4a2
Reviewed-on: https://go-review.googlesource.com/37141
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/pprof: vendor pprof from github.com/google/pprof
Raul Silvera [Fri, 10 Feb 2017 22:52:02 +0000 (14:52 -0800)]
cmd/pprof: vendor pprof from github.com/google/pprof

Import the github.com/google/pprof and github.com/ianlancetaylor/demangle
packages, without modification.

Build the golang version of pprof from cmd/pprof/pprof.go
by importing the packages from src/cmd/vendot/github.com/google/pprof

The versions upstreamed are:

github.com/ianlancetaylor/demangle 4883227f66371e02c4948937d3e2be1664d9be38
github.com/google/pprof            7eb5ba977f28f2ad8dd5f6bb82cc9b454e123cdc

Update misc/nacl/testzip.proto for new tests.

Change-Id: I076584856491353607a3b98b67d0ca6838be50d6
Reviewed-on: https://go-review.googlesource.com/36798
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
8 years agocmd/internal/browser: wait 3 seconds for non-zero exit codes
Chris Broadfoot [Wed, 22 Feb 2017 23:50:24 +0000 (15:50 -0800)]
cmd/internal/browser: wait 3 seconds for non-zero exit codes

Wait a short period between trying commands. Many commands
will return a non-zero exit code if the browser couldn't be launched.

For example, google-chrome returns quickly with a non-zero
exit code in a headless environment.

Updates #19131.

Change-Id: I0ae5356dd4447969d9e216615449cead7a8fd5c9
Reviewed-on: https://go-review.googlesource.com/37391
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: evaluate zero-sized values converted to interfaces
Josh Bleecher Snyder [Thu, 23 Feb 2017 05:19:29 +0000 (21:19 -0800)]
cmd/compile: evaluate zero-sized values converted to interfaces

CL 35562 substituted zerobase for the pointer for
interfaces containing zero-sized values.
However, it failed to evaluate the zero-sized value
expression for side-effects. Fix that.

The other similar interface value optimizations
are not affected, because they all actually use the
value one way or another.

Fixes #19246

Change-Id: I1168a99561477c63c29751d5cd04cf81b5ea509d
Reviewed-on: https://go-review.googlesource.com/37395
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/compile/internal/parser: improved a couple of error messages
Robert Griesemer [Wed, 22 Feb 2017 21:43:23 +0000 (13:43 -0800)]
cmd/compile/internal/parser: improved a couple of error messages

The new syntax tree introduced with 1.8 represents send statements
(ch <- x) as statements; the old syntax tree represented them as
expressions (and parsed them as such) but complained if they were
used in expression context. As a consequence, some of the errors
that in the past were of the form "ch <- x used as value" now look
like "unexpected <- ..." because a "<-" is not valid according to
Go syntax in those situations. Accept the new error message.

Also: Fine-tune handling of misformed for loop headers.

Also: Minor cleanups/better comments.

Fixes #17590.

Change-Id: Ia541dea1f2f015c1b21f5b3ae44aacdec60a8aba
Reviewed-on: https://go-review.googlesource.com/37386
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agovendor/golang_org/x/crypto/chacha20poly1305: revendor
Ian Lance Taylor [Fri, 24 Feb 2017 18:35:43 +0000 (10:35 -0800)]
vendor/golang_org/x/crypto/chacha20poly1305: revendor

Brings in chacha20poly1305 directory from golang.org/x/crypto revision
453249f01cfeb54c3d549ddb75ff152ca243f9d8, adding:

CL 35874: crypto/chacha20poly1305/internal/chacha20: add missing copyright header
CL 35875: crypto/chacha20poly1305: rename test vectors file

Fixes #19155.

Change-Id: I25cf83d060113f6b2a197f243a25614440008f7e
Reviewed-on: https://go-review.googlesource.com/37408
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agorun.bash: set GOPATH to $GOROOT/nil before running tests
Sean Christopherson [Wed, 22 Feb 2017 14:56:26 +0000 (06:56 -0800)]
run.bash: set GOPATH to $GOROOT/nil before running tests

Set $GOPATH to a semantically valid, non-empty string that cannot
conflict with $GOROOT to avoid false test failures that occur when
$GOROOT resides under $GOPATH.  Unsetting GOPATH is no longer viable
as Go now defines a default $GOPATH that may conflict with $GOROOT.

Fixes #19237

Change-Id: I376a2ad3b18e9c4098211b988dde7e76bc4725d2
Reviewed-on: https://go-review.googlesource.com/37396
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime/pprof: use more efficient hash table for staging profile
Russ Cox [Fri, 17 Feb 2017 02:13:15 +0000 (21:13 -0500)]
runtime/pprof: use more efficient hash table for staging profile

The old hash table was a place holder that allocates memory
during every lookup for key generation, even for keys that hit
in the the table.

Change-Id: I4f601bbfd349f0be76d6259a8989c9c17ccfac21
Reviewed-on: https://go-review.googlesource.com/37163
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
8 years agoruntime/pprof: use new profile buffers for CPU profiling
Russ Cox [Thu, 9 Feb 2017 20:57:57 +0000 (15:57 -0500)]
runtime/pprof: use new profile buffers for CPU profiling

This doesn't change the functionality of the current code,
but it sets us up for exporting the profiling labels into the profile.

The old code had a hash table of profile samples maintained
during the signal handler, with evictions going into a log.
The new code just logs every sample directly, leaving the
hash-based deduplication to an ordinary goroutine.

The new code also avoids storing the entire profile in two
forms in memory, an unfortunate regression introduced
when binary profile support was added. After this CL the
entire profile is only stored once in memory. We'd still like
to get back down to storing it zero times (streaming it to
the underlying io.Writer).

Change-Id: I0893a1788267c564aa1af17970d47377b2a43457
Reviewed-on: https://go-review.googlesource.com/36712
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
8 years agoruntime: do not allocate on every time.Sleep
Russ Cox [Fri, 17 Feb 2017 15:17:42 +0000 (10:17 -0500)]
runtime: do not allocate on every time.Sleep

It's common for some goroutines to loop calling time.Sleep.
Allocate once per goroutine, not every time.
This comes up in runtime/pprof's background reader.

Change-Id: I89d17dc7379dca266d2c9cd3aefc2382f5bdbade
Reviewed-on: https://go-review.googlesource.com/37162
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
8 years agocmd/doc: truncate long lists of arguments
Joe Tsai [Thu, 12 Jan 2017 00:53:49 +0000 (16:53 -0800)]
cmd/doc: truncate long lists of arguments

Some field-lists (especially in generated code) can be excessively long.
In the one-line printout, it does not make sense to print all elements
of the list if line-wrapping causes the "one-line" to become multi-line.

// Before:
var LongLine = newLongLine("someArgument1", "someArgument2", "someArgument3", "someArgument4", "someArgument5", "someArgument6", "someArgument7", "someArgument8")

// After:
var LongLine = newLongLine("someArgument1", "someArgument2", "someArgument3", "someArgument4", ...)

Change-Id: I4bbbe2dbd1d7be9f02d63431d213088c3dee332c
Reviewed-on: https://go-review.googlesource.com/36031
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
8 years agocrypto/tls: use io.ReadFull in conn_test.go
Joe Tsai [Fri, 24 Feb 2017 01:41:31 +0000 (17:41 -0800)]
crypto/tls: use io.ReadFull in conn_test.go

An io.Reader does not guarantee that it will read in the entire buffer.
To ensure that property, io.ReadFull should be used instead.

Change-Id: I0b863135ab9abc40e813f9dac07bfb2a76199950
Reviewed-on: https://go-review.googlesource.com/37403
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: repaired loop-finder to handle trickier nesting
David Chase [Thu, 23 Feb 2017 18:49:25 +0000 (13:49 -0500)]
cmd/compile: repaired loop-finder to handle trickier nesting

The loop-A-encloses-loop-C code did not properly handle the
case where really C was already known to be enclosed by B,
and A was nearest-outer to B, not C.

Fixes #19217.

Change-Id: I755dd768e823cb707abdc5302fed39c11cdb34d4
Reviewed-on: https://go-review.googlesource.com/37340
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agogo/types: fix doc string for Named.Obj
Robert Griesemer [Thu, 23 Feb 2017 20:52:33 +0000 (12:52 -0800)]
go/types: fix doc string for Named.Obj

Fixes #19249.

Change-Id: I6327192eca11fa24f1078c016c9669e4ba4bdb4e
Reviewed-on: https://go-review.googlesource.com/37399
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: silence superfluous assignment error message
David R. Jenni [Fri, 10 Feb 2017 18:25:58 +0000 (19:25 +0100)]
cmd/compile: silence superfluous assignment error message

Avoid printing a second error message when a field of an undefined
variable is accessed.

Fixes #8440.

Change-Id: I3fe0b11fa3423cec3871cb01b5951efa8ea7451a
Reviewed-on: https://go-review.googlesource.com/36751
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: new profile buffer implementation supporting label pointers
Russ Cox [Thu, 9 Feb 2017 18:58:48 +0000 (13:58 -0500)]
runtime: new profile buffer implementation supporting label pointers

The existing CPU profiling buffer is a slice of uintptr, but we want to
start including profiling label data in the profiles, and those labels need
to be pointers in order to let them describe rich information.

This CL implements a new profBuf type that holds both a slice of uint64
for data and a slice of unsafe.Pointer for profiling labels (aka tags).
Making the runtime use these buffers will happen in followup CLs.

Change-Id: I9ff16b532d8edaf4ce0cbba1098229a561834efc
Reviewed-on: https://go-review.googlesource.com/36713
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
8 years agowebsite: mention go1.8 in project page
Alberto Donizetti [Thu, 23 Feb 2017 18:36:34 +0000 (19:36 +0100)]
website: mention go1.8 in project page

Fixes #19253

Change-Id: Ia473f51bfe4cf42cf64938993a81d9b1dbc2594d
Reviewed-on: https://go-review.googlesource.com/37433
Reviewed-by: Chris Broadfoot <cbro@golang.org>
8 years agocmd/internal/browser: use xdg-open only from a desktop session
Chris Broadfoot [Wed, 22 Feb 2017 23:48:38 +0000 (15:48 -0800)]
cmd/internal/browser: use xdg-open only from a desktop session

xdg-open's man page says:
> xdg-open is for use inside a desktop session only.

