]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agocmd/compile: add s390x intrinsics for Ceil, Floor, Round and Trunc
Michael Munday [Thu, 14 Sep 2017 19:00:02 +0000 (20:00 +0100)]
cmd/compile: add s390x intrinsics for Ceil, Floor, Round and Trunc

Ceil, Floor and Trunc are pre-existing intrinsics. Round is a new
function and has been added as an intrinsic in this CL. All of the
functions can be implemented as a single 'LOAD FP INTEGER'
instruction, FIDBR, on s390x.

name   old time/op  new time/op  delta
Ceil   2.34ns ± 0%  0.85ns ± 0%  -63.74%  (p=0.000 n=5+4)
Floor  2.33ns ± 0%  0.85ns ± 1%  -63.35%  (p=0.008 n=5+5)
Round  4.23ns ± 0%  0.85ns ± 0%  -79.89%  (p=0.000 n=5+4)
Trunc  2.35ns ± 0%  0.85ns ± 0%  -63.83%  (p=0.029 n=4+4)

Change-Id: Idee7ba24a2899d12bf9afee4eedd6b4aaad3c510
Reviewed-on: https://go-review.googlesource.com/63890
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agofmt: Implement pp.WriteString method
Rajath Agasthya [Sun, 20 Aug 2017 01:05:10 +0000 (18:05 -0700)]
fmt: Implement pp.WriteString method

This allows io.WriteString to make use of WriteString method
implemented by pp when writing a string to fmt.State.

Fixes #20786

Change-Id: Ice7a92bf303127ad87f05562217fa076f5c589ad
Reviewed-on: https://go-review.googlesource.com/61430
Reviewed-by: Rob Pike <r@golang.org>
7 years agoall: correct location of go tool
Hiroshi Ioka [Tue, 19 Sep 2017 09:18:09 +0000 (18:18 +0900)]
all: correct location of go tool

In general, there are no guarantee that `go` command exist on $PATH.
This CL tries to get `go` command from $GOROOT/bin instead.

There are three kinds of code we should handle:
    For normal code, the CL implements goCmd() or goCmdName().
    For unit tests, the CL uses testenv.GoTool() or testenv.GoToolPath().
    For integration tests, the CL sets PATH=$GOROOT/bin:$PATH in cmd/dist.

Note that make.bash sets PATH=$GOROOT/bin:$PATH in the build process.
So this change is only useful when we use toolchain manually.

Updates #21875

Change-Id: I963b9f22ea732dd735363ececde4cf94a5db5ca2
Reviewed-on: https://go-review.googlesource.com/64650
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/dist: use bytes.Buffer for code generation
Marvin Stenger [Fri, 1 Sep 2017 16:17:10 +0000 (18:17 +0200)]
cmd/dist: use bytes.Buffer for code generation

This belongs to a series of clean-up changes (see below) for cmd/dist.
This is change (8).

These changes include:
(1)  apply minor fixes
(2)  restore behavior of branchtag
(3)  unleash bootstrap optimization for windows
(4)  use standard generated code header
(5)  remove trivial variables + functions
(6)  move functions for the better
(7)  simplify code segments
(8)  use bytes.Buffer for code generation
(9)  rename variables + functions
(10) remove doc.go

Change-Id: I2d5a071eb8e14690325612271432fdc5f43b108b
Reviewed-on: https://go-review.googlesource.com/61014
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/nm: add test case for go archives
Hiroshi Ioka [Tue, 19 Sep 2017 06:58:40 +0000 (15:58 +0900)]
cmd/nm: add test case for go archives

Also, rename some test cases, check (*os.File).Close

For #21706

Change-Id: Ie60c4d345b2259736c823dc6001c08affcdd86e7
Reviewed-on: https://go-review.googlesource.com/64510
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agohtml/template: prevent aliasing of parse Trees via AddParseTree
Samuel Tan [Tue, 19 Sep 2017 18:54:47 +0000 (11:54 -0700)]
html/template: prevent aliasing of parse Trees via AddParseTree

Check all associated templates in the set for an existing reference
to the given Tree in AddParseTree before assigning that reference
to a new or existing template. This prevents multiple html/template
Templates from referencing and modifying the same underlying Tree.

While there, fix a few existing unit tests so that they terminate
upon encountering unrecoverable failures.

Fixes #21844

Change-Id: I6b4f6996cf5467113ef94f7b91a6933dbbc21839
Reviewed-on: https://go-review.googlesource.com/64770
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
7 years agocmd/dist: simplify code segments
Marvin Stenger [Thu, 31 Aug 2017 10:50:57 +0000 (12:50 +0200)]
cmd/dist: simplify code segments

This belongs to a series of clean-up changes (see below) for cmd/dist.
This is change (7).

These changes include:
(1)  apply minor fixes
(2)  restore behavior of branchtag
(3)  unleash bootstrap optimization for windows
(4)  use standard generated code header
(5)  remove trivial variables + functions
(6)  move functions for the better
(7)  simplify code segments
(8)  use bytes.Buffer for code generation
(9)  rename variables + functions
(10) remove doc.go

Change-Id: Ia3c33ef060b4baaef354b729ba82ed0b28e52857
Reviewed-on: https://go-review.googlesource.com/61013
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/vet: don't warn on expected space in XML tag
Sam Whited [Sat, 9 Sep 2017 15:49:04 +0000 (10:49 -0500)]
cmd/vet: don't warn on expected space in XML tag

The change in https://golang.org/cl/43295 added warning about spaces in
struct tags. However, in XML tags it is expected that there will be a
space between the namespace and the local name.

Change-Id: Ic31c3bdae30797f406f25c737b83bbe2de1ed1db
Reviewed-on: https://go-review.googlesource.com/62570
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
7 years agointernal/testenv: take testing.TB instead of *testing.T in MustHave* and SkipFlaky*
Hiroshi Ioka [Tue, 19 Sep 2017 15:04:50 +0000 (00:04 +0900)]
internal/testenv: take testing.TB instead of *testing.T in MustHave* and SkipFlaky*

Change-Id: I16475e9bb055b934302870ccb5136174dc3bc817
Reviewed-on: https://go-review.googlesource.com/64670
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: remove Symbol wrappers from Aux fields
Keith Randall [Mon, 18 Sep 2017 21:53:56 +0000 (14:53 -0700)]
cmd/compile: remove Symbol wrappers from Aux fields

We used to have {Arg,Auto,Extern}Symbol structs with which we wrapped
a *gc.Node or *obj.LSym before storing them in the Aux field
of an ssa.Value.  This let the SSA part of the compiler distinguish
between autos and args, for example.  We no longer need the wrappers
as we can query the underlying objects directly.

There was also some sloppy usage, where VarDef had a *gc.Node
directly in its Aux field, whereas the use of that variable had
that *gc.Node wrapped in an AutoSymbol. Thus the Aux fields didn't
match (using ==) when they probably should.
This sloppy usage cleanup is the only thing in the CL that changes the
generated code - we can get rid of some more unused auto variables if
the matching happens reliably.

Removing this wrapper also lets us get rid of the varsyms cache
(which was used to prevent wrapping the same *gc.Node twice).

Change-Id: I0dedf8f82f84bfee413d310342b777316bd1d478
Reviewed-on: https://go-review.googlesource.com/64452
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoreflect: fix pointer past-the-end in Call with zero-sized return value
Cherry Zhang [Fri, 1 Sep 2017 02:02:37 +0000 (22:02 -0400)]
reflect: fix pointer past-the-end in Call with zero-sized return value

If a function with nonzero frame but zero-sized return value is
Call'd, we may write a past-the-end pointer in preparing the
return Values. Fix by return the zero value for zero-sized
return value.

Fixes #21717.

Change-Id: I5351cd86d898467170a888b4c3fc9392f0e7aa3b
Reviewed-on: https://go-review.googlesource.com/60811
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agolog: fix data race on log.Output
hagen1778 [Tue, 19 Sep 2017 17:28:11 +0000 (20:28 +0300)]
log: fix data race on log.Output

There was unprotected access to Logger.flag in log.Output which
could lead to data race in cases when log.SetFlags called simultaneously.
For example, "hot" switching on/off debug-mode for Logger by log.SetFlags
while application still writing logs.

Fixes #21935

Change-Id: I36be25f23cad44cde62ed1af28a30d276400e1b8
Reviewed-on: https://go-review.googlesource.com/64710
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix stack frame info for calls in receiver slot
Matthew Dempsky [Tue, 19 Sep 2017 06:02:02 +0000 (23:02 -0700)]
cmd/compile: fix stack frame info for calls in receiver slot

Previously, after inlining a call, we made a second pass to rewrite
the AST's position information to record the inlined stack frame. The
call arguments were part of this AST, but it would be incorrect to
rewrite them too, so extra effort was made to temporarily remove them
while the position rewriting was done.

However, this extra logic was only done for regular arguments: it was
not done for receiver arguments. Consequently if m was inlined in
"f().m(g(), h())", g and h would have correct call frames, but f would
appear to be called by m.

The fix taken by this CL is to merge setpos into inlsubst and only
rewrite position information for nodes that were actually copied from
the original function AST body. As a side benefit, this eliminates an
extra AST pass and some AST walking code.

Fixes #21879.

Change-Id: I22b25c208313fc25c358d3a2eebfc9b012400084
Reviewed-on: https://go-review.googlesource.com/64470
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/cgo: use a named type to indicate syntactic context
Bryan C. Mills [Thu, 14 Sep 2017 15:36:21 +0000 (11:36 -0400)]
cmd/cgo: use a named type to indicate syntactic context

We previously used bare strings, which made it difficult to see (and
to cross-reference) the set of allowed context values.

This change is purely cosmetic, but makes it easier for me to
understand how to address #21878.

updates #21878

Change-Id: I9027d94fd5997a0fe857c0055dea8719e1511f03
Reviewed-on: https://go-review.googlesource.com/63830
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: cleanup toolstash pacifier from OXFALL removal
Matthew Dempsky [Fri, 1 Sep 2017 21:58:32 +0000 (14:58 -0700)]
cmd/compile: cleanup toolstash pacifier from OXFALL removal

Change-Id: Ide7fe6b09247b7a6befbdfc2d6ce5988aa1df323
Reviewed-on: https://go-review.googlesource.com/61131
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: eliminate OXFALL
Matthew Dempsky [Fri, 1 Sep 2017 21:55:15 +0000 (14:55 -0700)]
cmd/compile: eliminate OXFALL

Previously, we used OXFALL vs OFALL to distinguish fallthrough
statements that had been validated. Because in the Node AST we flatten
statement blocks, OXCASE and OXFALL needed to keep track of their
block scopes for this purpose.

Now that we have an AST that keeps these separate, we can just perform
the validation earlier.

Passes toolstash-check.

Fixes #14540.

