]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agocrypto/aes: make the GHASH part of AES-GCM faster
Vlad Krasnov [Fri, 18 Aug 2017 19:49:59 +0000 (12:49 -0700)]
crypto/aes: make the GHASH part of AES-GCM faster

By processing 8 blocks in parallel GHASH achieves higher throughput on amd64

Results on Skylake i7:

benchmark                   old ns/op     new ns/op     delta
BenchmarkAESGCMSeal1K-8     316           314           -0.63%
BenchmarkAESGCMOpen1K-8     282           281           -0.35%
BenchmarkAESGCMSign8K-8     5611          1099          -80.41%
BenchmarkAESGCMSeal8K-8     1869          1922          +2.84%
BenchmarkAESGCMOpen8K-8     1718          1724          +0.35%

benchmark                   old MB/s     new MB/s     speedup
BenchmarkAESGCMSeal1K-8     3237.10      3260.94      1.01x
BenchmarkAESGCMOpen1K-8     3629.74      3638.10      1.00x
BenchmarkAESGCMSign8K-8     1459.82      7452.99      5.11x
BenchmarkAESGCMSeal8K-8     4382.45      4260.93      0.97x
BenchmarkAESGCMOpen8K-8     4766.41      4750.54      1.00x

Change-Id: I479f2a791a968caa1c516115b0b6b96a791a20d2
Reviewed-on: https://go-review.googlesource.com/57150
Reviewed-by: Adam Langley <agl@golang.org>
7 years agoruntime: fix false positive race in profile label reading
Austin Clements [Thu, 17 Aug 2017 22:40:07 +0000 (18:40 -0400)]
runtime: fix false positive race in profile label reading

Because profile labels are copied from the goroutine into the tag
buffer by the signal handler, there's a carefully-crafted set of race
detector annotations to create the necessary happens-before edges
between setting a goroutine's profile label and retrieving it from the
profile tag buffer.

Given the constraints of the signal handler, we have to approximate
the true synchronization behavior. Currently, that approximation is
too weak.

Ideally, runtime_setProfLabel would perform a store-release on
&getg().labels and copying each label into the profile would perform a
load-acquire on &getg().labels. This would create the necessary
happens-before edges through each individual g.labels object.

Since we can't do this in the signal handler, we instead synchronize
on a "labelSync" global. The problem occurs with the following
sequence:

1. Goroutine 1 calls setProfLabel, which does a store-release on
   labelSync.

2. Goroutine 2 calls setProfLabel, which does a store-release on
   labelSync.

3. Goroutine 3 reads the profile, which does a load-acquire on
   labelSync.

The problem is that the load-acquire only synchronizes with the *most
recent* store-release to labelSync, and the two store-releases don't
synchronize with each other. So, once goroutine 3 touches the label
set by goroutine 1, we report a race.

The solution is to use racereleasemerge. This is like a
read-modify-write, rather than just a store-release. Each RMW of
labelSync in runtime_setProfLabel synchronizes with the previous RMW
of labelSync, and this ultimately carries forward to the load-acquire,
so it synchronizes with *all* setProfLabel operations, not just the
most recent.

Change-Id: Iab58329b156122002fff12cfe64fbeacb31c9613
Reviewed-on: https://go-review.googlesource.com/56670
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
7 years agocmd/compile/internal/amd64: add ADD[Q|L]constmem
Ilya Tocar [Wed, 9 Aug 2017 20:43:25 +0000 (15:43 -0500)]
cmd/compile/internal/amd64: add ADD[Q|L]constmem

We can add a constant to loaction in memory with 1 instruction,
as opposed to load+add+store, so add a new op and relevent ssa rules.
Triggers in e. g. encoding/json isValidNumber:
NumberIsValid-6          36.4ns ± 0%    35.2ns ± 1%  -3.32%  (p=0.000 n=6+10)
Shaves ~2.5 kb from go tool.

Change-Id: I7ba576676c2522432360f77b290cecb9574a93c3
Reviewed-on: https://go-review.googlesource.com/54431
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/*: remove negative uint checks
Daniel Martí [Fri, 18 Aug 2017 15:05:33 +0000 (16:05 +0100)]
cmd/*: remove negative uint checks

All of these are uints of different sizes, so checking >= 0 or < 0 are
effectively no-ops.

Found with staticcheck.