Use the DISPLAY environment variable to detect this.

Updates #19131.

Change-Id: I3926b3e1042393939b2ec6aacd9b63ac8192df3b
Reviewed-on: https://go-review.googlesource.com/37390
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/vet/all: use -dolinkobj=false to speed up runs
Josh Bleecher Snyder [Wed, 22 Feb 2017 20:47:52 +0000 (12:47 -0800)]
cmd/vet/all: use -dolinkobj=false to speed up runs

When running on the host platform,
the standard library has almost certainly already been built.
However, all other platforms will probably need building.
Use the new -dolinkobj=false flag to cmd/compile
to only build the export data instead of doing a full compile.

Having partial object files could be confusing for people
doing subsequent cross-compiles, depending on what happens with #18369.
However, cmd/vet/all will mainly be run by builders
and core developers, who are probably fairly well-placed
to handle any such confusion.

This reduces the time on my machine for a cold run of
'go run main.go -all' by almost half:

benchmark           old ns/op        new ns/op        delta
BenchmarkVetAll     240670814551     130784517074     -45.66%

Change-Id: Ieb866ffb2cb714b361b0a6104077652f8eacd166
Reviewed-on: https://go-review.googlesource.com/37385
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: add -dolinkobj flag
Josh Bleecher Snyder [Wed, 22 Feb 2017 08:05:18 +0000 (00:05 -0800)]
cmd/compile: add -dolinkobj flag

When set to false, the -dolinkobj flag instructs the compiler
not to generate or emit linker information.

This is handy when you need the compiler's export data,
e.g. for use with go/importer,
but you want to avoid the cost of full compilation.

This must be used with care, since the resulting
files are unusable for linking.

This CL interacts with #18369,
where adding gcflags and ldflags to buildid has been mooted.
On the one hand, adding gcflags would make safe use of this
flag easier, since if the full object files were needed,
a simple 'go install' would fix it.
On the other hand, this would mean that
'go install -gcflags=-dolinkobj=false' would rebuild the object files,
although any existing object files would probably suffice.

Change-Id: I8dc75ab5a40095c785c1a4d2260aeb63c4d10f73
Reviewed-on: https://go-review.googlesource.com/37384
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agodoc: fix broken link in go1.8.html
Brad Fitzpatrick [Wed, 22 Feb 2017 22:23:35 +0000 (22:23 +0000)]
doc: fix broken link in go1.8.html

Fixes #19244

Change-Id: Ia6332941b229c83d6fd082af49f31003a66b90db
Reviewed-on: https://go-review.googlesource.com/37388
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/compile: rewrite pairs of shifts to extensions
Michael Munday [Mon, 20 Feb 2017 20:54:43 +0000 (15:54 -0500)]
cmd/compile: rewrite pairs of shifts to extensions

Replaces pairs of shifts with sign/zero extension where possible.

For example:
(uint64(x) << 32) >> 32 -> uint64(uint32(x))

Reduces the execution time of the following code by ~4.5% on s390x:

for i := 0; i < N; i++ {
        x += (uint64(i)<<32)>>32
}

Change-Id: Idb2d56f27e80a2e1366bc995922ad3fd958c51a7
Reviewed-on: https://go-review.googlesource.com/37292
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agosort: new example: Sorting slices with sort.SliceStable
Kenny Grant [Sat, 18 Feb 2017 16:56:45 +0000 (16:56 +0000)]
sort: new example: Sorting slices with sort.SliceStable

ExampleSliceStable echoes the sort.Slice example, to demonstrate sorting
on two fields together preserving order between sorts.

Change-Id: I8afc20c0203991bfd57260431eda73913c165355
Reviewed-on: https://go-review.googlesource.com/37196
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/internal/obj/ppc64: Fix RLDIMI
Carlos Eduardo Seo [Thu, 16 Feb 2017 16:58:47 +0000 (14:58 -0200)]
cmd/internal/obj/ppc64: Fix RLDIMI

Fix the encoding of the SH field for rldimi.

The SH field of rldimi is 6-bit wide and it is not contiguous in the instruction.
Bits 0-4 are placed in bit fields 16-20 in the instruction, while bit 5 is
placed in bit field 30. The current implementation does not consider this and,
therefore, any SH field between 32 and 63 are encoded wrongly in the instruciton.

Change-Id: I4d25a0a70f4219569be0e18160dea5505bd7fff0
Reviewed-on: https://go-review.googlesource.com/37350
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
8 years agonet: update IP.MarshalText documentation regarding len(ip) == 0
Yuval Pavel Zholkover [Wed, 22 Feb 2017 19:24:44 +0000 (21:24 +0200)]
net: update IP.MarshalText documentation regarding len(ip) == 0

Describe the difference from String encoding when len(ip) is zero.

Change-Id: Ia9b36b405d4fec3fee9a77498a839b6d90c2ec0d
Reviewed-on: https://go-review.googlesource.com/37379
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agomath: speed up and improve accuracy of Pow10
Martin Möhrmann [Sat, 10 Dec 2016 07:04:40 +0000 (08:04 +0100)]
math: speed up and improve accuracy of Pow10

Removes init function from the math package.

Allows stripping of arrays with pre-computed values
used for Pow10 from binaries if Pow10 is not used.
cmd/go shrinks by 128 bytes.

Fixed small values like 10**-323 being 0 instead of 1e-323.

Overall precision is increased but still not as good as
predefined constants for some inputs.

Samples:

Pow10(208)
before: 1.0000000000000006662e+208
after:  1.0000000000000000959e+208

Pow10(202)
before 1.0000000000000009895e+202
after  1.0000000000000001193e+202

Pow10(60)
before 1.0000000000000001278e+60
after  0.9999999999999999494e+60

Pow10(-100)
before 0.99999999999999938551e-100
after  0.99999999999999989309e-100

Pow10(-200)
before 0.9999999999999988218e-200
after  1.0000000000000001271e-200

name        old time/op  new time/op  delta
Pow10Pos-4  44.6ns ± 2%   1.2ns ± 1%  -97.39%  (p=0.000 n=19+17)
Pow10Neg-4  50.8ns ± 1%   4.1ns ± 2%  -92.02%  (p=0.000 n=17+19)

Change-Id: If094034286b8ac64be3a95fd9e8ffa3d4ad39b31
Reviewed-on: https://go-review.googlesource.com/36331
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/dist: fix negative test filtering
Michael Pratt [Wed, 22 Feb 2017 01:00:10 +0000 (17:00 -0800)]
cmd/dist: fix negative test filtering

std and race bench tests fail to check against t.runRxWant, so what
should be negative filters act as positive filters.

Fixes #19239

Change-Id: Icf02b2192bcd806a162fca9fb0af68a27ccfc936
Reviewed-on: https://go-review.googlesource.com/37336
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: suppress callsite signatures if any type is unknown
Emmanuel Odeke [Sat, 11 Feb 2017 06:41:53 +0000 (23:41 -0700)]
cmd/compile: suppress callsite signatures if any type is unknown

Fixes #19012.

Fallback to return signatures without detailed types.
These error message will be of the form of issue:
* https://golang.org/issues/4215
* https://golang.org/issues/6750

So:
func f(x int, y uint) {
    return x > y
}

f(10, "a" < 3)

will give errors:
too many errors to return
too many arguments in call to f

instead of:

too many errors to return
  have (<T>)
  want ()
too many arguments in call to f
  have (number, <T>)
  want (number, number)

Change-Id: I680abc7cdd8444400e234caddf3ff49c2d69f53d
Reviewed-on: https://go-review.googlesource.com/36806
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agomath: add more tests for special cases of Bessel functions Y0, Y1, Yn
Alexander Döring [Mon, 20 Feb 2017 19:34:36 +0000 (20:34 +0100)]
math: add more tests for special cases of Bessel functions Y0, Y1, Yn

Test finite negative x with Y0(-1), Y1(-1), Yn(2,-1), Yn(-3,-1).

Also test the special case Yn(0,0).

Fixes #19130.

Change-Id: I95f05a72e1c455ed8ddf202c56f4266f03f370fd
Reviewed-on: https://go-review.googlesource.com/37310
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocontext: document that Err is unspecified before Done
Ian Lance Taylor [Tue, 21 Feb 2017 20:48:45 +0000 (12:48 -0800)]
context: document that Err is unspecified before Done

It could have been defined the other way, but since the behavior has
been unspecified, this is the conservative approach for people writing
different implementations of the Context interface.

Change-Id: I7334a4c674bc2330cca6874f7cac1eb0eaea3cff
Reviewed-on: https://go-review.googlesource.com/37375
Reviewed-by: Matt Layher <mdlayher@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Sameer Ajmani <sameer@golang.org>
Run-TryBot: Sameer Ajmani <sameer@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: more detail for crash_test.go
Lynn Boger [Tue, 21 Feb 2017 16:28:49 +0000 (11:28 -0500)]
runtime: more detail for crash_test.go

This updates the testcase to display the timestamps for the
runtime.a, it dependent packages atomic.a and sys.a, and
source files.

Change-Id: Id2901b4e8aa8eb9775c4f404ac01cc07b394ba91
Reviewed-on: https://go-review.googlesource.com/37332
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/compile: zero extend when replacing load-hit-store on s390x
Michael Munday [Fri, 17 Feb 2017 00:53:08 +0000 (19:53 -0500)]
cmd/compile: zero extend when replacing load-hit-store on s390x

Keith pointed out that these rules should zero extend during the review
of CL 36845. In practice the generic rules are responsible for eliminating
most load-hit-stores and they do not have this problem. When the s390x
rules are triggered any cast following the elided load-hit-store is
kept because of the sequence the rules are applied in (i.e. the load is
removed before the zero extension gets a chance to be merged into the load).
It is therefore not clear that this issue results in any functional bugs.