Change-Id: I8421eaba16c2b3b72c9c5483b5cf20b14261385e
Reviewed-on: https://go-review.googlesource.com/61130
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/dist: test: use existing globals rather than environment variables
Hiroshi Ioka [Tue, 19 Sep 2017 13:13:22 +0000 (22:13 +0900)]
cmd/dist: test: use existing globals rather than environment variables

Change-Id: Ief6bad2d15461d455e7230eadd9b42b27d04ec8b
Reviewed-on: https://go-review.googlesource.com/64630
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agospec: state which predeclared types are defined or alias types (clarification)
griesemer [Tue, 19 Sep 2017 12:35:15 +0000 (14:35 +0200)]
spec: state which predeclared types are defined or alias types (clarification)

When we introduced the distinction between "defined" and "alias" types
we retained the notion of a "named" type (any type with a name). The
predeclared types (which all have names) simply remained named types.

This CL clarifies the spec by stating excplicitly which predeclared
types are defined types (or at least "act" like defined types), and
which ones are alias types.

Fixes #21785.

Change-Id: Ia8ae133509eb5d738e6757b3442c9992355e3535
Reviewed-on: https://go-review.googlesource.com/64591
Reviewed-by: Russ Cox <rsc@golang.org>
7 years agounicode: allow version to be passed by env var
Marcel van Lohuizen [Wed, 13 Sep 2017 15:45:51 +0000 (17:45 +0200)]
unicode: allow version to be passed by env var

This, in turn, to make it work with x/text’s
go generate.

Also eliminates need to manually update version
string in maketables.go.

Change-Id: Id5a8b8e27bdce5b1b5920eb9223a2d27b889149a
Reviewed-on: https://go-review.googlesource.com/63952
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Nigel Tao <nigeltao@golang.org>
7 years agocmd/go: fix file leak in TestBuildmodePIE
Hiroshi Ioka [Tue, 19 Sep 2017 07:31:03 +0000 (16:31 +0900)]
cmd/go: fix file leak in TestBuildmodePIE

Change-Id: Ifeb93f6dc75fb5f90d595211fb0f97a89bf74526
Reviewed-on: https://go-review.googlesource.com/64530
Reviewed-by: Dave Cheney <dave@cheney.net>
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agodoc: simplify Append example in "Effective Go"
Gabriel Aszalos [Mon, 18 Sep 2017 14:51:15 +0000 (16:51 +0200)]
doc: simplify Append example in "Effective Go"

Change-Id: I011486993b167e65c69da1c8390bbcc625ca58c3
Reviewed-on: https://go-review.googlesource.com/64331
Reviewed-by: Rob Pike <r@golang.org>
7 years agocmd/compile: simplify range expression
Tobias Klauser [Mon, 18 Sep 2017 13:20:31 +0000 (15:20 +0200)]
cmd/compile: simplify range expression

Found by running gofmt -s on the file in question.

Change-Id: I84511bd2bc75dff196930a7a87ecf5a2aca2fbb8
Reviewed-on: https://go-review.googlesource.com/64310
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix compiler crash on recursive types
Matthew Dempsky [Mon, 18 Sep 2017 18:54:03 +0000 (11:54 -0700)]
cmd/compile: fix compiler crash on recursive types

By setting both a valid size and alignment for broken recursive types,
we can appease some more safety checks and prevent compiler crashes.

Fixes #21882.

Change-Id: Ibaa137d8aa2c2a9d521462f144d7016c4abfd6e7
Reviewed-on: https://go-review.googlesource.com/64430
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agohtml/template: explain URL filtering
Samuel Tan [Wed, 2 Aug 2017 20:45:43 +0000 (13:45 -0700)]
html/template: explain URL filtering

Expand documentation in of the internal urlFilter function
to explain why URLs with schemes other than "http", "https",
and "mailto" are filtered out.

Fixes #20586

Change-Id: I1f65ff6e15fc4cd325489327c40f8c141904bf5c
Reviewed-on: https://go-review.googlesource.com/52853
Reviewed-by: Mike Samuel <mikesamuel@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agogo/types, constant: remove superfluous import comment
griesemer [Mon, 18 Sep 2017 09:49:47 +0000 (11:49 +0200)]
go/types, constant: remove superfluous import comment

The comment was a left-over from the long-past move
of these two packages from x/tools to the std lib.

Fixes #21791.

Change-Id: I65cbebf479e609be0204b58edb6506c6403aec9b
Reviewed-on: https://go-review.googlesource.com/64250
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agobytes: removed unnecessary slicing on copy
Gabriel Aszalos [Mon, 18 Sep 2017 15:07:21 +0000 (17:07 +0200)]
bytes: removed unnecessary slicing on copy

Change-Id: Ia42e3479c852a88968947411de8b32e5bcda5ae3
Reviewed-on: https://go-review.googlesource.com/64371
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoarchive/tar: add Reader.WriteTo and Writer.ReadFrom
Joe Tsai [Thu, 31 Aug 2017 21:52:20 +0000 (14:52 -0700)]
archive/tar: add Reader.WriteTo and Writer.ReadFrom

To support the efficient packing and extracting of sparse files,
add two new methods:
func Reader.WriteTo(io.Writer) (int64, error)
func Writer.ReadFrom(io.Reader) (int64, error)

If the current archive entry is sparse and the provided io.{Reader,Writer}
is also an io.Seeker, then use Seek to skip past the holes.
If the last region in a file entry is a hole, then we seek to 1 byte
before the EOF:
* for Reader.WriteTo to write a single byte
to ensure that the resulting filesize is correct.
* for Writer.ReadFrom to read a single byte
to verify that the input filesize is correct.

The downside of this approach is when the last region in the sparse file
is a hole. In the case of Reader.WriteTo, the 1-byte write will cause
the last fragment to have a single chunk allocated.
However, the goal of ReadFrom/WriteTo is *not* the ability to
exactly reproduce sparse files (in terms of the location of sparse holes),
but rather to provide an efficient way to create them.

File systems already impose their own restrictions on how the sparse file
will be created. Some filesystems (e.g., HFS+) don't support sparseness and
seeking forward simply causes the FS to write zeros. Other filesystems
have different chunk sizes, which will cause chunk allocations at boundaries
different from what was in the original sparse file. In either case,
it should not be a normal expectation of users that the location of holes
in sparse files exactly matches the source.

For users that really desire to have exact reproduction of sparse holes,
they can wrap os.File with their own io.WriteSeeker that discards the
final 1-byte write and uses File.Truncate to resize the file to the
correct size.

Other reasons we choose this approach over special-casing *os.File because:
* The Reader already has special-case logic for io.Seeker
* As much as possible, we want to decouple OS-specific logic from
Reader and Writer.
* This allows other abstractions over *os.File to also benefit from
the "skip past holes" logic.
* It is easier to test, since it is harder to mock an *os.File.

Updates #13548

Change-Id: I0a4f293bd53d13d154a946bc4a2ade28a6646f6a
Reviewed-on: https://go-review.googlesource.com/60872
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/internal/obj/x86: add ADDSUBPS/PD
Damien Lespiau [Tue, 9 May 2017 05:44:25 +0000 (06:44 +0100)]
cmd/internal/obj/x86: add ADDSUBPS/PD

These are the last instructions missing to complete SSE3 support.

For reference what was missing was found by a tool [1]:

$ x86db-gogen list --extension SSE3 --not-known
ADDSUBPD xmmreg,xmmrm [rm: 66 0f d0 /r] PRESCOTT,SSE3,SO
ADDSUBPS xmmreg,xmmrm [rm: f2 0f d0 /r] PRESCOTT,SSE3,SO

[1] https://github.com/dlespiau/x86db

Fixes #20293

Change-Id: Ib5a91bf64dcc5282cdb60eae740ae52b4db16ebd
Reviewed-on: https://go-review.googlesource.com/42990
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agocmd/compile: remove some redundant types in decls
Daniel Martí [Sat, 16 Sep 2017 21:11:43 +0000 (22:11 +0100)]
cmd/compile: remove some redundant types in decls

As per golint's suggestions.

Change-Id: Ie0c6ad9aa5dc69966a279562a341c7b095c47ede
Reviewed-on: https://go-review.googlesource.com/64192
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agobytes: add example for Len function of Reader
Andrzej Żeżel [Sat, 16 Sep 2017 14:58:10 +0000 (16:58 +0200)]
bytes: add example for Len function of Reader

Change-Id: If7ecdc57f190f647bfc673bde8e66b4ef12aa906
Reviewed-on: https://go-review.googlesource.com/64190
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: improve fastrand with a better generator
Giovanni Bajo [Sat, 9 Sep 2017 12:59:06 +0000 (14:59 +0200)]
runtime: improve fastrand with a better generator

The current generator is a simple LSFR, which showed strong
correlation in higher bits, as manifested by fastrandn().

Change it with xorshift64+, which is slightly more complex,
has a larger state, but has a period of 2^64-1 and is much better
at statistical tests. The version used here is capable of
passing Diehard and even SmallCrush.

Speed is slightly worse but is probably insignificant:

name                old time/op  new time/op  delta
Fastrand-4          0.77ns ±12%  0.91ns ±21%  +17.31%  (p=0.048 n=5+5)
FastrandHashiter-4  13.6ns ±21%  15.2ns ±17%     ~     (p=0.160 n=6+5)
Fastrandn/2-4       2.30ns ± 5%  2.45ns ±15%     ~     (p=0.222 n=5+5)
Fastrandn/3-4       2.36ns ± 7%  2.45ns ± 6%     ~     (p=0.222 n=5+5)
Fastrandn/4-4       2.33ns ± 8%  2.61ns ±30%     ~     (p=0.126 n=6+5)
Fastrandn/5-4       2.33ns ± 5%  2.48ns ± 9%     ~     (p=0.052 n=6+5)

Fixes #21806

Change-Id: I013bb37b463fdfc229a7f324df8fe2da8d286f33
Reviewed-on: https://go-review.googlesource.com/62530
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: test constant folded integer to/from float conversions
Michael Munday [Fri, 15 Sep 2017 20:50:28 +0000 (21:50 +0100)]
cmd/compile: test constant folded integer to/from float conversions

Improves test coverage of the rules added in CL 63795 and would have
detected the bug fixed by CL 63950.

Change-Id: I107ee8d8e0b6684ce85b2446bd5018c5a03d608a
Reviewed-on: https://go-review.googlesource.com/64130
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: optimize ARM code with MULAF/MULSF/MULAD/MULSD
Ben Shi [Thu, 14 Sep 2017 06:52:51 +0000 (06:52 +0000)]
cmd/compile: optimize ARM code with MULAF/MULSF/MULAD/MULSD

The go compiler can generate better ARM code with those more
efficient FP instructions. And there is little improvement
in total but big improvement in special cases.