Change-Id: I16ac900eb7007bc8f9018b302136d42e483a4180
Reviewed-on: https://go-review.googlesource.com/56950
Reviewed-by: Matt Layher <mdlayher@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Matt Layher <mdlayher@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: make evacDst a top level type
Josh Bleecher Snyder [Fri, 18 Aug 2017 01:25:54 +0000 (18:25 -0700)]
runtime: make evacDst a top level type

This will reduce duplication when evacuate is specialized.

Change-Id: I34cdfb7103442d3e0ea908c970fb46334b86d5c4
Reviewed-on: https://go-review.googlesource.com/56934
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Avelino <t@avelino.xxx>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: split advanceEvacuationMark from evacuate
Josh Bleecher Snyder [Thu, 17 Aug 2017 17:37:17 +0000 (10:37 -0700)]
runtime: split advanceEvacuationMark from evacuate

Minor refactoring. This is a step towards specializing evacuate
for mapfast key types.

Change-Id: Icffe2759b7d38e5c008d03941918d5a912ce62f6
Reviewed-on: https://go-review.googlesource.com/56933
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoruntime: tiny refactor in evacuate
Josh Bleecher Snyder [Thu, 17 Aug 2017 17:24:13 +0000 (10:24 -0700)]
runtime: tiny refactor in evacuate

Since oldbucket == h.nevacuate, we can just increment h.nevacuate here.
This removes oldbucket from scope, which will be useful shortly.

Change-Id: I70f81ec3995f17845ebf5d77ccd20ea4338f23e6
Reviewed-on: https://go-review.googlesource.com/56932
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Avelino <t@avelino.xxx>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: don't cache t.key.alg in evacuate
Josh Bleecher Snyder [Fri, 18 Aug 2017 01:01:05 +0000 (18:01 -0700)]
runtime: don't cache t.key.alg in evacuate

The number of times that alg has to be spilled
and restored makes it better to just reload it.

Change-Id: I2674752a889ecad59dab54da1d68fad03db1ca85
Reviewed-on: https://go-review.googlesource.com/56931
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: simplify evacuate's handling of NaNs
Josh Bleecher Snyder [Fri, 18 Aug 2017 00:41:44 +0000 (17:41 -0700)]
runtime: simplify evacuate's handling of NaNs

The new code is not quite equivalent to the old,
in that if newbit was very large it might have altered the new tophash.
The old behavior is unnecessary and probably undesirable.

Change-Id: I7fb3222520cb61081a857adcddfbb9078ead7122
Reviewed-on: https://go-review.googlesource.com/56930
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agobytes: clean-up of buffer.go
Marvin Stenger [Sun, 7 May 2017 14:23:04 +0000 (16:23 +0200)]
bytes: clean-up of buffer.go

Clean-up changes in no particular order:
- use uint8 instead of int for readOp
- remove duplicated code in ReadFrom()
- introduce (*Buffer).empty()
- remove naked returns

Change-Id: Ie6e673c20c398f980f8be0448969a36ad4778804
Reviewed-on: https://go-review.googlesource.com/42816
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: add rules handling unsigned div/mod by constant 1<<63
Cherry Zhang [Fri, 18 Aug 2017 12:37:58 +0000 (08:37 -0400)]
cmd/compile: add rules handling unsigned div/mod by constant 1<<63

Normally 64-bit div/mod is turned into runtime calls on 32-bit
arch, but the front end leaves power-of-two constant division
and hopes the SSA backend turns into a shift or AND. The SSA rule is

(Mod64u <t> n (Const64 [c])) && isPowerOfTwo(c) -> (And64 n (Const64 <t> [c-1]))

But isPowerOfTwo returns true only for positive int64, which leaves
out 1<<63 unhandled. Add a special case for 1<<63.

Fixes #21517.

Change-Id: I02d27dc7177d4af0ee8d7f5533714edecddf8c95
Reviewed-on: https://go-review.googlesource.com/56890
Reviewed-by: Keith Randall <khr@golang.org>
7 years agotime: add leap year test for Date
Kevin Burke [Thu, 17 Aug 2017 15:16:57 +0000 (08:16 -0700)]
time: add leap year test for Date

I'm writing a matching implementation of the time package and missed
the "add one day in a leap year" block. This test would have caught my
error.

I understand we can't add test cases for every Date but it seems like
"tripped up someone attempting to reimplement this" is a good
indicator it may trip up people in the future.