This CL includes a test, but it only tests the generic rules currently.

Change-Id: Idbc43c782097a3fb159be293ec3138c5b36858ad
Reviewed-on: https://go-review.googlesource.com/37154
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: add opcode flag hasSideEffects for do-not-remove
David Chase [Tue, 21 Feb 2017 20:22:52 +0000 (15:22 -0500)]
cmd/compile: add opcode flag hasSideEffects for do-not-remove

Added a flag to generic and various architectures' atomic
operations that are judged to have observable side effects
and thus cannot be dead-code-eliminated.

Test requires GOMAXPROCS > 1 without preemption in loop.

Fixes #19182.

Change-Id: Id2230031abd2cca0bbb32fd68fc8a58fb912070f
Reviewed-on: https://go-review.googlesource.com/37333
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agoreflect: fix bucketOf to only look at ptrdata entries in gcdata
Ian Lance Taylor [Thu, 16 Feb 2017 22:20:24 +0000 (14:20 -0800)]
reflect: fix bucketOf to only look at ptrdata entries in gcdata

The gcdata field only records ptrdata entries, not size entries.

Also fix an obsolete comment: the enforced limit on pointer maps is
now 2048 bytes, not 16 bytes.

I wasn't able to contruct a test case for this. It would require
building a type whose size is greater than 64 bytes but less than 128
bytes, with at least one pointer in first 64 bytes but no pointers
after the first 64 bytes, such that the linker arranges for the one
byte gcbits value to be immediately followed by a non-zero byte.

Change-Id: I9118d3e4ec6f07fd18b72f621c1e5f4fdfe5f80b
Reviewed-on: https://go-review.googlesource.com/37142
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
8 years agocmd/compile: update builtin writeBarrier to match runtime
Ian Lance Taylor [Tue, 21 Feb 2017 23:57:06 +0000 (15:57 -0800)]
cmd/compile: update builtin writeBarrier to match runtime

The definition of writeBarrier in the runtime was changed in CL 22855
to include padding. Update the definition built in to the compiler to match.
This doesn't affect the generated code, as the compiler sets the type
to use anyhow, but having them be different seems clearly wrong.

Change-Id: I8eac05bf70a424a0b2338ba5e9e41af231316de0
Reviewed-on: https://go-review.googlesource.com/37377
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agodoc: use appropriate type to describe return value
Kevin Burke [Tue, 21 Feb 2017 19:23:04 +0000 (11:23 -0800)]
doc: use appropriate type to describe return value

Fixes #19223.

Change-Id: I4cc8e81559a1313e1477ee36902e1b653155a888
Reviewed-on: https://go-review.googlesource.com/37374
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
8 years agocmd/compile: do not fold offset into load/store for args on ARM64
Cherry Zhang [Sun, 19 Feb 2017 02:03:15 +0000 (21:03 -0500)]
cmd/compile: do not fold offset into load/store for args on ARM64

Args may be not at 8-byte aligned offset to SP. When the stack
frame is large, folding the offset of args may cause large
unaligned offsets that does not fit in a machine instruction on
ARM64. Therefore disable folding offsets for args.

This has small performance impact (see below). A better fix would
be letting the assembler backend fix up the offset by loading it
into a register if it doesn't fit into an instruction. And the
compiler can simply generate large load/stores with offset. Since
in most of the cases the offset is aligned or the stack frame is
small, it can fit in an instruction and no fixup is needed. But
this is too complicated for Go 1.8.

name                     old time/op    new time/op    delta
BinaryTree17-8              8.30s ± 0%     8.31s ± 0%    ~     (p=0.579 n=10+10)
Fannkuch11-8                6.14s ± 0%     6.18s ± 0%  +0.53%  (p=0.000 n=9+10)
FmtFprintfEmpty-8           117ns ± 0%     117ns ± 0%    ~     (all equal)
FmtFprintfString-8          196ns ± 0%     197ns ± 0%  +0.72%  (p=0.000 n=10+10)
FmtFprintfInt-8             204ns ± 0%     205ns ± 0%  +0.49%  (p=0.000 n=9+10)
FmtFprintfIntInt-8          302ns ± 0%     307ns ± 1%  +1.46%  (p=0.000 n=10+10)
FmtFprintfPrefixedInt-8     329ns ± 2%     326ns ± 0%    ~     (p=0.083 n=10+10)
FmtFprintfFloat-8           540ns ± 0%     542ns ± 0%  +0.46%  (p=0.000 n=8+7)
FmtManyArgs-8              1.20µs ± 1%    1.19µs ± 1%  -1.02%  (p=0.000 n=10+10)
GobDecode-8                17.3ms ± 1%    17.8ms ± 0%  +2.75%  (p=0.000 n=10+7)
GobEncode-8                15.3ms ± 1%    15.4ms ± 0%  +0.57%  (p=0.004 n=9+10)
Gzip-8                      789ms ± 0%     803ms ± 0%  +1.78%  (p=0.000 n=9+10)
Gunzip-8                    128ms ± 0%     130ms ± 0%  +1.73%  (p=0.000 n=10+9)
HTTPClientServer-8          202µs ± 6%     201µs ±10%    ~     (p=0.739 n=10+10)
JSONEncode-8               42.0ms ± 0%    42.1ms ± 0%  +0.19%  (p=0.028 n=10+9)
JSONDecode-8                159ms ± 0%     161ms ± 0%  +1.05%  (p=0.000 n=9+10)
Mandelbrot200-8            10.1ms ± 0%    10.1ms ± 0%  -0.07%  (p=0.000 n=10+9)
GoParse-8                  8.46ms ± 1%    8.61ms ± 1%  +1.77%  (p=0.000 n=10+10)
RegexpMatchEasy0_32-8       227ns ± 1%     226ns ± 0%  -0.35%  (p=0.001 n=10+9)
RegexpMatchEasy0_1K-8      1.63µs ± 0%    1.63µs ± 0%  -0.13%  (p=0.000 n=10+9)
RegexpMatchEasy1_32-8       250ns ± 0%     249ns ± 0%  -0.40%  (p=0.001 n=8+9)
RegexpMatchEasy1_1K-8      2.07µs ± 0%    2.08µs ± 0%  +0.05%  (p=0.027 n=9+9)
RegexpMatchMedium_32-8      350ns ± 0%     350ns ± 0%    ~     (p=0.412 n=9+8)
RegexpMatchMedium_1K-8      104µs ± 0%     104µs ± 0%  +0.31%  (p=0.000 n=10+7)
RegexpMatchHard_32-8       5.82µs ± 0%    5.82µs ± 0%    ~     (p=0.937 n=9+9)
RegexpMatchHard_1K-8        176µs ± 0%     176µs ± 0%  +0.03%  (p=0.000 n=9+8)
Revcomp-8                   1.36s ± 1%     1.37s ± 1%    ~     (p=0.218 n=10+10)
Template-8                  151ms ± 1%     156ms ± 1%  +3.21%  (p=0.000 n=10+10)
TimeParse-8                 737ns ± 0%     758ns ± 2%  +2.74%  (p=0.000 n=10+10)
TimeFormat-8                801ns ± 2%     789ns ± 1%  -1.51%  (p=0.000 n=10+10)
[Geo mean]                  142µs          143µs       +0.50%

Fixes #19137.

Change-Id: Ib8a21ea98c0ffb2d282a586535b213cc163e1b67
Reviewed-on: https://go-review.googlesource.com/37251
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agomath/big: define Word as uint instead of uintptr
Robert Griesemer [Tue, 21 Feb 2017 18:22:05 +0000 (10:22 -0800)]
math/big: define Word as uint instead of uintptr

For compatibility with math/bits uint operations.

When math/big was written originally, the Go compiler used 32bit
int/uint values even on a 64bit machine. uintptr was the type that
represented the machine register size. Now, the int/uint types are
sized to the native machine register size, so they are the natural
machine Word type.

On most machines, the size of int/uint correspond to the size of
uintptr. On platforms where uint and uintptr have different sizes,
this change may lead to performance differences (e.g., amd64p32).

Change-Id: Ief249c160b707b6441848f20041e32e9e9d8d8ca
Reviewed-on: https://go-review.googlesource.com/37372
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: fix type of OffPtr generated by ODOTPTR
Michael Munday [Sun, 19 Feb 2017 17:10:18 +0000 (12:10 -0500)]
cmd/compile: fix type of OffPtr generated by ODOTPTR

The type of the OffPtr should be consistent with the type of the
following load. Before this CL it was typed as a pointer to the
struct.

Fixes #19164.

Change-Id: Ibcdec4411c6f719702f76f8dba3cce8691bfbe0c
Reviewed-on: https://go-review.googlesource.com/37254
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agofmt: add short note about %g precision
Alberto Donizetti [Tue, 21 Feb 2017 13:30:17 +0000 (14:30 +0100)]
fmt: add short note about %g precision

Fixes #18772

Change-Id: Ib5d9ffa0abd35b9d3ca83bac139aece0f3c9702d
Reviewed-on: https://go-review.googlesource.com/37313
Reviewed-by: Rob Pike <r@golang.org>
8 years agosyscall: fix linux/mipsx ret value FP offsets for Syscall9
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:58:36 +0000 (22:58 -0800)]
syscall: fix linux/mipsx ret value FP offsets for Syscall9

Found by vet.

Change-Id: Idf910405566816ddce6781c8e99f90b59f33d63c
Reviewed-on: https://go-review.googlesource.com/37308
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agosync/atomic: fix mipsx frame sizes
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:58:11 +0000 (22:58 -0800)]
sync/atomic: fix mipsx frame sizes

Found by vet.

Change-Id: Ied3089a2cc8757ae5377fb5fa05bbb385d26ad9c
Reviewed-on: https://go-review.googlesource.com/37307
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: use standard linux/mipsx clone variable names
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:57:45 +0000 (22:57 -0800)]
runtime: use standard linux/mipsx clone variable names

Change-Id: I62118e197190af1d11a89921d5769101ce6d2257
Reviewed-on: https://go-review.googlesource.com/37306
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocrypto/aes: minor ppc64 assembly naming improvements
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:57:10 +0000 (22:57 -0800)]
crypto/aes: minor ppc64 assembly naming improvements

doEncryptKeyAsm is tail-called from other assembly routines.
Give it a proper prototype so that vet can check it.
Adjust one assembly FP reference accordingly.