1. The size of pkg/linux_arm/math.a shrinks by 2.4%.

2. there is neither improvement nor regression in compilecmp benchmark.
name        old time/op       new time/op       delta
Template          2.32s ± 2%        2.32s ± 1%    ~     (p=1.000 n=9+10)
Unicode           1.32s ± 4%        1.32s ± 4%    ~     (p=0.912 n=10+10)
GoTypes           7.76s ± 1%        7.79s ± 1%    ~     (p=0.447 n=9+10)
Compiler          37.4s ± 2%        37.2s ± 2%    ~     (p=0.218 n=10+10)
SSA               84.8s ± 2%        85.0s ± 1%    ~     (p=0.604 n=10+9)
Flate             1.45s ± 2%        1.44s ± 2%    ~     (p=0.075 n=10+10)
GoParser          1.82s ± 1%        1.81s ± 1%    ~     (p=0.190 n=10+10)
Reflect           5.06s ± 1%        5.05s ± 1%    ~     (p=0.315 n=10+9)
Tar               2.37s ± 1%        2.37s ± 2%    ~     (p=0.912 n=10+10)
XML               2.56s ± 1%        2.58s ± 2%    ~     (p=0.089 n=10+10)
[Geo mean]        4.77s             4.77s       -0.08%

name        old user-time/op  new user-time/op  delta
Template          2.74s ± 2%        2.75s ± 2%    ~     (p=0.856 n=9+10)
Unicode           1.61s ± 4%        1.62s ± 3%    ~     (p=0.693 n=10+10)
GoTypes           9.55s ± 1%        9.49s ± 2%    ~     (p=0.056 n=9+10)
Compiler          45.9s ± 1%        45.8s ± 1%    ~     (p=0.345 n=9+10)
SSA                110s ± 1%         110s ± 1%    ~     (p=0.763 n=9+10)
Flate             1.68s ± 2%        1.68s ± 3%    ~     (p=0.616 n=10+10)
GoParser          2.14s ± 4%        2.14s ± 1%    ~     (p=0.825 n=10+9)
Reflect           5.95s ± 1%        5.97s ± 3%    ~     (p=0.951 n=9+10)
Tar               2.94s ± 3%        2.93s ± 2%    ~     (p=0.359 n=10+10)
XML               3.03s ± 3%        3.07s ± 6%    ~     (p=0.166 n=10+10)
[Geo mean]        5.76s             5.77s       +0.12%

name        old text-bytes    new text-bytes    delta
HelloSize         588kB ± 0%        588kB ± 0%    ~     (all equal)

name        old data-bytes    new data-bytes    delta
HelloSize        5.46kB ± 0%       5.46kB ± 0%    ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize        72.9kB ± 0%       72.9kB ± 0%    ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.03MB ± 0%       1.03MB ± 0%    ~     (all equal)

3. The performance of Mandelbrot200 improves 15%, though little
   improvement in total.
name                     old time/op    new time/op    delta
BinaryTree17-4              41.7s ± 1%     41.7s ± 1%     ~     (p=0.264 n=29+23)
Fannkuch11-4                24.2s ± 0%     24.1s ± 1%   -0.13%  (p=0.050 n=30+30)
FmtFprintfEmpty-4           826ns ± 1%     824ns ± 1%   -0.24%  (p=0.038 n=25+30)
FmtFprintfString-4         1.38µs ± 1%    1.38µs ± 0%   -0.42%  (p=0.000 n=27+25)
FmtFprintfInt-4            1.46µs ± 1%    1.46µs ± 0%     ~     (p=0.060 n=30+23)
FmtFprintfIntInt-4         2.11µs ± 1%    2.08µs ± 0%   -1.04%  (p=0.000 n=30+30)
FmtFprintfPrefixedInt-4    2.23µs ± 1%    2.22µs ± 1%   -0.51%  (p=0.000 n=30+30)
FmtFprintfFloat-4          4.49µs ± 1%    4.48µs ± 1%   -0.22%  (p=0.004 n=26+30)
FmtManyArgs-4              8.06µs ± 1%    8.12µs ± 1%   +0.68%  (p=0.000 n=25+30)
GobDecode-4                 104ms ± 1%     104ms ± 2%     ~     (p=0.362 n=29+29)
GobEncode-4                92.9ms ± 1%    92.8ms ± 2%     ~     (p=0.786 n=30+30)
Gzip-4                      4.12s ± 1%     4.12s ± 1%     ~     (p=0.314 n=30+30)
Gunzip-4                    602ms ± 1%     603ms ± 1%     ~     (p=0.164 n=30+30)
HTTPClientServer-4          659µs ± 1%     655µs ± 2%   -0.64%  (p=0.006 n=25+28)
JSONEncode-4                234ms ± 1%     235ms ± 1%   +0.29%  (p=0.050 n=30+30)
JSONDecode-4                912ms ± 0%     911ms ± 0%     ~     (p=0.385 n=18+24)
Mandelbrot200-4            49.2ms ± 0%    41.7ms ± 0%  -15.35%  (p=0.000 n=25+27)
GoParse-4                  46.3ms ± 1%    46.3ms ± 2%     ~     (p=0.572 n=30+30)
RegexpMatchEasy0_32-4      1.29µs ± 1%    1.27µs ± 0%   -1.59%  (p=0.000 n=30+30)
RegexpMatchEasy0_1K-4      7.62µs ± 4%    7.71µs ± 3%     ~     (p=0.074 n=30+30)
RegexpMatchEasy1_32-4      1.31µs ± 0%    1.30µs ± 1%   -0.71%  (p=0.000 n=23+30)
RegexpMatchEasy1_1K-4      10.3µs ± 3%    10.3µs ± 5%     ~     (p=0.105 n=30+30)
RegexpMatchMedium_32-4     2.06µs ± 1%    2.06µs ± 1%     ~     (p=0.100 n=30+30)
RegexpMatchMedium_1K-4      533µs ± 1%     534µs ± 1%     ~     (p=0.254 n=29+30)
RegexpMatchHard_32-4       28.9µs ± 0%    28.9µs ± 0%     ~     (p=0.154 n=30+30)
RegexpMatchHard_1K-4        868µs ± 1%     867µs ± 0%     ~     (p=0.729 n=30+23)
Revcomp-4                  66.9ms ± 1%    67.2ms ± 2%     ~     (p=0.102 n=28+29)
Template-4                  1.07s ± 1%     1.06s ± 1%   -0.53%  (p=0.000 n=30+30)
TimeParse-4                7.07µs ± 1%    7.01µs ± 0%   -0.85%  (p=0.000 n=30+25)
TimeFormat-4               13.1µs ± 0%    13.2µs ± 1%   +0.77%  (p=0.000 n=27+27)
[Geo mean]                  721µs          716µs        -0.70%

name                     old speed      new speed      delta
GobDecode-4              7.38MB/s ± 1%  7.37MB/s ± 2%     ~     (p=0.399 n=29+29)
GobEncode-4              8.26MB/s ± 1%  8.27MB/s ± 2%     ~     (p=0.790 n=30+30)
Gzip-4                   4.71MB/s ± 1%  4.71MB/s ± 1%     ~     (p=0.885 n=30+30)
Gunzip-4                 32.2MB/s ± 1%  32.2MB/s ± 1%     ~     (p=0.190 n=30+30)
JSONEncode-4             8.28MB/s ± 1%  8.25MB/s ± 1%     ~     (p=0.053 n=30+30)
JSONDecode-4             2.13MB/s ± 0%  2.12MB/s ± 1%     ~     (p=0.072 n=18+30)
GoParse-4                1.25MB/s ± 1%  1.25MB/s ± 2%     ~     (p=0.863 n=30+30)
RegexpMatchEasy0_32-4    24.8MB/s ± 0%  25.2MB/s ± 1%   +1.61%  (p=0.000 n=30+30)
RegexpMatchEasy0_1K-4     134MB/s ± 4%   133MB/s ± 3%     ~     (p=0.074 n=30+30)
RegexpMatchEasy1_32-4    24.5MB/s ± 0%  24.6MB/s ± 1%   +0.72%  (p=0.000 n=23+30)
RegexpMatchEasy1_1K-4    99.1MB/s ± 3%  99.8MB/s ± 5%     ~     (p=0.105 n=30+30)
RegexpMatchMedium_32-4    483kB/s ± 1%   487kB/s ± 1%   +0.83%  (p=0.002 n=30+30)
RegexpMatchMedium_1K-4   1.92MB/s ± 1%  1.92MB/s ± 1%     ~     (p=0.058 n=30+30)
RegexpMatchHard_32-4     1.10MB/s ± 0%  1.11MB/s ± 0%     ~     (p=0.804 n=30+30)
RegexpMatchHard_1K-4     1.18MB/s ± 0%  1.18MB/s ± 0%     ~     (all equal)
Revcomp-4                38.0MB/s ± 1%  37.8MB/s ± 2%     ~     (p=0.098 n=28+29)
Template-4               1.82MB/s ± 1%  1.83MB/s ± 1%   +0.55%  (p=0.000 n=29+29)
[Geo mean]               6.79MB/s       6.79MB/s        +0.09%

Change-Id: Ia91991c2c5c59c5df712de85a83b13a21c0a554b
Reviewed-on: https://go-review.googlesource.com/63770
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/internal/obj: change Prog.From3 to RestArgs ([]Addr)
isharipo [Wed, 13 Sep 2017 11:32:08 +0000 (14:32 +0300)]
cmd/internal/obj: change Prog.From3 to RestArgs ([]Addr)

This change makes it easier to express instructions
with arbitrary number of operands.

Rationale: previous approach with operand "hiding" does
not scale well, AVX and especially AVX512 have many
instructions with 3+ operands.

x86 asm backend is updated to handle up to 6 explicit operands.
It also fixes issue with 4-th immediate operand type checks.
All `ytab` tables are updated accordingly.

Changes to non-x86 backends only include these patterns:
`p.From3 = X` => `p.SetFrom3(X)`
`p.From3.X = Y` => `p.GetFrom3().X = Y`

Over time, other backends can adapt Prog.RestArgs
and reduce the amount of workarounds.

-- Performance --

x/benchmark/build:

$ benchstat upstream.bench patched.bench
name      old time/op                 new time/op                 delta
Build-48                  21.7s ± 2%                  21.8s ± 2%   ~     (p=0.218 n=10+10)

name      old binary-size             new binary-size             delta
Build-48                  10.3M ± 0%                  10.3M ± 0%   ~     (all equal)

name      old build-time/op           new build-time/op           delta
Build-48                  21.7s ± 2%                  21.8s ± 2%   ~     (p=0.218 n=10+10)

name      old build-peak-RSS-bytes    new build-peak-RSS-bytes    delta
Build-48                  145MB ± 5%                  148MB ± 5%   ~     (p=0.218 n=10+10)

name      old build-user+sys-time/op  new build-user+sys-time/op  delta
Build-48                  21.0s ± 2%                  21.2s ± 2%   ~     (p=0.075 n=10+10)

Microbenchmark shows a slight slowdown.

name        old time/op  new time/op  delta
AMD64asm-4  49.5ms ± 1%  49.9ms ± 1%  +0.67%  (p=0.001 n=23+15)

func BenchmarkAMD64asm(b *testing.B) {
  for i := 0; i < b.N; i++ {
    TestAMD64EndToEnd(nil)
    TestAMD64Encoder(nil)
  }
}

Change-Id: I4f1d37b5c2c966da3f2127705ccac9bff0038183
Reviewed-on: https://go-review.googlesource.com/63490
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: fix lexical block of captured variables
Alessandro Arzilli [Fri, 18 Aug 2017 08:22:19 +0000 (10:22 +0200)]
cmd/compile: fix lexical block of captured variables

Variables captured by a closure were always assigned to the root scope
in their declaration function. Using decl.Name.Defn.Pos will result in
the correct scope for both the declaration function and the capturing
function.