Change-Id: I4c3b51e52e269215ec0e52199afe604482326edb
Reviewed-on: https://go-review.googlesource.com/56490
Reviewed-by: Matt Layher <mdlayher@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Matt Layher <mdlayher@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/vendor/golang.org/x/arch: import latest (Aug 18 2017 ffd22fb365cd)
Russ Cox [Fri, 18 Aug 2017 16:11:11 +0000 (12:11 -0400)]
cmd/vendor/golang.org/x/arch: import latest (Aug 18 2017 ffd22fb365cd)

Fixes #21486.

Change-Id: I01794f008404f0e2d8a1408309ae6055513c5b49
Reviewed-on: https://go-review.googlesource.com/57030
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/asm: uncomment tests for amd64 PHADD{SW,W}, PHSUB{D,SW,W}
isharipo [Fri, 18 Aug 2017 11:39:53 +0000 (14:39 +0300)]
cmd/asm: uncomment tests for amd64 PHADD{SW,W}, PHSUB{D,SW,W}

Instructions added in https://golang.org/cl/18853

2nd change out of 3 to cover AMD64 SSSE3 instruction set in Go asm.
This commit does not actually add any new instructions, only
enables some test cases.

Change-Id: I9596435b31ee4c19460a51dd6cea4530aac9d198
Reviewed-on: https://go-review.googlesource.com/56835
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agocmd/dist: disable broken TestDeps
Russ Cox [Fri, 18 Aug 2017 15:06:20 +0000 (11:06 -0400)]
cmd/dist: disable broken TestDeps

Will debug soon but wanted to fix builders.

Change-Id: I921d58d1272370f3102ba1f86ad535f4c0f6b8db
Reviewed-on: https://go-review.googlesource.com/56970
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
7 years agoruntime: no need to protect key/value increments against end of bucket
Keith Randall [Fri, 18 Aug 2017 06:13:57 +0000 (23:13 -0700)]
runtime: no need to protect key/value increments against end of bucket

After the key and value arrays, we have an overflow pointer.
So there's no way a past-the-end key or value pointer could point
past the end of the containing bucket.

So we don't need this additional protection.

Update #21459

Change-Id: I7726140033b06b187f7a7d566b3af8cdcaeab0b0
Reviewed-on: https://go-review.googlesource.com/56772
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Avelino <t@avelino.xxx>
7 years agocmd/go: remove Package.Internal.Deps
Russ Cox [Fri, 9 Jun 2017 15:36:10 +0000 (11:36 -0400)]
cmd/go: remove Package.Internal.Deps

Package.Internal.Imports is enough in nearly all cases,
and not maintaining a separate Package.Internal.Deps
avoids the two lists ending up out of sync.
(In some synthesized packages created during go test,
only Internal.Imports is initialized.)

Change-Id: I83f6a3ec6e6cbd75382f1fa0e439d31feec32d5a
Reviewed-on: https://go-review.googlesource.com/56278
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: use objdir as consistent variable name for per-package work dir
Russ Cox [Fri, 9 Jun 2017 14:45:49 +0000 (10:45 -0400)]
cmd/go: use objdir as consistent variable name for per-package work dir

Before it was obj, but if you don't have everything paged in
that sounds a bit like an object file. Use objdir, which is more
clearly a directory and also matches the Action.Objdir struct field.

Change-Id: I268042800f9ca05721814d7f18c728acb4831232
Reviewed-on: https://go-review.googlesource.com/56277
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: document that BinaryOnly packages must have accurate import info
Russ Cox [Fri, 9 Jun 2017 16:29:43 +0000 (12:29 -0400)]
cmd/go: document that BinaryOnly packages must have accurate import info

Update BinaryOnly test by adding import _ "fmt".

Change-Id: I3a1dcfb83a27d8ff50a658060a46e1a3f481f6c7
Reviewed-on: https://go-review.googlesource.com/56276
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: rename local variable ImportPaths to importPaths
Russ Cox [Mon, 17 Jul 2017 16:07:26 +0000 (12:07 -0400)]
cmd/go: rename local variable ImportPaths to importPaths

ImportPaths is also the name of a top-level function.
It is confusing to have a capitalized local variable.

Change-Id: I1313e05ade4934d4ee250a67e5af6d1bd6229aca
Reviewed-on: https://go-review.googlesource.com/56275
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: add t.Helper calls to test helpers
Russ Cox [Mon, 17 Jul 2017 14:30:40 +0000 (10:30 -0400)]
cmd/go: add t.Helper calls to test helpers

Now that we have t.Helper, might as well use it to make the
reported failure lines more helpful.