Change-Id: I263fcb0191529214b16e6bd67330fadee492eef4
Reviewed-on: https://go-review.googlesource.com/37305
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/vet/all: update windows whitelist
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:55:51 +0000 (22:55 -0800)]
cmd/vet/all: update windows whitelist

A last-minute rollback of a change left some
unreachable code that we don't want to remove.

Change-Id: Ida0af5b18ed1a2e13ef66c303694afcc49d7bff4
Reviewed-on: https://go-review.googlesource.com/37304
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: update assembly var names after monotonic time changes
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:54:39 +0000 (22:54 -0800)]
runtime: update assembly var names after monotonic time changes

Change-Id: I721045120a4df41462c02252e2e5e8529ae2d694
Reviewed-on: https://go-review.googlesource.com/37303
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/vet/all: update whitelists for monotonic time changes
Josh Bleecher Snyder [Tue, 21 Feb 2017 06:54:07 +0000 (22:54 -0800)]
cmd/vet/all: update whitelists for monotonic time changes

Change-Id: Ib942cb9e0cb20821aea4274bc3ddc83a215afbcb
Reviewed-on: https://go-review.googlesource.com/37302
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/vet/all: add mips and mipsle
Josh Bleecher Snyder [Tue, 14 Feb 2017 00:19:51 +0000 (16:19 -0800)]
cmd/vet/all: add mips and mipsle

Change-Id: I689b2e8e214561350f88fa4e20c8f34cf69dc6a7
Reviewed-on: https://go-review.googlesource.com/37301
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/vet/all: work around vet printf checker deficiencies
Josh Bleecher Snyder [Tue, 14 Feb 2017 00:11:13 +0000 (16:11 -0800)]
cmd/vet/all: work around vet printf checker deficiencies

cmd/vet has a known deficiency in its handling of fmt.Formatters.
This causes a spurious printf error only for non-host platforms.
Since cmd/vet/all may get run on any given platform,
whitelists cannot help here.

Work around the issue by skipping printf tests entirely
for non-host platforms.

Work around the one known acceptable false positive from vet
by whitelisting the file that contains it.

Change-Id: Id74b3d4db0519cf9a670a065683715f856266e45
Reviewed-on: https://go-review.googlesource.com/36936
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoos/exec: deflake TestStdinCloseRace
Ian Lance Taylor [Tue, 21 Feb 2017 15:14:51 +0000 (07:14 -0800)]
os/exec: deflake TestStdinCloseRace

Stop reporting errors from cmd.Process.Kill; they don't matter for
purposes of this test, and they can occur if the process exits quickly.

Fixes #19211.
Fixes #19213.

Change-Id: I1a0bb9170220ca69199abb8e8811b1dde43e1897
Reviewed-on: https://go-review.googlesource.com/37309
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: fix storeOrder
Cherry Zhang [Sat, 18 Feb 2017 23:37:38 +0000 (18:37 -0500)]
cmd/compile: fix storeOrder

storeOrder visits values in DFS order. It should "break" after
pushing one argument to stack, instead of "continue".

Fixes #19179.

Change-Id: I561afb44213df40ebf8bf7d28e0fd00f22a81ac0
Reviewed-on: https://go-review.googlesource.com/37250
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
8 years agocmd/link: simplify peemitreloc
Alex Brainman [Wed, 8 Feb 2017 01:04:11 +0000 (12:04 +1100)]
cmd/link: simplify peemitreloc

No functional changes.

For #10776.

Change-Id: If9a5ef832af116c5802b06a38e0c050d7363f2d5
Reviewed-on: https://go-review.googlesource.com/36981
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/link: reorder pe sections
Alex Brainman [Wed, 8 Feb 2017 00:44:09 +0000 (11:44 +1100)]
cmd/link: reorder pe sections

dwarf writing code assumes that dwarf sections follow
.data and .bss, not .ctors. Make pe section writing code
match that assumption.

For #10776.

Change-Id: I128c3ad125f7d0db19e922f165704a054b2af7ba
Reviewed-on: https://go-review.googlesource.com/36980
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/link: do not add __image_base__ and _image_base__ if external linker
Alex Brainman [Wed, 8 Feb 2017 00:42:24 +0000 (11:42 +1100)]
cmd/link: do not add __image_base__ and _image_base__ if external linker

The symbols get in a way when using external linker. They are
not associated with a section. And linker fails when
generating relocations for them.

__image_base__ and _image_base__ have been added long time ago.
I do not think they are needed anymore. If I delete them, all
tests still PASS. I tried going back to the commit that added
them to see if I can reproduce original error, but I cannot
build it. I don't have hg version of go repo, and my gcc is
complaining about cc source code.

I wasted too much time with this, so I decided to leave them only
for internal linker. That is what they were originally added for.

For #10776.

Change-Id: Ibb72b04f3864947c782f964a7badc69f4b074e25
Reviewed-on: https://go-review.googlesource.com/36979
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/link: add all pe section names to pe symbol table
Alex Brainman [Wed, 8 Feb 2017 02:59:19 +0000 (13:59 +1100)]
cmd/link: add all pe section names to pe symbol table

dwarf relocations refer to dwarf section symbols, so dwarf
section symbols must be present in pe symbol table before we
write dwarf relocations.

.ctors pe section already refer to .text symbol.

Write all pe section name symbols into symbol table, so we
can use them whenever we need them.

This CL also simplified some code.

For #10776.

Change-Id: I9b8c680ea75904af90c797a06bbb1f4df19e34b6
Reviewed-on: https://go-review.googlesource.com/36978
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/link: introduce shNames
Alex Brainman [Wed, 8 Feb 2017 02:58:21 +0000 (13:58 +1100)]
cmd/link: introduce shNames

Introduce a slice that keeps long pe section names as we add them.
It will be used later to output pe symbol table and dwarf relocations.

For #10776.

Change-Id: I02f808a456393659db2354031baf1d4f9e0b2d61
Reviewed-on: https://go-review.googlesource.com/36977
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/link: set VirtualAddress to 0 for external linker
Alex Brainman [Mon, 6 Feb 2017 04:18:26 +0000 (15:18 +1100)]
cmd/link: set VirtualAddress to 0 for external linker

This is what gcc does when it generates object files.
And pecoff.doc says: "for simplicity, compilers should
 set this to zero". It is easier to count everything,
when it starts from 0. Make go linker do the same.

For #10776.

Change-Id: Iffa4b3ad86160624ed34adf1c6ba13feba34c658
Reviewed-on: https://go-review.googlesource.com/36976
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime/pprof: mark TestMutexProfile as flaky for now
Brad Fitzpatrick [Mon, 20 Feb 2017 19:48:28 +0000 (19:48 +0000)]
runtime/pprof: mark TestMutexProfile as flaky for now

Flaky tests hurt productivity. Disable for now.

Updates #19139

Change-Id: I2e3040bdf0e53597a1c4f925b788e3268ea284c1
Reviewed-on: https://go-review.googlesource.com/37291
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Peter Weinberger <pjw@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/objdump: make test independent of inlining
David Lazar [Sun, 19 Feb 2017 19:01:42 +0000 (14:01 -0500)]
cmd/objdump: make test independent of inlining

Fixes #19189.

Change-Id: Ice69216c7fc2eaeb3dbbdcd08a8284204c7f52ef
Reviewed-on: https://go-review.googlesource.com/37237
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: remove unused constant divide strength reduction code
Martin Möhrmann [Sun, 19 Feb 2017 10:34:29 +0000 (11:34 +0100)]
cmd/compile: remove unused constant divide strength reduction code

Change list https://golang.org/cl/37015/ moved the optimization
of division by constants to the generic ssa backend.
This removes the old now unused code that was used
for this optimization outside of the ssa backend.

Change-Id: I86223e56742e48dbb372ba8d779681e66448c513
Reviewed-on: https://go-review.googlesource.com/37198
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agomath/bits: faster OnesCount
Robert Griesemer [Sat, 18 Feb 2017 19:14:35 +0000 (11:14 -0800)]
math/bits: faster OnesCount

Using some additional suggestions per "Hacker's Delight".
Added documentation and extra tests.

Measured on 1.7 GHz Intel Core i7, running macOS 10.12.3.

benchmark                  old ns/op     new ns/op     delta
BenchmarkOnesCount-4       7.34          5.38          -26.70%
BenchmarkOnesCount8-4      2.03          1.98          -2.46%
BenchmarkOnesCount16-4     2.56          2.50          -2.34%
BenchmarkOnesCount32-4     2.98          2.39          -19.80%
BenchmarkOnesCount64-4     4.22          2.96          -29.86%

Change-Id: I566b0ef766e55cf5776b1662b6016024ebe5d878
Reviewed-on: https://go-review.googlesource.com/37223
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agofmt: remove unused global variable byteType
Martin Möhrmann [Sun, 19 Feb 2017 09:42:34 +0000 (10:42 +0100)]
fmt: remove unused global variable byteType

Change list https://golang.org/cl/20686/ removed the last use
of the variable byteType.

Change-Id: I4ea79095136a49a9d22767b37f48f3404da05056
Reviewed-on: https://go-review.googlesource.com/37197
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: amd64, allow XCHG on stack pointers
Keith Randall [Sun, 19 Feb 2017 08:16:27 +0000 (00:16 -0800)]
cmd/compile: amd64, allow XCHG on stack pointers

XCHG needs to allow the stack pointer as an argument because we have a
rewrite that incorporates the address of a local variable into the
instruction.