Fixes #21515

Change-Id: I3960aface3c4fc97e15b36191a74a7bed5b5ebc1
Reviewed-on: https://go-review.googlesource.com/56830
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoruntime: return deltimer early if timer.timersBucket is unset
Emmanuel Odeke [Fri, 15 Sep 2017 01:12:26 +0000 (19:12 -0600)]
runtime: return deltimer early if timer.timersBucket is unset

Return early from deltimer, with false as the result,
to indicate that we couldn't delete the timer since its
timersBucket was nil(not set) in the first place.

That happens in such a case where a user created
the timer from a Ticker with:

  t := time.Ticker{C: c}

The above usage skips the entire setup of assigning
the appropriate underlying runtimeTimer and timersBucket,
steps that are done for us by time.NewTicker.

CL 34784 introduced this bug with an optimization, by changing
stopTimer to retrieve the timersBucket from the timer itself
(which is unset with the mentioned usage pattern above),
whereas the old  behavior relied on indexing
by goroutine ID into the global slice of runtime
timers, to retrieve the appropriate timersBucket.

Fixes #21874

Change-Id: Ie9ccc6bdee685414b2430dc4aa74ef618cea2b33
Reviewed-on: https://go-review.googlesource.com/63970
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/internal/objabi: remove unused flag funcs
Matthew Dempsky [Fri, 15 Sep 2017 19:04:34 +0000 (12:04 -0700)]
cmd/internal/objabi: remove unused flag funcs

Change-Id: I728c5606882ece949d58e86f9558fc16ae4ffd85
Reviewed-on: https://go-review.googlesource.com/64052
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/link: replace unrolled Cput loops with Cwrite/Cwritestring
Matthew Dempsky [Fri, 15 Sep 2017 18:38:24 +0000 (11:38 -0700)]
cmd/link: replace unrolled Cput loops with Cwrite/Cwritestring

Passes toolstash-check -all.

Change-Id: I1c85a2c0390517f4e9cdbddddbf3c353edca65b3
Reviewed-on: https://go-review.googlesource.com/64051
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: change lockedg/lockedm to guintptr/muintptr
Ian Lance Taylor [Wed, 13 Sep 2017 17:14:02 +0000 (10:14 -0700)]
runtime: change lockedg/lockedm to guintptr/muintptr

This change has no real effect in itself. This is to prepare for a
followup change that will call lockOSThread during a cgo callback when
there is no p assigned, and therefore when lockOSThread can not use a
write barrier.

Change-Id: Ia122d41acf54191864bcb68f393f2ed3b2f87abc
Reviewed-on: https://go-review.googlesource.com/63630
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/compile: replace GOROOT in //line directives
David Crawshaw [Wed, 13 Sep 2017 23:04:25 +0000 (19:04 -0400)]
cmd/compile: replace GOROOT in //line directives

The compiler replaces any path of the form /path/to/goroot/src/net/port.go
with GOROOT/src/net/port.go so that the same object file is
produced if the GOROOT is moved. It was skipping this transformation
for any absolute path into the GOROOT that came from //line directives,
such as those generated by cmd/cgo.

Fixes #21373
Fixes #21720
Fixes #21825

Change-Id: I2784c701b4391cfb92e23efbcb091a84957d61dd
Reviewed-on: https://go-review.googlesource.com/63693
Run-TryBot: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: fix typo in floating point rule
Todd Neal [Fri, 15 Sep 2017 00:47:18 +0000 (19:47 -0500)]
cmd/compile: fix typo in floating point rule

Change-Id: Idfb64fcb26f48d5b70bab872f9a3d96a036be681
Reviewed-on: https://go-review.googlesource.com/63950
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoall: fix article typos
Kunpei Sakai [Fri, 15 Sep 2017 01:24:47 +0000 (10:24 +0900)]
all: fix article typos

a -> an

Change-Id: I7362bdc199e83073a712be657f5d9ba16df3077e
Reviewed-on: https://go-review.googlesource.com/63850
Reviewed-by: Rob Pike <r@golang.org>
7 years agocmd/go: correctly report that -msan needs CGO_ENABLED=1
Emmanuel Odeke [Thu, 14 Sep 2017 23:16:58 +0000 (17:16 -0600)]
cmd/go: correctly report that -msan needs CGO_ENABLED=1

Previously, if CGO_ENABLED=0 was set when building
with -msan, the error message printed was:

  -race requires cgo; enable cgo by setting CGO_ENABLED=1

yet the instrumentation flag passed in was -msan. This CL
fixes that message to correctly report that -msan needed
CGO_ENABLED=1, and likewise if -race, report -race needed it.

Fixes #21895

Change-Id: If423d520daae7847fb38cc97c3192ada5d960f9d
Reviewed-on: https://go-review.googlesource.com/63930
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
7 years agocmd/compile: improve floating point constant propagation
Michael Munday [Thu, 14 Sep 2017 10:04:37 +0000 (11:04 +0100)]
cmd/compile: improve floating point constant propagation

Add generic rules to propagate floating point constants through
comparisons and integer conversions. These new rules seldom trigger
in the standard library so there is no performance change, however
I think it is worth adding them anyway for completeness.

Change-Id: I9db5222746508a2996f1cafb72f4e0cf2541de07
Reviewed-on: https://go-review.googlesource.com/63795
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile,math: improve int<->float conversions on ppc64x
Lynn Boger [Tue, 12 Sep 2017 16:33:21 +0000 (12:33 -0400)]
cmd/compile,math: improve int<->float conversions on ppc64x

The functions Float64bits and Float64frombits perform
poorly on ppc64x because the int<->float conversions
often result in load and store sequences to handle the
type change. This patch adds more rules to recognize
those sequences and use register to register moves
and avoid unnecessary loads and stores where possible.

There were some existing rules to improve these conversions,
but this provides additional improvements. Included here:

- New instruction FCFIDS to improve on conversion to 32 bit
- Rename Xf2i64 and Xi2f64 as MTVSRD, MFVSRD, to match the asm
- Add rules to lower some of the load/store sequences for
- Added new go asm to ppc64.s testcase.
conversions

Improvements:

BenchmarkAbs-16                2.16          0.93          -56.94%
BenchmarkCopysign-16           2.66          1.18          -55.64%
BenchmarkRound-16              4.82          2.69          -44.19%
BenchmarkSignbit-16            1.71          1.14          -33.33%
BenchmarkFrexp-16              11.4          7.94          -30.35%
BenchmarkLogb-16               10.4          7.34          -29.42%
BenchmarkLdexp-16              15.7          11.2          -28.66%
BenchmarkIlogb-16              10.2          7.32          -28.24%
BenchmarkPowInt-16             69.6          55.9          -19.68%
BenchmarkModf-16               10.1          8.19          -18.91%
BenchmarkLog2-16               17.4          14.3          -17.82%
BenchmarkCbrt-16               45.0          37.3          -17.11%
BenchmarkAtanh-16              57.6          48.3          -16.15%
BenchmarkRemainder-16          76.6          65.4          -14.62%
BenchmarkGamma-16              26.0          22.5          -13.46%
BenchmarkPowFrac-16            197           174           -11.68%
BenchmarkMod-16                112           99.8          -10.89%
BenchmarkAsinh-16              59.9          53.7          -10.35%
BenchmarkAcosh-16              44.8          40.3          -10.04%

Updates #21390

Change-Id: I56cc991fc2e55249d69518d4e1ba76cc23904e35
Reviewed-on: https://go-review.googlesource.com/63290
Reviewed-by: Michael Munday <mike.munday@ibm.com>
7 years agocmd/compile: expand inlining test to multiple pkgs
Daniel Martí [Wed, 13 Sep 2017 19:03:20 +0000 (21:03 +0200)]
cmd/compile: expand inlining test to multiple pkgs

Rework the test to work with any number of std packages. This was done
to include a few funcs from unicode/utf8. Adding more will be much
simpler too.

While at it, add more runtime funcs by searching for "inlined" or
"inlining" in the git log of its directory. These are: addb, subtractb,
fastrand and noescape.

Updates #21851.

Change-Id: I4fb2bd8aa6a5054218f9b36cb19d897ac533710e
Reviewed-on: https://go-review.googlesource.com/63611
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: limit the number of simultaneously opened files to avoid EMFILE/ENFILE...
zhongtao.chen [Tue, 22 Aug 2017 02:33:10 +0000 (10:33 +0800)]
cmd/compile: limit the number of simultaneously opened files to avoid EMFILE/ENFILE errors

If the Go packages with enough source files,it will cause EMFILE/ENFILE error,
Fix this by limiting the number of simultaneously opened files.

Fixes #21621

Change-Id: I8555d79242d2f90771e37e073b7540fc7194a64a
Reviewed-on: https://go-review.googlesource.com/57751
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/errors: don't pass -C to compiler
Ian Lance Taylor [Thu, 14 Sep 2017 03:41:45 +0000 (20:41 -0700)]
misc/cgo/errors: don't pass -C to compiler

It's not needed, and the current expectation is that it will go away
in the future.

Change-Id: I5f46800e748d9ffa484bda6d1738290c8e00ac2b
Reviewed-on: https://go-review.googlesource.com/63751
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomath/rand: fix comment for Shuffle
Burak Guven [Thu, 14 Sep 2017 02:54:40 +0000 (19:54 -0700)]
math/rand: fix comment for Shuffle

Shuffle panics if n < 0, not n <= 0. The comment for the (*Rand).Shuffle
function is already accurate.

Change-Id: I073049310bca9632e50e9ca3ff79eec402122793
Reviewed-on: https://go-review.googlesource.com/63750
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/errors: test that the Go rune type is not identical to C.int
Bryan C. Mills [Wed, 13 Sep 2017 21:21:07 +0000 (17:21 -0400)]
misc/cgo/errors: test that the Go rune type is not identical to C.int

rune has a well-defined size, but C.int is implementation-specified.
Using one as the other should require an explicit conversion.

updates #13467

Change-Id: I53ab2478427dca790efdcc197f6b8d9fbfbd1847
Reviewed-on: https://go-review.googlesource.com/63730
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/errors: fix erroneous regexp detection
Bryan C. Mills [Wed, 13 Sep 2017 19:14:05 +0000 (15:14 -0400)]
misc/cgo/errors: fix erroneous regexp detection

I had passed 1 instead of 2 to the SplitAfterN call in
errorstest.check, so all of the cases were erroneously falling through
to the non-regexp case (and passing even if the actual error didn't
match).