Change-Id: I2a0c64e9ca7bdc0eaf2b62f9f855c41467767084
Reviewed-on: https://go-review.googlesource.com/56274
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
7 years agocmd/go: make TestBuildDashIInstallsDependencies not depend only on time
Russ Cox [Sun, 16 Jul 2017 18:32:19 +0000 (14:32 -0400)]
cmd/go: make TestBuildDashIInstallsDependencies not depend only on time

When we make the go command pay attention to content
instead of time, we want this test to continue working.

Change-Id: Ib7d9d0d62bfe87810d71bdfc4f29561a8c70eccc
Reviewed-on: https://go-review.googlesource.com/56273
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: rewrite TestCgoFlagContainsSpace not to use a fake CC
Russ Cox [Fri, 9 Jun 2017 15:44:31 +0000 (11:44 -0400)]
cmd/go: rewrite TestCgoFlagContainsSpace not to use a fake CC

Using a fake CC fails today if runtime/cgo is stale, because the
build will try to rebuild runtime/cgo using the fake CC, and the
fake CC is not a working C compiler.

Worse, in the future, when the go command is sensitive to details like
the fact that different CCs produce different outputs, putting in
the fake CC will make runtime/cgo look stale even if it was
formerly up-to-date.

Fix both problems by not overriding CC and instead looking at
the command being run to make sure the flags are quoted as expected.

Change-Id: I4417e35cfab33a07546cc90748ddb6119d8fdb2d
Reviewed-on: https://go-review.googlesource.com/56272
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: change testMainDeps from map to slice
Russ Cox [Thu, 8 Jun 2017 14:17:47 +0000 (10:17 -0400)]
cmd/go: change testMainDeps from map to slice

This makes the construction of pmain.Internal.Imports consistently ordered.

Change-Id: I82348a18c7824378aa7e5bc5b6bcd550d4b758da
Reviewed-on: https://go-review.googlesource.com/56271
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/internal/obj/arm: support new arm instructions
Ben Shi [Tue, 8 Aug 2017 08:14:24 +0000 (08:14 +0000)]
cmd/internal/obj/arm: support new arm instructions

There are two changes in this CL.

1. Add new forms of MOVH/MOVHS/MOVHU.
   MOVHS R0<<0(R1), R2   // ldrsh
   MOVH  R0<<0(R1), R2   // ldrsh
   MOVHU R0<<0(R1), R2   // ldrh
   MOVHS R2, R5<<0(R1)   // strh
   MOVH  R2, R5<<0(R1)   // strh
   MOVHU R2, R5<<0(R1)   // strh

2. Simpify "MVN $0xffffffaa, Rn" to "MOVW $0x55, Rn".
   It is originally assembled to two instructions.
   "MOVW offset(PC), R11"
   "MVN R11, Rn"

Change-Id: I8e863dcfb2bd8f21a04c5d627fa7beec0afe65fb
Reviewed-on: https://go-review.googlesource.com/53690
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agoruntime: fix usleep by correctly setting nanoseconds parameter for pselect6
pvoicu [Fri, 18 Aug 2017 10:32:49 +0000 (03:32 -0700)]
runtime: fix usleep by correctly setting nanoseconds parameter for pselect6

Fixes #21518

Change-Id: Idd67e3f0410d0ce991b34dcc0c8f15e0d5c529c9
Reviewed-on: https://go-review.googlesource.com/56850
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomath: implement the erfinv function
Lakshay Garg [Wed, 28 Jun 2017 16:37:55 +0000 (22:07 +0530)]
math: implement the erfinv function

This commit defines the inverse of error function (erfinv) in the
math package. The function is based on the rational approximation
of percentage points of normal distribution available at
https://www.jstor.org/stable/pdf/2347330.pdf.

Fixes #6359

Change-Id: Icfe4508f623e0574c7fffdbf7aa929540fd4c944
Reviewed-on: https://go-review.googlesource.com/46990
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agomath/big: recognize squaring for Floats
Brian Kessler [Fri, 18 Aug 2017 06:26:31 +0000 (23:26 -0700)]
math/big: recognize squaring for Floats

Updates #13745

Recognize z.Mul(x, x) as squaring for Floats and use
the internal z.sqr(x) method for nat on the mantissa.

Change-Id: I0f792157bad93a13cae1aecc4c10bd20c6397693
Reviewed-on: https://go-review.googlesource.com/56774
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomath/big: use internal square for Rat
Brian Kessler [Fri, 18 Aug 2017 06:44:34 +0000 (23:44 -0700)]
math/big: use internal square for Rat

updates #13745

A squared rational is always positive and can not
be reduced since the numerator and denominator had
no previous common factors.  The nat multiplication
can be performed using the internal sqr method.