Fixes #19184

Change-Id: Ic438e6e1946332cdce3864d15abecd41b911b2a9
Reviewed-on: https://go-review.googlesource.com/37253
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/go/internal/envcmd: report PKG_CONFIG after the CGO group
Jaana Burcu Dogan [Sat, 18 Feb 2017 23:07:13 +0000 (15:07 -0800)]
cmd/go/internal/envcmd: report PKG_CONFIG after the CGO group

Before the change, `go env` reports PKG_CONFIG in between the
CGO env group:

    GOARCH="amd64"
    GOBIN=""
    GOEXE=""
    GOHOSTARCH="amd64"
    GOHOSTOS="darwin"
    GOOS="darwin"
    GOPATH="/Users/jbd"
    GORACE=""
    GOROOT="/Users/jbd/go"
    GOTOOLDIR="/Users/jbd/go/pkg/tool/darwin_amd64"
    GCCGO="gccgo"
    CC="clang"
    GOGCCFLAGS="-fPIC -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/lq/qcn67khn4_1b41_g48x3zchh005d21/T/go-build184491598=/tmp/go-build -gno-record-gcc-switches -fno-common"
    CXX="clang++"
    CGO_ENABLED="1"
    PKG_CONFIG="pkg-config"
    CGO_CFLAGS="-g -O2"
    CGO_CPPFLAGS=""
    CGO_CXXFLAGS="-g -O2"
    CGO_FFLAGS="-g -O2"
    CGO_LDFLAGS="-g -O2"

The change makes PKG_CONFIG to be reported as the final item,
and not breaking the CGO_* group apart.

Change-Id: I1e7ed6bdec83009ff118f85c9f0f7b78a67fdd76
Reviewed-on: https://go-review.googlesource.com/37228
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agofmt: support sharp flag for float and complex value printing
Martin Möhrmann [Wed, 15 Feb 2017 11:41:02 +0000 (12:41 +0100)]
fmt: support sharp flag for float and complex value printing

Added an alternate form of printing floats and complex values
by specifying the sharp flag.

Output formatted using the the verbs v, e, E, f, F, g and G in
combination with the sharp flag will always include a decimal point.

The alternate form specified by the sharp flag for %g and %G verbs
will not truncate trailing zeros and assume a default precision of 6.

Fixes #18857.

Change-Id: I4d776239e06d7a6a90f2d8556240a359888cb7c3
Reviewed-on: https://go-review.googlesource.com/37051
Reviewed-by: Rob Pike <r@golang.org>
8 years agonet/url: document that Query returns only valid values
Kenny Grant [Sat, 18 Feb 2017 07:28:49 +0000 (07:28 +0000)]
net/url: document that Query returns only valid values

Fixes #19110

Change-Id: I291fa4ec3c61145162acd019e3f0e5dd3d7c97e9
Reviewed-on: https://go-review.googlesource.com/37194
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agomath: protect benchmarked functions from being optimized away
Martin Möhrmann [Sat, 18 Feb 2017 08:32:46 +0000 (09:32 +0100)]
math: protect benchmarked functions from being optimized away

Add exported global variables and store the results of benchmarked
functions in them. This prevents the current compiler optimizations
from removing the instructions that are needed to compute the return
values of the benchmarked functions.

Change-Id: If8b08424e85f3796bb6dd73e761c653abbabcc5e
Reviewed-on: https://go-review.googlesource.com/37195
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoos: remove incorrect detection of O_CLOEXEC flag on darwin
Martin Möhrmann [Sat, 18 Feb 2017 06:46:41 +0000 (07:46 +0100)]
os: remove incorrect detection of O_CLOEXEC flag on darwin

The below range loop will not stop when encountering
the first '.' character in a Darwin version string like "15.6.0".

for i = range osver {
   if osver[i] != '.' {
         continue
      }
   }
}

Therefore, the condition i > 2 was always satisfied and
supportsCloseOnExec was always set to true.

Since the minimum supported version of OSX for go is currently 10.8
and O_CLOEXEC is implemented from OSX 10.7 on the detection code
can be removed and support for O_CLOEXEC is always assumed to exist.

Change-Id: Idd10094d8385dd4adebc8d7a6d9e9a8f29455867
Reviewed-on: https://go-review.googlesource.com/37193
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agogo/doc: allow : in godoc links
Kenny Grant [Sun, 18 Dec 2016 07:24:58 +0000 (07:24 +0000)]
go/doc: allow : in godoc links

The emphasize function used a complex regexp to find URLs, which
truncated some types of URL and did not match others.
This has been simplified and adjusted to allow valid punctuation
like :: or ! in the path part and :[] in the host part.
Comments were added to clarify what this regexp allows.
The path part matches query and fragment also so document this.
Removed news, telnet, wais, and prospero protocols.

Tests were added for:
 IPV6 URLs
 URLs surrounded by brackets
 URLs containing ::
 URLs containing :;!- in the path

In order to allow punctuation and yet preserve current behaviour,
URLs are not permitted to end in .,:;?! to allow the use of
normal punctuation surrounding URLs in comments.

Fixes #18139

Change-Id: I38b2d7a85fe0d171e4bf4aac420f8c2d3ced8a2f
Reviewed-on: https://go-review.googlesource.com/37192
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agomath/bits: added benchmarks for Leading/TrailingZeros
Robert Griesemer [Fri, 17 Feb 2017 23:21:36 +0000 (15:21 -0800)]
math/bits: added benchmarks for Leading/TrailingZeros

BenchmarkLeadingZeros-8       200000000          8.80 ns/op
BenchmarkLeadingZeros8-8      200000000          8.21 ns/op
BenchmarkLeadingZeros16-8     200000000          7.49 ns/op
BenchmarkLeadingZeros32-8     200000000          7.80 ns/op
BenchmarkLeadingZeros64-8     200000000          8.67 ns/op

BenchmarkTrailingZeros-8      1000000000          2.05 ns/op
BenchmarkTrailingZeros8-8     2000000000          1.94 ns/op
BenchmarkTrailingZeros16-8    2000000000          1.94 ns/op
BenchmarkTrailingZeros32-8    2000000000          1.92 ns/op
BenchmarkTrailingZeros64-8    2000000000          2.03 ns/op

Change-Id: I45497bf2d6369ba6cfc88ded05aa735908af8908
Reviewed-on: https://go-review.googlesource.com/37220
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agomath/bits: faster Rotate functions, added respective benchmarks
Robert Griesemer [Fri, 17 Feb 2017 23:02:49 +0000 (15:02 -0800)]
math/bits: faster Rotate functions, added respective benchmarks

Measured on 2.3 GHz Intel Core i7, running maxOS 10.12.3.

benchmark                    old ns/op     new ns/op     delta
BenchmarkRotateLeft-8        7.87          7.00          -11.05%
BenchmarkRotateLeft8-8       8.41          4.52          -46.25%
BenchmarkRotateLeft16-8      8.07          4.55          -43.62%
BenchmarkRotateLeft32-8      8.36          4.73          -43.42%
BenchmarkRotateLeft64-8      7.93          4.78          -39.72%

BenchmarkRotateRight-8       8.23          6.72          -18.35%
BenchmarkRotateRight8-8      8.76          4.39          -49.89%
BenchmarkRotateRight16-8     9.07          4.44          -51.05%
BenchmarkRotateRight32-8     8.85          4.46          -49.60%
BenchmarkRotateRight64-8     8.11          4.43          -45.38%

Change-Id: I79ea1e9e6fc65f95794a91f860a911efed3aa8a1
Reviewed-on: https://go-review.googlesource.com/37219
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agomath/bits: faster OnesCount, added respective benchmarks
Robert Griesemer [Fri, 17 Feb 2017 22:20:11 +0000 (14:20 -0800)]
math/bits: faster OnesCount, added respective benchmarks

Also: Changed Reverse/ReverseBytes implementations to use
the same (smaller) masks as OnesCount.

BenchmarkOnesCount-8          37.0          6.26          -83.08%
BenchmarkOnesCount8-8         7.24          1.99          -72.51%
BenchmarkOnesCount16-8        11.3          2.47          -78.14%
BenchmarkOnesCount32-8        18.4          3.02          -83.59%
BenchmarkOnesCount64-8        40.0          3.78          -90.55%
BenchmarkReverse-8            6.69          6.22          -7.03%
BenchmarkReverse8-8           1.64          1.64          +0.00%
BenchmarkReverse16-8          2.26          2.18          -3.54%
BenchmarkReverse32-8          2.88          2.87          -0.35%
BenchmarkReverse64-8          5.64          4.34          -23.05%
BenchmarkReverseBytes-8       2.48          2.17          -12.50%
BenchmarkReverseBytes16-8     0.63          0.95          +50.79%
BenchmarkReverseBytes32-8     1.13          1.24          +9.73%
BenchmarkReverseBytes64-8     2.50          2.16          -13.60%

OnesCount-8       37.0ns ± 0%   6.3ns ± 0%   ~             (p=1.000 n=1+1)
OnesCount8-8      7.24ns ± 0%  1.99ns ± 0%   ~             (p=1.000 n=1+1)
OnesCount16-8     11.3ns ± 0%   2.5ns ± 0%   ~             (p=1.000 n=1+1)
OnesCount32-8     18.4ns ± 0%   3.0ns ± 0%   ~             (p=1.000 n=1+1)
OnesCount64-8     40.0ns ± 0%   3.8ns ± 0%   ~             (p=1.000 n=1+1)
Reverse-8         6.69ns ± 0%  6.22ns ± 0%   ~             (p=1.000 n=1+1)
Reverse8-8        1.64ns ± 0%  1.64ns ± 0%   ~     (all samples are equal)
Reverse16-8       2.26ns ± 0%  2.18ns ± 0%   ~             (p=1.000 n=1+1)
Reverse32-8       2.88ns ± 0%  2.87ns ± 0%   ~             (p=1.000 n=1+1)
Reverse64-8       5.64ns ± 0%  4.34ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes-8    2.48ns ± 0%  2.17ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes16-8  0.63ns ± 0%  0.95ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes32-8  1.13ns ± 0%  1.24ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes64-8  2.50ns ± 0%  2.16ns ± 0%   ~             (p=1.000 n=1+1)

Change-Id: I591b0ffc83fc3a42828256b6e5030f32c64f9497
Reviewed-on: https://go-review.googlesource.com/37218
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile/internal/ssa: combine load + op on AMD64
Ilya Tocar [Fri, 10 Feb 2017 19:17:20 +0000 (13:17 -0600)]
cmd/compile/internal/ssa: combine load + op on AMD64

On AMD64 Most operation can have one operand in memory.
Combine load and dependand operation into one new operation,
where possible. I've seen no significant performance changes on go1,
but this allows to remove ~1.8kb code from go tool. And in math package
I see e. g.:

Remainder-6            70.0ns ± 0%   64.6ns ± 0%   -7.76%  (p=0.000 n=9+1
Change-Id: I88b8602b1d55da8ba548a34eb7da4b25d59a297e
Reviewed-on: https://go-review.googlesource.com/36793
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: fix 32-bit unsigned division on 64-bit machines
Keith Randall [Fri, 17 Feb 2017 19:36:08 +0000 (11:36 -0800)]
cmd/compile: fix 32-bit unsigned division on 64-bit machines

The type of an intermediate multiply was wrong.  When that
intermediate multiply was spilled, the top 32 bits were lost.