Now, we use bytes.HasSuffix to check for the non-regexp case, so we
will not incorrectly match a regexp comment to the non-regexp case.

updates #13467

Change-Id: Ia6be928a495425f2b7bae5001bd01346e115dcfa
Reviewed-on: https://go-review.googlesource.com/63692
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoos: avoid crashing with a thundering herd in TestPipeThreads
Ian Lance Taylor [Wed, 13 Sep 2017 18:16:40 +0000 (11:16 -0700)]
os: avoid crashing with a thundering herd in TestPipeThreads

Fixes #21559

Change-Id: I3393c4bee4c84fe0724a9c9aeb1a809b1a92eea6
Reviewed-on: https://go-review.googlesource.com/63650
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Joe Tsai <joetsai@google.com>
7 years agocmd/compile: simplify exporting ONAME nodes
Matthew Dempsky [Tue, 12 Sep 2017 20:36:44 +0000 (13:36 -0700)]
cmd/compile: simplify exporting ONAME nodes

These two special cases are unnecessary:

1) "~b%d" references only appear during walk, to handle "return"
statements implicitly assigning to blank result parameters. Even if
they could appear, the "inlined and customized version" accidentally
diverged from p.sym in golang.org/cl/33911.

2) The Vargen case is already identical to the default case, and it
never overlaps with the remaining "T.method" case.

Passes toolstash-check.

Change-Id: I03f7e5b75b707b43afc8ed6eb90f43ba93ed17ae
Reviewed-on: https://go-review.googlesource.com/63272
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agoarchive/tar: populate Devmajor and Devminor in FileInfoHeader on *BSD
Tobias Klauser [Wed, 13 Sep 2017 07:11:49 +0000 (09:11 +0200)]
archive/tar: populate Devmajor and Devminor in FileInfoHeader on *BSD

Extract device major/minor number on all the BSDs and set Devmajor and
Devminor in FileInfoHeader. Code based on the corresponding Major/Minor
implementations in golang.org/x/sys/unix.

Change-Id: Ieffa7ce0cdbe6481950de666b2f5f88407a32382
Reviewed-on: https://go-review.googlesource.com/63470
Reviewed-by: Joe Tsai <joetsai@google.com>
7 years agocmd/compile: simplify exporting OTYPE nodes
Matthew Dempsky [Tue, 12 Sep 2017 20:42:10 +0000 (13:42 -0700)]
cmd/compile: simplify exporting OTYPE nodes

We only export packages that typechecked successfully, and OTYPE nodes
will always have their Type field set.

Changes the package export format, but only in the compiler-specific
section. No version bump necessary.

Change-Id: I722f5827e73948fceb0432bc8b3b22471fea8f61
Reviewed-on: https://go-review.googlesource.com/63273
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agocmd/compile: add TestIntendedInlining from runtime
Daniel Martí [Wed, 13 Sep 2017 13:04:16 +0000 (15:04 +0200)]
cmd/compile: add TestIntendedInlining from runtime

Move it from the runtime package, as we will soon add more packages and
functions for it to check.

The test used the testEnv func, which cleaned certain environment
variables from a command, so it was moved to internal/testenv under a
more descriptive (and less ambiguous) name. Add a simple godoc to it
too.

For #21851.

Change-Id: I6f39c1f23b45377718355fafe66ffd87047d8ab6
Reviewed-on: https://go-review.googlesource.com/63550
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agomisc/cgo/errors: port test.bash to Go
Bryan C. Mills [Tue, 12 Sep 2017 22:28:02 +0000 (18:28 -0400)]
misc/cgo/errors: port test.bash to Go

This makes the test easier to run in isolation and easier to change,
and simplifies the code to run the tests in parallel.

updates #13467

Change-Id: I5622b5cc98276970347da18e95d071dbca3c5cc1
Reviewed-on: https://go-review.googlesource.com/63276
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoencoding/xml: add decode wrapper
Sam Whited [Wed, 29 Mar 2017 18:19:29 +0000 (13:19 -0500)]
encoding/xml: add decode wrapper

Fixes #19480

Change-Id: I5a621507279d5bb1f3991b7a412d9a63039d464b
Reviewed-on: https://go-review.googlesource.com/38791
Run-TryBot: Sam Whited <sam@samwhited.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocontainer/heap: call t.Helper() in verify()
Tyler Bui-Palsulich [Wed, 13 Sep 2017 15:09:21 +0000 (08:09 -0700)]
container/heap: call t.Helper() in verify()

I modified verify() to fail every time to test the change. Before adding
t.Helper() (line 37 is in verify()):
/.../go/src/container/heap/heap_test.go:37: forced failure
FAIL

Afer adding t.Helper() (line 67 is where verify() is called):
/.../go/src/container/heap/heap_test.go:67: forced failure
FAIL

Fixes #21863

Change-Id: I46f0c8ec413cc664358c568fc53e48bb4a1d03d0
Reviewed-on: https://go-review.googlesource.com/63570
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotext/template: check ignored error in a test
Daniel Martí [Mon, 11 Sep 2017 19:46:11 +0000 (21:46 +0200)]
text/template: check ignored error in a test

Found with staticcheck. Not terribly important since the test would
likely fail anyway, but at least it will fail with a better explanation
now.

Change-Id: Ic3f9a94a2152404b7873cc8cd47b6db79d78c2e6
Reviewed-on: https://go-review.googlesource.com/62990
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc: mention s390x in environment section of install-source.html
Michael Munday [Wed, 23 Aug 2017 11:45:27 +0000 (12:45 +0100)]
doc: mention s390x in environment section of install-source.html

Change-Id: I521ba94356165efc3252ca4b3d364a2bbabd4aab
Reviewed-on: https://go-review.googlesource.com/58010
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: refactor hmap.extra.overflow array into two separate fields
Martin Möhrmann [Sun, 10 Sep 2017 10:55:16 +0000 (12:55 +0200)]
runtime: refactor hmap.extra.overflow array into two separate fields

This makes it easier to deduce from the field names which overflow
field corresponds to h.buckets and which to h.oldbuckets by aligning
the naming with the buckets fields in hmap.

Change-Id: I8d6a729229a190db0212bac012ead1a3c13cf5d0
Reviewed-on: https://go-review.googlesource.com/62411
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoruntime: move evacuateX evacuateY relation check from makemap to evacuate
Martin Möhrmann [Fri, 8 Sep 2017 21:05:19 +0000 (23:05 +0200)]
runtime: move evacuateX evacuateY relation check from makemap to evacuate

Move the check near the code in evacuate that relies on
the relation evacuateX+1 == evacuateY.

If the relation is fullfilled the check is known to be true
at compile time and removed by the compiler.

Change-Id: I711b75e09047bf347819ccaeec41d244a5883867
Reviewed-on: https://go-review.googlesource.com/62410
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/vet: fix go vet on parentheses of assembly function flag
Wei Xiao [Mon, 11 Sep 2017 07:59:30 +0000 (07:59 +0000)]
cmd/vet: fix go vet on parentheses of assembly function flag

Current implementation doesn't recognize parentheses that may appear in flags
of assembly function as shown below:

TEXT ·makeFuncStub(SB),(NOSPLIT|WRAPPER),$24

It results in vet reporting false positives and a lot of whitelists are added
for suppressing the false alarms.

This CL fixes the issue and eliminates the redundant whitelists.

Change-Id: Idbc1b42965b31cea8ee7c23d1a6f62feb68e844c
Reviewed-on: https://go-review.googlesource.com/62850
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
7 years agomisc/cgo/test: set the traceback level instead of failing the test
Bryan C. Mills [Tue, 12 Sep 2017 20:45:26 +0000 (16:45 -0400)]
misc/cgo/test: set the traceback level instead of failing the test

Previously, test7978 failed if the user did not invoke it with
GOTRACEBACK=2 already set in their environment. Environment-sensitive
test are awkward, and in this case there is a very simple workaround:
set the traceback level to the necessary value explicitly.

Change-Id: I7d576f24138aa8a41392148eae11bbeaef558573
Reviewed-on: https://go-review.googlesource.com/63275
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agounicode: speed-up is16/is32
Ilya Tocar [Tue, 12 Sep 2017 20:31:09 +0000 (15:31 -0500)]
unicode: speed-up is16/is32

Avoid division in common case. There are 5438 ranges in unicode/tables.go
4110 of them have stride 1.
Stride 1 case got significantly faster. Other stride is a bit slower.
Measured by

import (
"testing"
"unicode"
)

func BenchmarkDiv1(b *testing.B) {
rtb := &unicode.RangeTable{
R16: []unicode.Range16{
{0xa800, 0xdfff, 1}, // or 3
},
}
for i := 0; i < b.N; i++ {
unicode.Is(rtb, rune(0xc700))
}
}

Div1-6  15.6ns ± 1%   9.9ns ± 1%  -36.54%  (p=0.000 n=10+10)
Div3-6  15.5ns ± 1%  16.1ns ± 1%   +3.67%  (p=0.000 n=10+10)

Helps a bit with xml parsing from issue #21823

XMLsax-6   30.9s ± 0%   29.6s ± 0%  -4.15%  (p=0.000 n=10+9)

Change-Id: Ibac1a91d7b9474d0c134b0add83e56caa62daa20
Reviewed-on: https://go-review.googlesource.com/63390
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agounicode/utf8: make FullRune inlinable
Ilya Tocar [Tue, 12 Sep 2017 19:23:09 +0000 (14:23 -0500)]
unicode/utf8: make FullRune inlinable

This has same readability and allows to inline FullRune for massive
performance gain:

FullASCIIRune-6                      4.36ns ± 0%  1.25ns ± 0%  -71.33%  (p=0.000 n=8+10)
FullJapaneseRune-6                   4.70ns ± 0%  1.42ns ± 1%  -69.68%  (p=0.000 n=9+10)

Change-Id: I95edd6292417a28aac244e40afb713596a087d93
Reviewed-on: https://go-review.googlesource.com/63332
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: remove redundant switch label
Daniel Martí [Mon, 11 Sep 2017 19:10:43 +0000 (21:10 +0200)]
cmd/compile: remove redundant switch label

This label was added automatically by grind to remove gotos. As of
today, it's completely useless, as none of its uses need a label to
begin with.

While at it, remove all the redundant breaks too. Leave those that are
the single statement in a case clause body, as that's the style used
throughout std and cmd to clarify when cases are empty.

Change-Id: I3e20068b66b759614e903beab1cc9b2709b31063
Reviewed-on: https://go-review.googlesource.com/62950
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoencoding/json: update documentation for MarshalIndent
tbunyk [Tue, 5 Sep 2017 16:01:16 +0000 (19:01 +0300)]
encoding/json: update documentation for MarshalIndent

Make arguments semantics clear without the need to look for
json.Indent documentation.