Change-Id: I558f5b38e379bfd26ff163c9489006d7e5a9cfaa
Reviewed-on: https://go-review.googlesource.com/56776
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix language in makeslice comment
Martin Möhrmann [Mon, 14 Aug 2017 08:35:17 +0000 (10:35 +0200)]
cmd/compile: fix language in makeslice comment

Change-Id: I1929ea7e4ed88631ef729472ffe474016efec3e8
Reviewed-on: https://go-review.googlesource.com/56370
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoall: unindent some big chunks of code
Daniel Martí [Thu, 17 Aug 2017 14:51:35 +0000 (15:51 +0100)]
all: unindent some big chunks of code

Found with mvdan.cc/unindent. Prioritized the ones with the biggest wins
for now.

Change-Id: I2b032e45cdd559fc9ed5b1ee4c4de42c4c92e07b
Reviewed-on: https://go-review.googlesource.com/56470
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/testcshared: fix syntax error in the test
Alex Brainman [Fri, 18 Aug 2017 04:55:05 +0000 (14:55 +1000)]
misc/cgo/testcshared: fix syntax error in the test

Another attempt to fix build

Change-Id: I26137c115ad4b5f5a69801ed981c146adf6e824c
Reviewed-on: https://go-review.googlesource.com/56750
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomisc/cgo/testcshared: use adb instead of ./adb on android
Alex Brainman [Fri, 18 Aug 2017 03:53:42 +0000 (13:53 +1000)]
misc/cgo/testcshared: use adb instead of ./adb on android

Hopefully fixes build.

Change-Id: If0629b95b923a65e4507073cf7aa44a5e178fc0f
Reviewed-on: https://go-review.googlesource.com/56711
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

7 years agomisc/cgo/testcshared: rewrite test.bash in Go
Christopher Nelson [Mon, 28 Nov 2016 00:05:01 +0000 (19:05 -0500)]
misc/cgo/testcshared: rewrite test.bash in Go

Change-Id: Id717054cb3c4537452f8ff848445b0c20196a373
Reviewed-on: https://go-review.googlesource.com/33579
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
7 years agodebug/macho: support LC_RPATH
Hiroshi Ioka [Thu, 17 Aug 2017 00:11:36 +0000 (09:11 +0900)]
debug/macho: support LC_RPATH

Updates #21487

Change-Id: Ia549a87a8a305cc80da11ea9bd904402f1a14689
Reviewed-on: https://go-review.googlesource.com/56321
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: show native relocation type name in error messages
Hiroshi Ioka [Thu, 17 Aug 2017 00:45:20 +0000 (09:45 +0900)]
cmd/link: show native relocation type name in error messages

Change-Id: I7f7b1e7ef832d53a93562b08ae914d023247c2c0
Reviewed-on: https://go-review.googlesource.com/56312
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: rename SSA Register.Name to Register.String
Keith Randall [Thu, 17 Aug 2017 19:23:34 +0000 (12:23 -0700)]
cmd/compile: rename SSA Register.Name to Register.String

Just to get rid of lots of .Name() stutter in printf calls.

Change-Id: I86cf00b3f7b2172387a1c6a7f189c1897fab6300
Reviewed-on: https://go-review.googlesource.com/56630
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
7 years agoruntime: improve makechan memory checks and allocation calls
Martin Möhrmann [Sat, 12 Aug 2017 12:37:19 +0000 (14:37 +0200)]
runtime: improve makechan memory checks and allocation calls

Use mallogc instead of newarray to save some overhead since
makechan already checks for _MaxMem constraints.

Flattens the if else construct that determines if buf and hchan struct
should be allocated in one mallocgc call and where buf should point to.