Fixes #19153

Change-Id: Ib29350a4351efa405935b7f7ee3c112668e64108
Reviewed-on: https://go-review.googlesource.com/37212
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agomath/bits: faster Reverse, ReverseBytes
Robert Griesemer [Fri, 17 Feb 2017 21:27:09 +0000 (13:27 -0800)]
math/bits: faster Reverse, ReverseBytes

- moved from: x&m>>k | x&^m<<k to: x&m>>k | x<<k&m
  This permits use of the same constant m twice (*) which may be
  better for machines that can't use large immediate constants
  directly with an AND instruction and have to load them explicitly.
  *) CPUs don't usually have a &^ instruction, so x&^m becomes x&(^m)

- simplified returns
  This improves the generated code because the compiler recognizes
  x>>k | x<<k as ROT when k is the bitsize of x.

The 8-bit versions of these instructions can be significantly faster
still if they are replaced with table lookups, as long as the table
is in cache. If the table is not in cache, table-lookup is probably
slower, hence the choice of an explicit register-only implementation
for now.

BenchmarkReverse-8            8.50          6.86          -19.29%
BenchmarkReverse8-8           2.17          1.74          -19.82%
BenchmarkReverse16-8          2.89          2.34          -19.03%
BenchmarkReverse32-8          3.55          2.95          -16.90%
BenchmarkReverse64-8          6.81          5.57          -18.21%
BenchmarkReverseBytes-8       3.49          2.48          -28.94%
BenchmarkReverseBytes16-8     0.93          0.62          -33.33%
BenchmarkReverseBytes32-8     1.55          1.13          -27.10%
BenchmarkReverseBytes64-8     2.47          2.47          +0.00%

Reverse-8         8.50ns ± 0%  6.86ns ± 0%   ~             (p=1.000 n=1+1)
Reverse8-8        2.17ns ± 0%  1.74ns ± 0%   ~             (p=1.000 n=1+1)
Reverse16-8       2.89ns ± 0%  2.34ns ± 0%   ~             (p=1.000 n=1+1)
Reverse32-8       3.55ns ± 0%  2.95ns ± 0%   ~             (p=1.000 n=1+1)
Reverse64-8       6.81ns ± 0%  5.57ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes-8    3.49ns ± 0%  2.48ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes16-8  0.93ns ± 0%  0.62ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes32-8  1.55ns ± 0%  1.13ns ± 0%   ~             (p=1.000 n=1+1)
ReverseBytes64-8  2.47ns ± 0%  2.47ns ± 0%   ~     (all samples are equal)

Change-Id: I0064de8c7e0e568ca7885d6f7064344bef91a06d
Reviewed-on: https://go-review.googlesource.com/37215
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/gc: remove Node.IsStatic field
Matthew Dempsky [Thu, 16 Feb 2017 05:16:49 +0000 (21:16 -0800)]
cmd/compile/internal/gc: remove Node.IsStatic field

We can immediately emit static assignment data rather than queueing
them up to be processed during SSA building.

Passes toolstash -cmp.

Change-Id: I8bcea4b72eafb0cc0b849cd93e9cde9d84f30d5e
Reviewed-on: https://go-review.googlesource.com/37024
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: check both syms when folding address into load/store on ARM64
Cherry Zhang [Fri, 17 Feb 2017 15:27:43 +0000 (10:27 -0500)]
cmd/compile: check both syms when folding address into load/store on ARM64

The rules for folding addresses into load/stores checks sym1 is
not on stack (because the stack offset is not known at that point).
But sym1 could be nil, which invalidates the check. Check merged
sym instead.

Fixes #19137.

Change-Id: I8574da22ced1216bb5850403d8f08ec60a8d1005
Reviewed-on: https://go-review.googlesource.com/37145
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agomath/bits: fix benchmarks (make sure calls don't get optimized away)
Robert Griesemer [Fri, 17 Feb 2017 20:37:38 +0000 (12:37 -0800)]
math/bits: fix benchmarks (make sure calls don't get optimized away)

Sum up function results and store them in an exported (global)
variable. This prevents the compiler from optimizing away the
otherwise side-effect free function calls.

We now have more realistic set of benchmark numbers...

Measured on 2.3 GHz Intel Core i7, running maxOS 10.12.3.

Note: These measurements are based on the same "old"
implementation as the prior measurements (commit 7d5c003).

benchmark                     old ns/op     new ns/op     delta
BenchmarkReverse-8            72.9          8.50          -88.34%
BenchmarkReverse8-8           13.2          2.17          -83.56%
BenchmarkReverse16-8          21.2          2.89          -86.37%
BenchmarkReverse32-8          36.3          3.55          -90.22%
BenchmarkReverse64-8          71.3          6.81          -90.45%
BenchmarkReverseBytes-8       11.2          3.49          -68.84%
BenchmarkReverseBytes16-8     6.24          0.93          -85.10%
BenchmarkReverseBytes32-8     7.40          1.55          -79.05%
BenchmarkReverseBytes64-8     10.5          2.47          -76.48%

Reverse-8         72.9ns ± 0%   8.5ns ± 0%   ~     (p=1.000 n=1+1)
Reverse8-8        13.2ns ± 0%   2.2ns ± 0%   ~     (p=1.000 n=1+1)
Reverse16-8       21.2ns ± 0%   2.9ns ± 0%   ~     (p=1.000 n=1+1)
Reverse32-8       36.3ns ± 0%   3.5ns ± 0%   ~     (p=1.000 n=1+1)
Reverse64-8       71.3ns ± 0%   6.8ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes-8    11.2ns ± 0%   3.5ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes16-8  6.24ns ± 0%  0.93ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes32-8  7.40ns ± 0%  1.55ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes64-8  10.5ns ± 0%   2.5ns ± 0%   ~     (p=1.000 n=1+1)

Change-Id: I8aef1334b84f6cafd25edccad7e6868b37969efb
Reviewed-on: https://go-review.googlesource.com/37213
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agomath/bits: much faster ReverseBytes, added respective benchmarks
Robert Griesemer [Fri, 17 Feb 2017 19:32:08 +0000 (11:32 -0800)]
math/bits: much faster ReverseBytes, added respective benchmarks

Measured on 2.3 GHz Intel Core i7, running maxOS 10.12.3.

benchmark                     old ns/op     new ns/op     delta
BenchmarkReverseBytes-8       11.4          3.51          -69.21%
BenchmarkReverseBytes16-8     6.87          0.64          -90.68%
BenchmarkReverseBytes32-8     7.79          0.65          -91.66%
BenchmarkReverseBytes64-8     11.6          0.64          -94.48%

name              old time/op  new time/op  delta
ReverseBytes-8    11.4ns ± 0%   3.5ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes16-8  6.87ns ± 0%  0.64ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes32-8  7.79ns ± 0%  0.65ns ± 0%   ~     (p=1.000 n=1+1)
ReverseBytes64-8  11.6ns ± 0%   0.6ns ± 0%   ~     (p=1.000 n=1+1)

Change-Id: I67b529652b3b613c61687e9e185e8d4ee40c51a2
Reviewed-on: https://go-review.googlesource.com/37211
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agomath/bits: much faster Reverse, added respective benchmarks
Robert Griesemer [Fri, 17 Feb 2017 19:12:49 +0000 (11:12 -0800)]
math/bits: much faster Reverse, added respective benchmarks

Measured on 2.3 GHz Intel Core i7, running maxOS 10.12.3.

name         old time/op  new time/op  delta
Reverse-8    76.6ns ± 0%   8.1ns ± 0%   ~     (p=1.000 n=1+1)
Reverse8-8   12.6ns ± 0%   0.6ns ± 0%   ~     (p=1.000 n=1+1)
Reverse16-8  20.8ns ± 0%   0.6ns ± 0%   ~     (p=1.000 n=1+1)
Reverse32-8  36.5ns ± 0%   0.6ns ± 0%   ~     (p=1.000 n=1+1)
Reverse64-8  74.0ns ± 0%   6.4ns ± 0%   ~     (p=1.000 n=1+1)

benchmark                old ns/op     new ns/op     delta
BenchmarkReverse-8       76.6          8.07          -89.46%
BenchmarkReverse8-8      12.6          0.64          -94.92%
BenchmarkReverse16-8     20.8          0.64          -96.92%
BenchmarkReverse32-8     36.5          0.64          -98.25%
BenchmarkReverse64-8     74.0          6.38          -91.38%

Change-Id: I6b99b10cee2f2babfe79342b50ee36a45a34da30
Reviewed-on: https://go-review.googlesource.com/37149
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: fix some types in SSA
Cherry Zhang [Mon, 6 Feb 2017 04:41:41 +0000 (23:41 -0500)]
cmd/compile: fix some types in SSA

These seem not to really matter, but good to be correct.