Change-Id: If9adfe9f477a30d426ae83790b0f2578c0a809b7
Reviewed-on: https://go-review.googlesource.com/61670
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocrypto: simplify amd64 asm for sha{1,256,512} a bit
Ilya Tocar [Wed, 6 Sep 2017 19:33:36 +0000 (14:33 -0500)]
crypto: simplify amd64 asm for sha{1,256,512} a bit

Use constants directly, instead of loading address to e. g. AX
and using (AX). Shouldn't affect performance, but makes code a bit
nicer.

Change-Id: Ifa138e54d3d2b2f4ad71e4ef4b9368ea79eb30f4
Reviewed-on: https://go-review.googlesource.com/62010
Reviewed-by: Adam Langley <agl@golang.org>
7 years agodoc: unify colons and spacing in diagnostics guide
Tobias Klauser [Tue, 12 Sep 2017 12:20:17 +0000 (14:20 +0200)]
doc: unify colons and spacing in diagnostics guide

Unify colons (outside of <strong></strong>) and add a missing space in
the list of groups of diagnostics solutions.

Change-Id: Icbcd94427d4905dd88c4ea82aaa5dbf064c00990
Reviewed-on: https://go-review.googlesource.com/63111
Reviewed-by: JBD <jbd@google.com>
7 years agoruntime: optimize siftupTimer and siftdownTimer a bit
Aliaksandr Valialkin [Tue, 12 Sep 2017 11:04:34 +0000 (14:04 +0300)]
runtime: optimize siftupTimer and siftdownTimer a bit

Write the moving timer only once, since it is overwritten
by swapped timers on all the iterations except the last one.

Additionally, explicitly pass timers heap into siftupTimer
and siftdownTimer in order to make the code more clear.

Relevant benchmark results on linux/amd64:

Stop                      700µs ± 7%   608µs ± 1%  -13.13%  (p=0.000 n=10+10)
Stop-2                    440µs ± 4%   376µs ± 4%  -14.48%  (p=0.000 n=10+10)
Stop-4                    339µs ± 2%   330µs ± 3%   -2.66%  (p=0.015 n=10+10)
SimultaneousAfterFunc     702µs ± 9%   709µs ± 1%     ~     (p=0.436 n=9+9)
SimultaneousAfterFunc-2   573µs ± 2%   546µs ± 2%   -4.71%  (p=0.000 n=10+10)
SimultaneousAfterFunc-4   387µs ± 1%   368µs ± 1%   -4.89%  (p=0.000 n=8+10)
StartStop                 268µs ± 0%   270µs ± 0%   +0.91%  (p=0.000 n=9+9)
StartStop-2               155µs ± 6%   145µs ± 6%   -6.70%  (p=0.000 n=10+10)
StartStop-4               125µs ± 1%   124µs ± 1%     ~     (p=0.065 n=10+9)

Change-Id: I3685835b5e3e82844e2e5e73ee03a1e22100bf0e
Reviewed-on: https://go-review.googlesource.com/63110
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: in cpuProfile.addExtra, set p.lostExtra to 0 after flush
Michael Matloob [Tue, 12 Sep 2017 16:22:22 +0000 (12:22 -0400)]
runtime: in cpuProfile.addExtra, set p.lostExtra to 0 after flush

After the number of lost extra events are written to the the cpuprof log,
the number of lost extra events should be set to zero, or else, the next
time time addExtra is logged, lostExtra will be overcounted. This change
resets lostExtra after its value is written to the log.

Fixes #21836

Change-Id: I8a6ac9c61e579e7a5ca7bdb0f3463f8ae8b9f863
Reviewed-on: https://go-review.googlesource.com/63270
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: improve timers scalability on multi-CPU systems
Aliaksandr Valialkin [Fri, 6 Jan 2017 13:44:17 +0000 (15:44 +0200)]
runtime: improve timers scalability on multi-CPU systems

Use per-P timers, so each P may work with its own timers.

This CL improves performance on multi-CPU systems
in the following cases:

- When serving high number of concurrent connections
  with read/write deadlines set (for instance, highly loaded
  net/http server).

- When using high number of concurrent timers. These timers
  may be implicitly created via context.WithDeadline
  or context.WithTimeout.

Production servers should usually set timeout on connections
and external requests in order to prevent from resource leakage.
See https://blog.cloudflare.com/the-complete-guide-to-golang-net-http-timeouts/

Below are relevant benchmark results for various GOMAXPROCS values
on linux/amd64:

context package:

name                                     old time/op  new time/op  delta
WithTimeout/concurrency=40      4.92µs ± 0%  5.17µs ± 1%  +5.07%  (p=0.000 n=9+9)
WithTimeout/concurrency=4000    6.03µs ± 1%  6.49µs ± 0%  +7.63%  (p=0.000 n=8+10)
WithTimeout/concurrency=400000  8.58µs ± 7%  9.02µs ± 4%  +5.02%  (p=0.019 n=10+10)

name                                     old time/op  new time/op  delta
WithTimeout/concurrency=40-2      3.70µs ± 1%  2.78µs ± 4%  -24.90%  (p=0.000 n=8+9)
WithTimeout/concurrency=4000-2    4.49µs ± 4%  3.67µs ± 5%  -18.26%  (p=0.000 n=10+10)
WithTimeout/concurrency=400000-2  6.16µs ±10%  5.15µs ±13%  -16.30%  (p=0.000 n=10+10)

name                                     old time/op  new time/op  delta
WithTimeout/concurrency=40-4      3.58µs ± 1%  2.64µs ± 2%  -26.13%  (p=0.000 n=9+10)
WithTimeout/concurrency=4000-4    4.17µs ± 0%  3.32µs ± 1%  -20.36%  (p=0.000 n=10+10)
WithTimeout/concurrency=400000-4  5.57µs ± 9%  4.83µs ±10%  -13.27%  (p=0.001 n=10+10)

time package:

name                     old time/op  new time/op  delta
AfterFunc                6.15ms ± 3%  6.07ms ± 2%     ~     (p=0.133 n=10+9)
AfterFunc-2              3.43ms ± 1%  3.56ms ± 1%   +3.91%  (p=0.000 n=10+9)
AfterFunc-4              5.04ms ± 2%  2.36ms ± 0%  -53.20%  (p=0.000 n=10+9)
After                    6.54ms ± 2%  6.49ms ± 3%     ~     (p=0.393 n=10+10)
After-2                  3.68ms ± 1%  3.87ms ± 0%   +5.14%  (p=0.000 n=9+9)
After-4                  6.66ms ± 1%  2.87ms ± 1%  -56.89%  (p=0.000 n=10+10)
Stop                      698µs ± 2%   689µs ± 1%   -1.26%  (p=0.011 n=10+10)
Stop-2                    729µs ± 2%   434µs ± 3%  -40.49%  (p=0.000 n=10+10)
Stop-4                    837µs ± 3%   333µs ± 2%  -60.20%  (p=0.000 n=10+10)
SimultaneousAfterFunc     694µs ± 1%   692µs ± 7%     ~     (p=0.481 n=10+10)
SimultaneousAfterFunc-2   714µs ± 3%   569µs ± 2%  -20.33%  (p=0.000 n=10+10)
SimultaneousAfterFunc-4   782µs ± 2%   386µs ± 2%  -50.67%  (p=0.000 n=10+10)
StartStop                 267µs ± 3%   274µs ± 0%   +2.64%  (p=0.000 n=8+9)
StartStop-2               238µs ± 2%   140µs ± 3%  -40.95%  (p=0.000 n=10+8)
StartStop-4               320µs ± 1%   125µs ± 1%  -61.02%  (p=0.000 n=9+9)
Reset                    75.0µs ± 1%  77.5µs ± 2%   +3.38%  (p=0.000 n=10+10)
Reset-2                   150µs ± 2%    40µs ± 5%  -73.09%  (p=0.000 n=10+9)
Reset-4                   226µs ± 1%    33µs ± 1%  -85.42%  (p=0.000 n=10+10)
Sleep                     857µs ± 6%   878µs ± 9%     ~     (p=0.079 n=10+9)
Sleep-2                   617µs ± 4%   585µs ± 2%   -5.21%  (p=0.000 n=10+10)
Sleep-4                   689µs ± 3%   465µs ± 4%  -32.53%  (p=0.000 n=10+10)
Ticker                   55.9ms ± 2%  55.9ms ± 2%     ~     (p=0.971 n=10+10)
Ticker-2                 28.7ms ± 2%  28.1ms ± 1%   -2.06%  (p=0.000 n=10+10)
Ticker-4                 14.6ms ± 0%  13.6ms ± 1%   -6.80%  (p=0.000 n=9+10)

Fixes #15133

Change-Id: I6f4b09d2db8c5bec93146db6501b44dbfe5c0ac4
Reviewed-on: https://go-review.googlesource.com/34784
Reviewed-by: Austin Clements <austin@google.com>
7 years agogo/*: use sort.Slice to simplify some code
Daniel Martí [Tue, 12 Sep 2017 14:32:33 +0000 (16:32 +0200)]
go/*: use sort.Slice to simplify some code

Skip the ones that have multiple uses for now. Also had to rename the
importComment variable as it shadowed the top-level func by the same
name.

Change-Id: I796285aa7b4fdf2c39e652666390427d37b063ee
Reviewed-on: https://go-review.googlesource.com/63150
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoall: remove strings.Contains check around Replace
Daniel Martí [Mon, 11 Sep 2017 19:51:38 +0000 (21:51 +0200)]
all: remove strings.Contains check around Replace

It doesn't change the outcome. It might have been useful at some point
to avoid Replace from doing work or allocating. However, nowadays the
func returns early without doing any work if Count returns 0.

Change-Id: Id69dc74042a6e39672b405016484db8b50f43d58
Reviewed-on: https://go-review.googlesource.com/62991
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
7 years agocmd/compile: improve single blank variable handling in walkrange
Martin Möhrmann [Sun, 3 Sep 2017 06:20:57 +0000 (08:20 +0200)]
cmd/compile: improve single blank variable handling in walkrange

Refactor walkrange to treat "for _ = range a" as "for range a".

This avoids generating some later discarded nodes in the compiler.

Passes toolstash -cmp.

Change-Id: Ifb2e1ca3b8519cbb67e8ad5aad514af9d18f1ec4
Reviewed-on: https://go-review.googlesource.com/61017
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agonet/http: allow reuse of http.Request objects
Luan Santos [Mon, 11 Sep 2017 15:37:50 +0000 (08:37 -0700)]
net/http: allow reuse of http.Request objects

Calling response.Body.Close() early would generarate a race before this.
Since closing would return early before the main code path had a chance
to reset the request canceler. Having a non-nil request canceler at the
start of the next request would cause a "request canceled" error.

Here we simply wait for the eofc channel to be closed before returning
from earlyCloseFn, ensuring that the caller won't be re-using that
Request object before we have a chance to reset the request canceler to
nil.