Uses maxSliceCap to avoid divisions similar to makeslice.

name                old time/op  new time/op  delta
MakeChan/Byte       82.0ns ± 8%  81.4ns ± 7%    ~     (p=0.643 n=10+10)
MakeChan/Int        97.9ns ± 2%  96.6ns ± 2%  -1.40%  (p=0.009 n=10+10)
MakeChan/Ptr         128ns ± 3%   120ns ± 1%  -6.63%  (p=0.000 n=10+10)
MakeChan/Struct/0   66.7ns ± 4%  66.4ns ± 2%    ~     (p=0.697 n=10+10)
MakeChan/Struct/32   136ns ± 1%   130ns ± 0%  -4.42%  (p=0.000 n=10+10)
MakeChan/Struct/40   150ns ± 1%   150ns ± 1%    ~     (p=0.725 n=10+10)

Change-Id: Ibb5675d0843a072aae2bfa58ecd39cf4cd926533
Reviewed-on: https://go-review.googlesource.com/55132
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoruntime: avoid zeroing hmap fields in makemap twice
Martin Möhrmann [Wed, 16 Aug 2017 21:52:16 +0000 (23:52 +0200)]
runtime: avoid zeroing hmap fields in makemap twice

Stack allocated hmap structs are explicitly zeroed before being
passed by pointer to makemap.

Heap allocated hmap structs are created with newobject
which also zeroes on allocation.

Therefore, setting the hmap fields to 0 or nil is redundant
since they will have been zeroed when hmap was allocated.

Change-Id: I5fc55b75e9dc5ba69f5e3588d6c746f53b45ba66
Reviewed-on: https://go-review.googlesource.com/56291
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/go: add fossil to general server regexp in get
Kyle Shannon [Wed, 16 Aug 2017 22:51:22 +0000 (16:51 -0600)]
cmd/go: add fossil to general server regexp in get

Fix a missed change from:

https://golang.org/cl/56190

pointed out on the fossil mailing list shortly after submission
of the change mentioned above.  See:

http://www.mail-archive.com/fossil-users@lists.fossil-scm.org/msg25736.html

This change adds fossil to the general regular expression that is checked last
in the import path check.

For #10010

Change-Id: I6b711cdb1a8d4d767f61e1e28dc29dce529e0fad
Reviewed-on: https://go-review.googlesource.com/56491
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years ago{net,os/user,plugin}: eliminate unnecessary C round-trips
Bryan C. Mills [Sat, 13 May 2017 04:01:50 +0000 (00:01 -0400)]
{net,os/user,plugin}: eliminate unnecessary C round-trips

We're making two extra round-trips to C to malloc and free strings
that originate in Go and don't escape. Skip those round-trips by
allocating null-terminated slices in Go memory instead.

Change-Id: I9e4c5ad999a7924ba50b82293c52073ec75518be
Reviewed-on: https://go-review.googlesource.com/56530
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotest: add test that caused gccgo compiler crash
Ian Lance Taylor [Tue, 1 Aug 2017 22:06:36 +0000 (15:06 -0700)]
test: add test that caused gccgo compiler crash

Updates #21253

Change-Id: Iece71a27207b578618cafb378dac2362517363d0
Reviewed-on: https://go-review.googlesource.com/52531
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile/internal/ssa: combine more const stores
Ilya Tocar [Wed, 16 Aug 2017 19:01:48 +0000 (14:01 -0500)]
cmd/compile/internal/ssa: combine more const stores

We already combine const stores up-to MOVQstoreconst.
Combine 2 64-bit stores of const zero into 1 sse store of 128-bit zero.

Shaves significant (>1%) amount of code from go tool:
/localdisk/itocar/golang/bin/go 10334877
go_old 10388125 [53248 bytes]

global text (code) = 51041 bytes (1.343944%)
read-only data = 663 bytes (0.039617%)
Total difference 51704 bytes (0.873981%)

Change-Id: I7bc40968023c3a69f379b10fbb433cdb11364f1b
Reviewed-on: https://go-review.googlesource.com/56250
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/asm: uncomment tests for PCMPESTRI, PHMINPOSUW
isharipo [Tue, 15 Aug 2017 17:31:45 +0000 (20:31 +0300)]
cmd/asm: uncomment tests for PCMPESTRI, PHMINPOSUW

Instructions are implemented in the following revisions:
PCMPESTRI - https://golang.org/cl/22337
PHMINPOSUW - https://golang.org/cl/18853

It is unknown when x86test will be updated/re-run, but tests are useful
to check which x86 instructions are not yet supported.
As an example of tool that uses this information, there is Damien
Lespiau x86db.

Part of the mission to add missing amd64 SSE4 instructions to Go asm.

Change-Id: I512ff26040f47a0976b3e37000fb1f37eac5b762
Reviewed-on: https://go-review.googlesource.com/55830
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agomisc/cgo/testsanitizers: convert test.bash to Go
Bryan C. Mills [Tue, 1 Aug 2017 23:30:57 +0000 (19:30 -0400)]
misc/cgo/testsanitizers: convert test.bash to Go

This makes it much easier to run individual failing subtests.