Change-Id: I02edb9797c3d6739725cfbe4723c75f151acd05e
Reviewed-on: https://go-review.googlesource.com/36837
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: redo writebarrier pass
Cherry Zhang [Wed, 1 Feb 2017 19:27:40 +0000 (14:27 -0500)]
cmd/compile: redo writebarrier pass

SSA's writebarrier pass requires WB store ops are always at the
end of a block. If we move write barrier insertion into SSA and
emits normal Store ops when building SSA, this requirement becomes
impractical -- it will create too many blocks for all the Store
ops.

Redo SSA's writebarrier pass, explicitly order values in store
order, so it no longer needs this requirement.

Updates #17583.
Fixes #19067.

Change-Id: I66e817e526affb7e13517d4245905300a90b7170
Reviewed-on: https://go-review.googlesource.com/36834
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agocmd/compile: re-enable nilcheck removal in same block
Cherry Zhang [Fri, 20 Jan 2017 15:38:05 +0000 (10:38 -0500)]
cmd/compile: re-enable nilcheck removal in same block

Nil check removal in the same block is disabled due to issue 18725:
because the values are not ordered, a nilcheck may influence a
value that is logically before it. This CL re-enables same-block
nilcheck removal by ordering values in store order first.

Updates #18725.

Change-Id: I287a38525230c14c5412cbcdbc422547dabd54f6
Reviewed-on: https://go-review.googlesource.com/35496
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agomath/bits: expand doc strings for all functions
Robert Griesemer [Thu, 16 Feb 2017 22:15:23 +0000 (14:15 -0800)]
math/bits: expand doc strings for all functions

Follow-up on https://go-review.googlesource.com/36315.
No functionality change.

For #18616.

Change-Id: Id4df34dd7d0381be06eea483a11bf92f4a01f604
Reviewed-on: https://go-review.googlesource.com/37140
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoall: fix a few typos in comments
Koki Ide [Thu, 16 Feb 2017 15:50:41 +0000 (00:50 +0900)]
all: fix a few typos in comments

Change-Id: I0455ffaa51c661803d8013c7961910f920d3c3cc
Reviewed-on: https://go-review.googlesource.com/37043
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agosync: make Mutex more fair
Dmitry Vyukov [Tue, 13 Dec 2016 15:45:55 +0000 (16:45 +0100)]
sync: make Mutex more fair

Add new starvation mode for Mutex.
In starvation mode ownership is directly handed off from
unlocking goroutine to the next waiter. New arriving goroutines
don't compete for ownership.
Unfair wait time is now limited to 1ms.
Also fix a long standing bug that goroutines were requeued
at the tail of the wait queue. That lead to even more unfair
acquisition times with multiple waiters.

Performance of normal mode is not considerably affected.

Fixes #13086

On the provided in the issue lockskew program:

done in 1.207853ms
done in 1.177451ms
done in 1.184168ms
done in 1.198633ms
done in 1.185797ms
done in 1.182502ms
done in 1.316485ms
done in 1.211611ms
done in 1.182418ms

name                    old time/op  new time/op   delta
MutexUncontended-48     0.65ns ± 0%   0.65ns ± 1%     ~           (p=0.087 n=10+10)
Mutex-48                 112ns ± 1%    114ns ± 1%   +1.69%        (p=0.000 n=10+10)
MutexSlack-48            113ns ± 0%     87ns ± 1%  -22.65%         (p=0.000 n=8+10)
MutexWork-48             149ns ± 0%    145ns ± 0%   -2.48%         (p=0.000 n=9+10)
MutexWorkSlack-48        149ns ± 0%    122ns ± 3%  -18.26%         (p=0.000 n=6+10)
MutexNoSpin-48           103ns ± 4%    105ns ± 3%     ~           (p=0.089 n=10+10)
MutexSpin-48             490ns ± 4%    515ns ± 6%   +5.08%        (p=0.006 n=10+10)
Cond32-48               13.4µs ± 6%   13.1µs ± 5%   -2.75%        (p=0.023 n=10+10)
RWMutexWrite100-48      53.2ns ± 3%   41.2ns ± 3%  -22.57%        (p=0.000 n=10+10)
RWMutexWrite10-48       45.9ns ± 2%   43.9ns ± 2%   -4.38%        (p=0.000 n=10+10)
RWMutexWorkWrite100-48   122ns ± 2%    134ns ± 1%   +9.92%        (p=0.000 n=10+10)
RWMutexWorkWrite10-48    206ns ± 1%    188ns ± 1%   -8.52%         (p=0.000 n=8+10)
Cond32-24               12.1µs ± 3%   12.4µs ± 3%   +1.98%         (p=0.043 n=10+9)
MutexUncontended-24     0.74ns ± 1%   0.75ns ± 1%     ~           (p=0.650 n=10+10)
Mutex-24                 122ns ± 2%    124ns ± 1%   +1.31%        (p=0.007 n=10+10)
MutexSlack-24           96.9ns ± 2%  102.8ns ± 2%   +6.11%        (p=0.000 n=10+10)
MutexWork-24             146ns ± 1%    135ns ± 2%   -7.70%         (p=0.000 n=10+9)
MutexWorkSlack-24        135ns ± 1%    128ns ± 2%   -5.01%         (p=0.000 n=10+9)
MutexNoSpin-24           114ns ± 3%    110ns ± 4%   -3.84%        (p=0.000 n=10+10)
MutexSpin-24             482ns ± 4%    475ns ± 8%     ~           (p=0.286 n=10+10)
RWMutexWrite100-24      43.0ns ± 3%   43.1ns ± 2%     ~           (p=0.956 n=10+10)
RWMutexWrite10-24       43.4ns ± 1%   43.2ns ± 1%     ~            (p=0.085 n=10+9)
RWMutexWorkWrite100-24   130ns ± 3%    131ns ± 3%     ~           (p=0.747 n=10+10)
RWMutexWorkWrite10-24    191ns ± 1%    192ns ± 1%     ~           (p=0.210 n=10+10)
Cond32-12               11.5µs ± 2%   11.7µs ± 2%   +1.98%        (p=0.002 n=10+10)
MutexUncontended-12     1.48ns ± 0%   1.50ns ± 1%   +1.08%        (p=0.004 n=10+10)
Mutex-12                 141ns ± 1%    143ns ± 1%   +1.63%        (p=0.000 n=10+10)
MutexSlack-12            121ns ± 0%    119ns ± 0%   -1.65%          (p=0.001 n=8+9)
MutexWork-12             141ns ± 2%    150ns ± 3%   +6.36%         (p=0.000 n=9+10)
MutexWorkSlack-12        131ns ± 0%    138ns ± 0%   +5.73%         (p=0.000 n=9+10)
MutexNoSpin-12          87.0ns ± 1%   83.7ns ± 1%   -3.80%        (p=0.000 n=10+10)
MutexSpin-12             364ns ± 1%    377ns ± 1%   +3.77%        (p=0.000 n=10+10)
RWMutexWrite100-12      42.8ns ± 1%   43.9ns ± 1%   +2.41%         (p=0.000 n=8+10)
RWMutexWrite10-12       39.8ns ± 4%   39.3ns ± 1%     ~            (p=0.433 n=10+9)
RWMutexWorkWrite100-12   131ns ± 1%    131ns ± 0%     ~            (p=0.591 n=10+9)
RWMutexWorkWrite10-12    173ns ± 1%    174ns ± 0%     ~            (p=0.059 n=10+8)
Cond32-6                10.9µs ± 2%   10.9µs ± 2%     ~           (p=0.739 n=10+10)
MutexUncontended-6      2.97ns ± 0%   2.97ns ± 0%     ~     (all samples are equal)
Mutex-6                  122ns ± 6%    122ns ± 2%     ~           (p=0.668 n=10+10)
MutexSlack-6             149ns ± 3%    142ns ± 3%   -4.63%        (p=0.000 n=10+10)
MutexWork-6              136ns ± 3%    140ns ± 5%     ~           (p=0.077 n=10+10)
MutexWorkSlack-6         152ns ± 0%    138ns ± 2%   -9.21%         (p=0.000 n=6+10)
MutexNoSpin-6            150ns ± 1%    152ns ± 0%   +1.50%         (p=0.000 n=8+10)
MutexSpin-6              726ns ± 0%    730ns ± 1%     ~           (p=0.069 n=10+10)
RWMutexWrite100-6       40.6ns ± 1%   40.9ns ± 1%   +0.91%         (p=0.001 n=8+10)
RWMutexWrite10-6        37.1ns ± 0%   37.0ns ± 1%     ~            (p=0.386 n=9+10)
RWMutexWorkWrite100-6    133ns ± 1%    134ns ± 1%   +1.01%         (p=0.005 n=9+10)
RWMutexWorkWrite10-6     152ns ± 0%    152ns ± 0%     ~     (all samples are equal)
Cond32-2                7.86µs ± 2%   7.95µs ± 2%   +1.10%        (p=0.023 n=10+10)
MutexUncontended-2      8.10ns ± 0%   9.11ns ± 4%  +12.44%         (p=0.000 n=9+10)
Mutex-2                 32.9ns ± 9%   38.4ns ± 6%  +16.58%        (p=0.000 n=10+10)
MutexSlack-2            93.4ns ± 1%   98.5ns ± 2%   +5.39%         (p=0.000 n=10+9)
MutexWork-2             40.8ns ± 3%   43.8ns ± 7%   +7.38%         (p=0.000 n=10+9)
MutexWorkSlack-2        98.6ns ± 5%  108.2ns ± 2%   +9.80%         (p=0.000 n=10+8)
MutexNoSpin-2            399ns ± 1%    398ns ± 2%     ~             (p=0.463 n=8+9)
MutexSpin-2             1.99µs ± 3%   1.97µs ± 1%   -0.81%          (p=0.003 n=9+8)
RWMutexWrite100-2       37.6ns ± 5%   46.0ns ± 4%  +22.17%         (p=0.000 n=10+8)
RWMutexWrite10-2        50.1ns ± 6%   36.8ns ±12%  -26.46%         (p=0.000 n=9+10)
RWMutexWorkWrite100-2    136ns ± 0%    134ns ± 2%   -1.80%          (p=0.001 n=7+9)
RWMutexWorkWrite10-2     140ns ± 1%    138ns ± 1%   -1.50%        (p=0.000 n=10+10)
Cond32                  5.93µs ± 1%   5.91µs ± 0%     ~            (p=0.411 n=9+10)
MutexUncontended        15.9ns ± 0%   15.8ns ± 0%   -0.63%          (p=0.000 n=8+8)
Mutex                   15.9ns ± 0%   15.8ns ± 0%   -0.44%        (p=0.003 n=10+10)
MutexSlack              26.9ns ± 3%   26.7ns ± 2%     ~           (p=0.084 n=10+10)
MutexWork               47.8ns ± 0%   47.9ns ± 0%   +0.21%          (p=0.014 n=9+8)
MutexWorkSlack          54.9ns ± 3%   54.5ns ± 3%     ~           (p=0.254 n=10+10)
MutexNoSpin              786ns ± 2%    765ns ± 1%   -2.66%        (p=0.000 n=10+10)
MutexSpin               3.87µs ± 1%   3.83µs ± 0%   -0.85%          (p=0.005 n=9+8)
RWMutexWrite100         21.2ns ± 2%   21.0ns ± 1%   -0.88%         (p=0.018 n=10+9)
RWMutexWrite10          22.6ns ± 1%   22.6ns ± 0%     ~             (p=0.471 n=9+9)
RWMutexWorkWrite100      132ns ± 0%    132ns ± 0%     ~     (all samples are equal)
RWMutexWorkWrite10       124ns ± 0%    123ns ± 0%     ~           (p=0.656 n=10+10)