Fixes #21838

Change-Id: I641815526c6ac63d1816c9b6ad49d73715f7a5cb
Reviewed-on: https://go-review.googlesource.com/62891
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Tom Bergan <tombergan@google.com>
7 years agoexpvar: make (*Map).Init clear existing keys
Bryan C. Mills [Fri, 8 Sep 2017 18:36:43 +0000 (14:36 -0400)]
expvar: make (*Map).Init clear existing keys

fixes #21619

Change-Id: I5bb513dfc8cac875b06a262eec40b5863ae23a4c
Reviewed-on: https://go-review.googlesource.com/62370
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/link: enable -buildmode=plugin for ppc64le
Lynn Boger [Tue, 15 Aug 2017 18:34:53 +0000 (14:34 -0400)]
cmd/link: enable -buildmode=plugin for ppc64le

This enables support for the buildmode plugin on
ppc64le.

Fixes #20756

Change-Id: I83241ff63f9b5c366fe0496cf46a3f67d75d08ac
Reviewed-on: https://go-review.googlesource.com/55850
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agocrypto/sha1: speed up sha1 for very small blocks
Ilya Tocar [Wed, 9 Aug 2017 18:21:20 +0000 (13:21 -0500)]
crypto/sha1: speed up sha1 for very small blocks

For very small blocks significant time is spent in checkSum function,
adding necessary padding. Instead of writing it byte by byte, copy
encoding/binary PutUint functions, which are compiled into single mov.

name            old time/op    new time/op    delta
Hash8Bytes-6       344ns ± 0%     310ns ± 0%   -9.78%  (p=0.000 n=10+9)
Hash320Bytes-6    1.28µs ± 0%    1.25µs ± 0%   -2.58%  (p=0.000 n=10+10)
Hash1K-6          2.51µs ± 0%    2.47µs ± 0%   -1.67%  (p=0.000 n=10+10)
Hash8K-6          15.8µs ± 0%    15.7µs ± 1%   -0.21%  (p=0.023 n=10+10)

name            old speed      new speed      delta
Hash8Bytes-6    23.2MB/s ± 0%  25.7MB/s ± 0%  +10.77%  (p=0.000 n=10+9)
Hash320Bytes-6   249MB/s ± 0%   256MB/s ± 0%   +2.65%  (p=0.000 n=10+10)
Hash1K-6         408MB/s ± 0%   414MB/s ± 0%   +1.70%  (p=0.000 n=10+10)

Change-Id: I3975ee929465c7dd137d0ca757ad3792a004e1a3
Reviewed-on: https://go-review.googlesource.com/54391
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Filippo Valsorda <hi@filippo.io>
7 years agocmd/compile: simplify a few early var declarations
Daniel Martí [Mon, 11 Sep 2017 19:23:44 +0000 (21:23 +0200)]
cmd/compile: simplify a few early var declarations

These were likely written in C or added by an automated tool. Either
way, they're unnecessary now. Clean up the code.

Change-Id: I56de2c7bb60ebab8c500803a8b6586bdf4bf75c7
Reviewed-on: https://go-review.googlesource.com/62951
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile, math/bits: add rotate rules to PPC64.rules
Lynn Boger [Tue, 29 Aug 2017 15:49:08 +0000 (11:49 -0400)]
cmd/compile, math/bits: add rotate rules to PPC64.rules

This adds rules to match the code in math/bits RotateLeft,
RotateLeft32, and RotateLef64 to allow them to be inlined.

The rules are complicated because the code in these function
use different types, and the non-const version of these
shifts generate Mask and Carry instructions that become
subexpressions during the match process.

Also adds a testcase to asm_test.go.

Improvement in math/bits:

BenchmarkRotateLeft-16       1.57     1.32      -15.92%
BenchmarkRotateLeft32-16     1.60     1.37      -14.37%
BenchmarkRotateLeft64-16     1.57     1.32      -15.92%

Updates #21390

Change-Id: Ib6f17669ecc9cab54f18d690be27e2225ca654a4
Reviewed-on: https://go-review.googlesource.com/59932
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/go: fix regexp
Joe Tsai [Mon, 11 Sep 2017 17:39:40 +0000 (10:39 -0700)]
cmd/go: fix regexp

The regular expression "A-za-z" is most likely a typo and
the intent seems to be "A-Za-z" instead.

Using "A-z" matches certain characters like: [\]^_`

Updates #10010

Change-Id: If2d064c56ef613f2e46285d8d4e5998e83aed43a
Reviewed-on: https://go-review.googlesource.com/62910
Reviewed-by: Matt Layher <mdlayher@gmail.com>
Run-TryBot: Matt Layher <mdlayher@gmail.com>

7 years agocrypto/elliptic: fix incomplete addition used in CombinedMult.
Adam Langley [Thu, 4 May 2017 01:20:12 +0000 (18:20 -0700)]
crypto/elliptic: fix incomplete addition used in CombinedMult.

The optimised P-256 includes a CombinedMult function, which doesn't do
dual-scalar multiplication, but does avoid an affine conversion for
ECDSA verification.

However, it currently uses an assembly point addition function that
doesn't handle exceptional cases.

Fixes #20215.

Change-Id: I4ba2ca1a546d883364a9bb6bf0bdbc7f7b44c94a
Reviewed-on: https://go-review.googlesource.com/42611
Run-TryBot: Adam Langley <agl@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
7 years agocmd/vet: print: permit '#' flag in %e, %f, and %g
Alan Donovan [Fri, 8 Sep 2017 18:45:59 +0000 (14:45 -0400)]
cmd/vet: print: permit '#' flag in %e, %f, and %g

+ Test

Change-Id: I7b42ff70f26a58e1cf58cbbc53d02a65623456ae
Reviewed-on: https://go-review.googlesource.com/62371
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: request r12 for indirect calls on ppc64le
Lynn Boger [Mon, 11 Sep 2017 14:39:24 +0000 (10:39 -0400)]
cmd/compile: request r12 for indirect calls on ppc64le

On ppc64le, functions compiled with -shared expect r12 to
hold the function's address for indirect calls. Previously
this was enforced by generating a move instruction if the
address wasn't already in r12. This change avoids that extra
move by requesting r12 in the CALL ops that do indirect calls.

As a result of adding support for plugins on ppc64le, it was
discovered that there would be more cases where this extra
move was needed, so this seemed like a better solution.

Updates #20756

Change-Id: I6770885a46990f78c6d2902a715dcdaa822192a1
Reviewed-on: https://go-review.googlesource.com/62890
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocrypto/elliptic: temporarily disable s390x assembly
Michael Munday [Sat, 9 Sep 2017 23:22:27 +0000 (00:22 +0100)]
crypto/elliptic: temporarily disable s390x assembly

This disables the s390x assembly. It will be re-enabled when #20215
is resolved on s390x.

Change-Id: I789eca2dd478004956107359fae98ed012f04abb
Reviewed-on: https://go-review.googlesource.com/62292
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc: hide content blocked by GFW when served in CN
Andrew Bonventre [Tue, 29 Aug 2017 20:23:49 +0000 (16:23 -0400)]
doc: hide content blocked by GFW when served in CN

Change-Id: Ia3d0fa0517d96f0672d3c5ce7e7b7bb8bd0ce093
Reviewed-on: https://go-review.googlesource.com/60070
Reviewed-by: Chris Broadfoot <cbro@golang.org>
7 years agocmd/internal/obj/arm: support more ARM VFP instructions
Ben Shi [Sat, 9 Sep 2017 14:17:05 +0000 (14:17 +0000)]
cmd/internal/obj/arm: support more ARM VFP instructions

Add support of more ARM VFP instructions in the assembler.
They were introduced in ARM VFPv4.

"FMULAF/FMULAD   Fm, Fn, Fd": Fd = Fd + Fn*Fm
"FNMULAF/FNMULAD Fm, Fn, Fd": Fd = -(Fd + Fn*Fm)
"FMULSF/FMULSD   Fm, Fn, Fd": Fd = Fd - Fn*Fm
"FNMULSF/FNMULSD Fm, Fn, Fd": Fd = -(Fd - Fn*Fm)

The multiplication results are not rounded.

Change-Id: Id9cc52fd8e1b9a708103cd1e514c85a9e1cb3f47
Reviewed-on: https://go-review.googlesource.com/62550
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: optimize ARM code with NMULF/NMULD
Ben Shi [Sat, 2 Sep 2017 08:14:08 +0000 (08:14 +0000)]
cmd/compile: optimize ARM code with NMULF/NMULD

NMULF and NMULD are efficient FP instructions, and the go compiler can
use them to generate better code.

The benchmark tests of my patch did not show general change, but big
improvement in special cases.

1.A special test case improved 12.6%.
https://github.com/benshi001/ugo1/blob/master/fpmul_test.go
name                     old time/op    new time/op    delta
FPMul-4                     398µs ± 1%     348µs ± 1%  -12.64%  (p=0.000 n=40+40)

2. the compilecmp test showed little change.
name        old time/op       new time/op       delta
Template          2.30s ± 1%        2.31s ± 1%    ~     (p=0.754 n=17+19)
Unicode           1.31s ± 3%        1.32s ± 5%    ~     (p=0.265 n=20+20)
GoTypes           7.73s ± 2%        7.73s ± 1%    ~     (p=0.925 n=20+20)
Compiler          37.0s ± 1%        37.3s ± 2%  +0.79%  (p=0.002 n=19+20)
SSA               83.8s ± 4%        83.5s ± 2%    ~     (p=0.964 n=20+17)
Flate             1.43s ± 2%        1.44s ± 1%    ~     (p=0.602 n=20+20)
GoParser          1.82s ± 2%        1.81s ± 2%    ~     (p=0.141 n=19+20)
Reflect           5.08s ± 2%        5.08s ± 3%    ~     (p=0.835 n=20+19)
Tar               2.36s ± 1%        2.35s ± 1%    ~     (p=0.195 n=18+17)
XML               2.57s ± 2%        2.56s ± 1%    ~     (p=0.283 n=20+17)
[Geo mean]        4.74s             4.75s       +0.05%

name        old user-time/op  new user-time/op  delta
Template          2.75s ± 2%        2.75s ± 0%    ~     (p=0.620 n=20+15)
Unicode           1.59s ± 4%        1.60s ± 4%    ~     (p=0.479 n=20+19)
GoTypes           9.48s ± 1%        9.47s ± 1%    ~     (p=0.743 n=20+20)
Compiler          45.7s ± 1%        45.7s ± 1%    ~     (p=0.482 n=19+20)
SSA                109s ± 1%         109s ± 2%    ~     (p=0.800 n=18+20)
Flate             1.67s ± 3%        1.67s ± 3%    ~     (p=0.598 n=19+18)
GoParser          2.15s ± 4%        2.13s ± 3%    ~     (p=0.153 n=20+20)
Reflect           5.95s ± 2%        5.95s ± 2%    ~     (p=0.961 n=19+20)
Tar               2.93s ± 2%        2.92s ± 3%    ~     (p=0.242 n=20+19)
XML               3.02s ± 3%        3.04s ± 3%    ~     (p=0.233 n=19+18)
[Geo mean]        5.74s             5.74s       -0.04%

name        old text-bytes    new text-bytes    delta
HelloSize         588kB ± 0%        588kB ± 0%    ~     (all equal)

name        old data-bytes    new data-bytes    delta
HelloSize        5.46kB ± 0%       5.46kB ± 0%    ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize        72.9kB ± 0%       72.9kB ± 0%    ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.03MB ± 0%       1.03MB ± 0%    ~     (all equal)