Use $(go env CC) instead of always defaulting to clang; this makes it
easier to test with other compilers.

Run C binaries to detect incompatible compiler/kernel pairings instead
of sniffing versions.

updates #21196

Change-Id: I0debb3cc4a4244df44b825157ffdc97b5c09338d
Reviewed-on: https://go-review.googlesource.com/52910
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomath: fix inaccurate result of Exp(1)
crvv [Tue, 18 Jul 2017 06:37:01 +0000 (14:37 +0800)]
math: fix inaccurate result of Exp(1)

The existing implementation is translated from C, which uses a
polynomial coefficient very close to 1/6. If the function uses
1/6 as this coeffient, the result of Exp(1) will be more accurate.
And this change doesn't introduce more error to Exp function.

Fixes #20319

Change-Id: I94c236a18cf95570ebb69f7fb99884b0d7cf5f6e
Reviewed-on: https://go-review.googlesource.com/49294
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: tweaks to unindent some code
Daniel Martí [Thu, 10 Aug 2017 14:41:17 +0000 (23:41 +0900)]
cmd/compile: tweaks to unindent some code

Prioritized the chunks of code with 8 or more levels of indentation.
Basically early breaks/returns and joining nested ifs.

Change-Id: I6817df1303226acf2eb904a29f2db720e4f7427a
Reviewed-on: https://go-review.googlesource.com/55630
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agocmd/link: correct runtime/cgo detection for -linkshared
Hiroshi Ioka [Wed, 16 Aug 2017 22:26:07 +0000 (07:26 +0900)]
cmd/link: correct runtime/cgo detection for -linkshared

Current code detect runtime/cgo iff the package or sub packages imports
runtime/cgo directly. However, when we are using linkshared, imported
shared libraries might have already included runtime/cgo.
This CL handles later case by looking an actual runtime/cgo symbol.

Change-Id: I35e7dfdb5e1a939eafc95a0259ee1af9782bc864
Reviewed-on: https://go-review.googlesource.com/56310
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 agodebug/macho: add comments for ambiguious LoadCommand
Hiroshi Ioka [Thu, 17 Aug 2017 00:16:32 +0000 (09:16 +0900)]
debug/macho: add comments for ambiguious LoadCommand

While LoadCmdDylib represents LC_LOAD_DYLIB,
LoadCmdDylinker represents LC_ID_DYLINKER.
This is confusing because there is another command called LC_LOAD_DYLINKER.
LC_ID_DYLINKER is not included in normal binary, it is only used for
/usr/lib/dyld as far as I know. So, perhaps this is a mistake.

Change-Id: I6ea61664a26998962742914af5688e094a233541
Reviewed-on: https://go-review.googlesource.com/56330
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 agodebug/macho: make tests more comprehensive
Hiroshi Ioka [Wed, 16 Aug 2017 23:54:39 +0000 (08:54 +0900)]
debug/macho: make tests more comprehensive

add tests for LC_LOAD_DYLIB.

Change-Id: Ic4b7a0f6296709175e9a75240aecd1d5291ade4b
Reviewed-on: https://go-review.googlesource.com/56311
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/dist: add test that deps.go is up to date
Ian Lance Taylor [Tue, 15 Aug 2017 23:58:51 +0000 (16:58 -0700)]
cmd/dist: add test that deps.go is up to date

Test is not run in short mode, except on builders.

Change-Id: I4456830770188951e05ac13669e834a25bf569ae
Reviewed-on: https://go-review.googlesource.com/55973
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Joe Tsai <joetsai@google.com>
Reviewed-by: Joe Tsai <joetsai@google.com>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/internal/obj/x86: don't apply workaround for solaris to darwin
Hiroshi Ioka [Sat, 5 Aug 2017 11:03:32 +0000 (20:03 +0900)]
cmd/internal/obj/x86: don't apply workaround for solaris to darwin

Currently, we have a workaround for solaris that enforce aboslute
addressing for external symbols. However, We don't want to use the
workaround for darwin.
This CL also refactors code a little bit, because the original function
name is not appropriate now.