Change-Id: I66412a3a0980df1233ad7a5a0cd9723b4274528b
Reviewed-on: https://go-review.googlesource.com/34310
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
8 years agosyscall: only call setgroups if we need to
Wander Lairson Costa [Fri, 10 Feb 2017 06:10:48 +0000 (04:10 -0200)]
syscall: only call setgroups if we need to

If the caller set ups a Credential in os/exec.Command,
os/exec.Command.Start will end up calling setgroups(2), even if no
supplementary groups were given.

Only root can call setgroups(2) on BSD kernels, which causes Start to
fail for non-root users when they try to set uid and gid for the new
process.

We fix by introducing a new field to syscall.Credential named
NoSetGroups, and setgroups(2) is only called if it is false.
We make this field with inverted logic to preserve backward
compatibility.

RELNOTES=yes

Change-Id: I3cff1f21c117a1430834f640ef21fd4e87e06804
Reviewed-on: https://go-review.googlesource.com/36697
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/compile: move constant divide strength reduction to SSA rules
Keith Randall [Tue, 14 Feb 2017 00:00:09 +0000 (16:00 -0800)]
cmd/compile: move constant divide strength reduction to SSA rules

Currently the conversion from constant divides to multiplies is mostly
done during the walk pass.  This is suboptimal because SSA can
determine that the value being divided by is constant more often
(e.g. after inlining).

Change-Id: If1a9b993edd71be37396b9167f77da271966f85f
Reviewed-on: https://go-review.googlesource.com/37015
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: simplify needwritebarrier
Matthew Dempsky [Thu, 16 Feb 2017 02:43:34 +0000 (18:43 -0800)]
cmd/compile: simplify needwritebarrier

Currently, whether we need a write barrier is simply a property of the
pointer slot being written to.

The only optimization we currently apply using the value being written
is that pointers to stack variables can omit write barriers because
they're only written to stack slots... but we already omit write
barriers for all writes to the stack anyway.

Passes toolstash -cmp.

Change-Id: I7f16b71ff473899ed96706232d371d5b2b7ae789
Reviewed-on: https://go-review.googlesource.com/37109
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agomath: fix typos in Bessel function docs
Shenghou Ma [Sat, 28 Jan 2017 07:01:50 +0000 (02:01 -0500)]
math: fix typos in Bessel function docs

While we're at it, also document Yn(0, 0) = -Inf for completeness.

Fixes #18823.

Change-Id: Ib6db68f76d29cc2373c12ebdf3fab129cac8c167
Reviewed-on: https://go-review.googlesource.com/35970
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agomath/bits: added package for bit-level counting and manipulation
Robert Griesemer [Fri, 3 Feb 2017 00:59:34 +0000 (16:59 -0800)]
math/bits: added package for bit-level counting and manipulation

Initial platform-independent implementation.

For #18616.

Change-Id: I4585c55b963101af9059c06c1b8a866cb384754c
Reviewed-on: https://go-review.googlesource.com/36315
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
8 years agocmd/compile/internal/syntax: better errors and recovery for invalid character literals
Robert Griesemer [Thu, 16 Feb 2017 20:52:01 +0000 (12:52 -0800)]
cmd/compile/internal/syntax: better errors and recovery for invalid character literals

Fixes #15611.

Change-Id: I352b145026466cafef8cf87addafbd30716bda24
Reviewed-on: https://go-review.googlesource.com/37138
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoruntime: use balanced tree for addr lookup in semaphore implementation
Russ Cox [Sun, 12 Feb 2017 18:19:02 +0000 (13:19 -0500)]
runtime: use balanced tree for addr lookup in semaphore implementation

CL 36792 fixed #17953, a linear scan caused by n goroutines piling into
two different locks that hashed to the same bucket in the semaphore table.
In that CL, n goroutines contending for 2 unfortunately chosen locks
went from O(n²) to O(n).

This CL fixes a different linear scan, when n goroutines are contending for
n/2 different locks that all hash to the same bucket in the semaphore table.
In this CL, n goroutines contending for n/2 unfortunately chosen locks
goes from O(n²) to O(n log n). This case is much less likely, but any linear
scan eventually hurts, so we might as well fix it while the problem is fresh
in our minds.

The new test in this CL checks for both linear scans.

The effect of this CL on the sync benchmarks is negligible
(but it fixes the new test).

name                      old time/op    new time/op    delta
Cond1-48                     576ns ±10%     575ns ±13%     ~     (p=0.679 n=71+71)
Cond2-48                    1.59µs ± 8%    1.61µs ± 9%     ~     (p=0.107 n=73+69)
Cond4-48                    4.56µs ± 7%    4.55µs ± 7%     ~     (p=0.670 n=74+72)
Cond8-48                    9.87µs ± 9%    9.90µs ± 7%     ~     (p=0.507 n=69+73)
Cond16-48                   20.4µs ± 7%    20.4µs ±10%     ~     (p=0.588 n=69+71)
Cond32-48                   45.4µs ±10%    45.4µs ±14%     ~     (p=0.944 n=73+73)
UncontendedSemaphore-48     19.7ns ±12%    19.7ns ± 8%     ~     (p=0.589 n=65+63)
ContendedSemaphore-48       55.4ns ±26%    54.9ns ±32%     ~     (p=0.441 n=75+75)
MutexUncontended-48         0.63ns ± 0%    0.63ns ± 0%     ~     (all equal)
Mutex-48                     210ns ± 6%     213ns ±10%   +1.30%  (p=0.035 n=70+74)
MutexSlack-48                210ns ± 7%     211ns ± 9%     ~     (p=0.184 n=71+72)
MutexWork-48                 299ns ± 5%     300ns ± 5%     ~     (p=0.678 n=73+75)
MutexWorkSlack-48            302ns ± 6%     300ns ± 5%     ~     (p=0.149 n=74+72)
MutexNoSpin-48               135ns ± 6%     135ns ±10%     ~     (p=0.788 n=67+75)
MutexSpin-48                 693ns ± 5%     689ns ± 6%     ~     (p=0.092 n=65+74)
Once-48                     0.22ns ±25%    0.22ns ±24%     ~     (p=0.882 n=74+73)
Pool-48                     5.88ns ±36%    5.79ns ±24%     ~     (p=0.655 n=69+69)
PoolOverflow-48             4.79µs ±18%    4.87µs ±20%     ~     (p=0.233 n=75+75)
SemaUncontended-48          0.80ns ± 1%    0.82ns ± 8%   +2.46%  (p=0.000 n=60+74)
SemaSyntNonblock-48          103ns ± 4%     102ns ± 5%   -1.11%  (p=0.003 n=75+75)
SemaSyntBlock-48             104ns ± 4%     104ns ± 5%     ~     (p=0.231 n=71+75)
SemaWorkNonblock-48          128ns ± 4%     129ns ± 6%   +1.51%  (p=0.000 n=63+75)
SemaWorkBlock-48             129ns ± 8%     130ns ± 7%     ~     (p=0.072 n=75+74)
RWMutexUncontended-48       2.35ns ± 1%    2.35ns ± 0%     ~     (p=0.144 n=70+55)
RWMutexWrite100-48           139ns ±18%     141ns ±21%     ~     (p=0.071 n=75+73)
RWMutexWrite10-48            145ns ± 9%     145ns ± 8%     ~     (p=0.553 n=75+75)
RWMutexWorkWrite100-48       297ns ±13%     297ns ±15%     ~     (p=0.519 n=75+74)
RWMutexWorkWrite10-48        588ns ± 7%     585ns ± 5%     ~     (p=0.173 n=73+70)
WaitGroupUncontended-48     0.87ns ± 0%    0.87ns ± 0%     ~     (all equal)
WaitGroupAddDone-48         63.2ns ± 4%    62.7ns ± 4%   -0.82%  (p=0.027 n=72+75)
WaitGroupAddDoneWork-48      109ns ± 5%     109ns ± 4%     ~     (p=0.233 n=75+75)
WaitGroupWait-48            0.17ns ± 0%    0.16ns ±16%   -8.55%  (p=0.000 n=56+75)
WaitGroupWaitWork-48        1.78ns ± 1%    2.08ns ± 5%  +16.92%  (p=0.000 n=74+70)
WaitGroupActuallyWait-48    52.0ns ± 3%    50.6ns ± 5%   -2.70%  (p=0.000 n=71+69)

https://perf.golang.org/search?q=upload:20170215.1

Change-Id: Ia29a8bd006c089e401ec4297c3038cca656bcd0a
Reviewed-on: https://go-review.googlesource.com/37103
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>