3. The go1 benchmark showed little change in total.
name                     old time/op    new time/op    delta
BinaryTree17-4              41.8s ± 1%     41.8s ± 1%    ~     (p=0.388 n=40+39)
Fannkuch11-4                24.1s ± 1%     24.1s ± 1%    ~     (p=0.077 n=40+40)
FmtFprintfEmpty-4           834ns ± 1%     831ns ± 1%  -0.31%  (p=0.002 n=40+37)
FmtFprintfString-4         1.34µs ± 1%    1.34µs ± 0%    ~     (p=0.387 n=40+40)
FmtFprintfInt-4            1.44µs ± 1%    1.44µs ± 1%    ~     (p=0.421 n=40+40)
FmtFprintfIntInt-4         2.09µs ± 0%    2.09µs ± 1%    ~     (p=0.589 n=40+39)
FmtFprintfPrefixedInt-4    2.32µs ± 1%    2.33µs ± 1%  +0.15%  (p=0.001 n=40+40)
FmtFprintfFloat-4          4.51µs ± 0%    4.44µs ± 1%  -1.50%  (p=0.000 n=40+40)
FmtManyArgs-4              7.94µs ± 0%    7.97µs ± 0%  +0.36%  (p=0.001 n=32+40)
GobDecode-4                 104ms ± 1%     102ms ± 2%  -1.27%  (p=0.000 n=39+37)
GobEncode-4                90.5ms ± 1%    90.9ms ± 2%  +0.40%  (p=0.006 n=37+40)
Gzip-4                      4.10s ± 2%     4.08s ± 1%  -0.30%  (p=0.004 n=40+40)
Gunzip-4                    603ms ± 0%     602ms ± 1%    ~     (p=0.303 n=37+40)
HTTPClientServer-4          672µs ± 3%     658µs ± 2%  -2.08%  (p=0.000 n=39+37)
JSONEncode-4                238ms ± 1%     239ms ± 0%  +0.26%  (p=0.001 n=40+25)
JSONDecode-4                884ms ± 1%     885ms ± 1%  +0.16%  (p=0.012 n=40+40)
Mandelbrot200-4            49.3ms ± 0%    49.3ms ± 0%    ~     (p=0.588 n=40+38)
GoParse-4                  46.3ms ± 1%    46.4ms ± 2%    ~     (p=0.487 n=40+40)
RegexpMatchEasy0_32-4      1.28µs ± 1%    1.28µs ± 0%  +0.12%  (p=0.003 n=40+40)
RegexpMatchEasy0_1K-4      7.78µs ± 5%    7.78µs ± 4%    ~     (p=0.825 n=40+40)
RegexpMatchEasy1_32-4      1.29µs ± 1%    1.29µs ± 0%    ~     (p=0.659 n=40+40)
RegexpMatchEasy1_1K-4      10.3µs ± 3%    10.4µs ± 2%    ~     (p=0.266 n=40+40)
RegexpMatchMedium_32-4     2.05µs ± 1%    2.05µs ± 0%  -0.18%  (p=0.002 n=40+28)
RegexpMatchMedium_1K-4      533µs ± 1%     534µs ± 1%    ~     (p=0.397 n=37+40)
RegexpMatchHard_32-4       28.9µs ± 1%    28.9µs ± 1%  -0.22%  (p=0.002 n=40+40)
RegexpMatchHard_1K-4        868µs ± 1%     870µs ± 1%  +0.21%  (p=0.015 n=40+40)
Revcomp-4                  67.3ms ± 1%    67.2ms ± 2%    ~     (p=0.262 n=38+39)
Template-4                  1.07s ± 1%     1.07s ± 1%    ~     (p=0.276 n=40+40)
TimeParse-4                7.16µs ± 1%    7.16µs ± 1%    ~     (p=0.610 n=39+40)
TimeFormat-4               13.3µs ± 1%    13.3µs ± 1%    ~     (p=0.617 n=38+40)
[Geo mean]                  720µs          719µs       -0.13%

name                     old speed      new speed      delta
GobDecode-4              7.39MB/s ± 1%  7.49MB/s ± 2%  +1.25%  (p=0.000 n=39+38)
GobEncode-4              8.48MB/s ± 1%  8.45MB/s ± 2%  -0.40%  (p=0.005 n=37+40)
Gzip-4                   4.74MB/s ± 2%  4.75MB/s ± 1%  +0.30%  (p=0.018 n=40+40)
Gunzip-4                 32.2MB/s ± 0%  32.2MB/s ± 1%    ~     (p=0.272 n=36+40)
JSONEncode-4             8.15MB/s ± 1%  8.13MB/s ± 0%  -0.26%  (p=0.003 n=40+25)
JSONDecode-4             2.19MB/s ± 1%  2.19MB/s ± 1%    ~     (p=0.676 n=40+40)
GoParse-4                1.25MB/s ± 2%  1.25MB/s ± 2%    ~     (p=0.823 n=40+40)
RegexpMatchEasy0_32-4    25.1MB/s ± 1%  25.1MB/s ± 0%  -0.12%  (p=0.006 n=40+40)
RegexpMatchEasy0_1K-4     132MB/s ± 5%   132MB/s ± 5%    ~     (p=0.821 n=40+40)
RegexpMatchEasy1_32-4    24.7MB/s ± 1%  24.7MB/s ± 0%    ~     (p=0.630 n=40+40)
RegexpMatchEasy1_1K-4    99.1MB/s ± 3%  98.8MB/s ± 2%    ~     (p=0.268 n=40+40)
RegexpMatchMedium_32-4    487kB/s ± 2%   490kB/s ± 0%  +0.51%  (p=0.001 n=40+40)
RegexpMatchMedium_1K-4   1.92MB/s ± 1%  1.92MB/s ± 1%    ~     (p=0.208 n=39+40)
RegexpMatchHard_32-4     1.11MB/s ± 1%  1.11MB/s ± 0%  +0.36%  (p=0.000 n=40+33)
RegexpMatchHard_1K-4     1.18MB/s ± 1%  1.18MB/s ± 1%    ~     (p=0.207 n=40+37)
Revcomp-4                37.8MB/s ± 1%  37.8MB/s ± 2%    ~     (p=0.276 n=38+39)
Template-4               1.82MB/s ± 1%  1.81MB/s ± 1%    ~     (p=0.122 n=38+40)
[Geo mean]               6.81MB/s       6.81MB/s       +0.06%

fixes #19843

Change-Id: Ief3a0c2b15f59d40c7b40f2784eeb71196685b59
Reviewed-on: https://go-review.googlesource.com/61150
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomisc/cgo/testcshared: actually run test executable on android
Alex Brainman [Mon, 11 Sep 2017 07:19:24 +0000 (17:19 +1000)]
misc/cgo/testcshared: actually run test executable on android

CL 62593 broken TestExportedSymbols and TestUnexportedSymbols
because it started executing android test binary on host.
Make them run on android.

Hopefully fixes android build.

Change-Id: Ic0bb9f0cbbefca23828574282caa33a03ef72431
Reviewed-on: https://go-review.googlesource.com/62830
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Elias Naur <elias.naur@gmail.com>
7 years agodoc: fix typo in the diagnostics guide
Jaana Burcu Dogan [Sat, 9 Sep 2017 08:36:25 +0000 (01:36 -0700)]
doc: fix typo in the diagnostics guide

Change-Id: Ib1b93e55788168b574badeb84d0e6c4f3795055e
Reviewed-on: https://go-review.googlesource.com/62510
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/testcshared: simlpify cshared_test.go
Alex Brainman [Mon, 11 Sep 2017 01:58:17 +0000 (11:58 +1000)]
misc/cgo/testcshared: simlpify cshared_test.go

Change-Id: Ib35bb7fc9c5b4ccc9b8e1bd16443e0b307be9406
Reviewed-on: https://go-review.googlesource.com/62593
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/cgo: treat simple C typedefs as go aliases
Anthony Sottile [Sun, 10 Sep 2017 04:38:51 +0000 (21:38 -0700)]
cmd/cgo: treat simple C typedefs as go aliases

Fixes #21809

Change-Id: Ic43077c6bea3c7cdc9611e74abf07b6deab70433
Reviewed-on: https://go-review.googlesource.com/62670
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotest: add test case that gccgo crashed on
Ian Lance Taylor [Sun, 10 Sep 2017 15:52:33 +0000 (08:52 -0700)]
test: add test case that gccgo crashed on

Change-Id: I4d5d40e1ed3f58b2cdecd6248cb25c8ae9a1b9a1
Reviewed-on: https://go-review.googlesource.com/62531
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agocmd/compile: preserve escape information for map literals
Martin Möhrmann [Sun, 10 Sep 2017 17:36:38 +0000 (19:36 +0200)]
cmd/compile: preserve escape information for map literals

While some map literals were marked non-escaping that information
was lost when creating the corresponding OMAKE node which made map
literals always heap allocated.

Copying the escape information to the corresponding OMAKE node allows
stack allocation of hmap and a map bucket for non escaping map literals.

Fixes #21830

Change-Id: Ife0b020fffbc513f1ac009352f2ecb110d6889c9
Reviewed-on: https://go-review.googlesource.com/62790
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agointernal/cpu: add support for x86 FMA cpu feature detection
Martin Möhrmann [Sat, 9 Sep 2017 20:41:30 +0000 (22:41 +0200)]
internal/cpu: add support for x86 FMA cpu feature detection

Change-Id: I88ea39de01b07e6afa1c187c0df6a258da4aa8e4
Reviewed-on: https://go-review.googlesource.com/62650
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/link: don't generate runtime.text twice for buildmode=plugin on darwin
Hiroshi Ioka [Sun, 10 Sep 2017 22:51:57 +0000 (07:51 +0900)]
cmd/link: don't generate runtime.text twice for buildmode=plugin on darwin

https://golang.org/cl/29394 changed to include runtime.text and
runtime.etext in ctxt.Textp as a work around.
But it seems that the CL forgot to change genasmsym.
As a result, we are generating runtime.text and runtime.etext twice.

Change-Id: If7f8faf496c1c489ffa4804da712f91a3d3f4be4
Reviewed-on: https://go-review.googlesource.com/62810
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomisc/cgo/testcshared: run tests in parallel
Alex Brainman [Sun, 10 Sep 2017 05:26:19 +0000 (15:26 +1000)]
misc/cgo/testcshared: run tests in parallel

Change-Id: Id1b5939cfcd210a0cb5f61915ce2d077c7fcec11
Reviewed-on: https://go-review.googlesource.com/62592
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>