Updates #17490

Change-Id: Id21f9cdf33dca6a40647226be49010c2c324ee24
Reviewed-on: https://go-review.googlesource.com/54871
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 agodebug/macho: rearrange code
Hiroshi Ioka [Wed, 16 Aug 2017 11:50:26 +0000 (20:50 +0900)]
debug/macho: rearrange code

* group load command structs.
* use hex literal for LoadCommand.
  Decimal number is not a proper representation for some commands.
  (e.g. LC_RPATH = 0x8000001c)
* move Symbol struct from macho.go to file.go.
  Symbol is a high level representation, not in Mach-O.

Change-Id: I3c69923cb464fb1211f2e766c02e1b537e0b5de2
Reviewed-on: https://go-review.googlesource.com/56130
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/go: add support for Fossil SCM to go get
Kyle Shannon [Wed, 9 Aug 2017 16:27:48 +0000 (10:27 -0600)]
cmd/go: add support for Fossil SCM to go get

Fixes #10010.

Change-Id: Ib13ac28eafed72c456d8b5b6549015cdf5fdda94
Reviewed-on: https://go-review.googlesource.com/56190
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: unify cgo output for gc and gccgo
Michael Steinert [Mon, 17 Jul 2017 16:14:23 +0000 (11:14 -0500)]
cmd/cgo: unify cgo output for gc and gccgo

When calling a Go function that returns multiple values from C, cgo
generates a structure to hold the values. According to the documentation
this structure is called `struct <function-name>_return`. When compiling
for gccgo the generated structure name is `struct <function-name>_result`.
This change updates the output for gccgo to match the documentation and
output for gc.

Fixes #20910

Change-Id: Iaea8030a695a7aaf9d9f317447fc05615d8e4adc
Reviewed-on: https://go-review.googlesource.com/49350
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/dist: update deps.go for current dependencies
Ian Lance Taylor [Tue, 15 Aug 2017 23:10:12 +0000 (16:10 -0700)]
cmd/dist: update deps.go for current dependencies

Fixes #21456

Change-Id: Iba7bc608686536b2d4fe3d23409fa84b59cea640
Reviewed-on: https://go-review.googlesource.com/55971
Reviewed-by: Joe Tsai <joetsai@google.com>
7 years agobytes: avoid overflow in (*Buffer).Grow and ReadFrom
Bryan C. Mills [Wed, 16 Aug 2017 18:58:19 +0000 (14:58 -0400)]
bytes: avoid overflow in (*Buffer).Grow and ReadFrom

fixes #21481

Change-Id: I26717876a1c0ee25a86c81159c6b3c59563dfec6
Reviewed-on: https://go-review.googlesource.com/56230
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoarchive/tar: add reader/writer benchmarks
Agniva De Sarker [Sat, 12 Aug 2017 15:05:40 +0000 (20:35 +0530)]
archive/tar: add reader/writer benchmarks

According to the discussion on golang.org/cl/55210,
adding benchmarks for reading from and writing to tar archives.

Splitting the benchmarks into 3 sections of USTAR, GNU, PAX each.

Results ran with -cpu=1 -count=10 on an amd64 machine (i5-5200U CPU @ 2.20GHz)
name           time/op
/Writer/USTAR  5.31µs ± 0%
/Writer/GNU    5.01µs ± 1%
/Writer/PAX    11.0µs ± 2%
/Reader/USTAR  3.22µs ± 1%
/Reader/GNU    3.04µs ± 1%
/Reader/PAX    7.48µs ± 1%

name           alloc/op
/Writer/USTAR  1.20kB ± 0%
/Writer/GNU    1.15kB ± 0%
/Writer/PAX    2.61kB ± 0%
/Reader/USTAR  1.38kB ± 0%
/Reader/GNU    1.35kB ± 0%
/Reader/PAX    4.91kB ± 0%

name           allocs/op
/Writer/USTAR    53.0 ± 0%
/Writer/GNU      47.0 ± 0%
/Writer/PAX       107 ± 0%
/Reader/USTAR    32.0 ± 0%
/Reader/GNU      30.0 ± 0%
/Reader/PAX      67.0 ± 0%

Change-Id: I58b1b85b52e58cbd566736aae4d722a3ddf2395b
Reviewed-on: https://go-review.googlesource.com/55254
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/link: skip TestFieldOverlap on Plan 9
David du Colombier [Wed, 16 Aug 2017 18:30:26 +0000 (20:30 +0200)]
cmd/link: skip TestFieldOverlap on Plan 9

TestSizes has been added in CL 50170. This test is
failing on Plan 9 because executables don't have
a DWARF symbol table.

Fixes #21480.

Change-Id: I51079abdc18ad944617bdbcfe2dad970a0cea0f2
Reviewed-on: https://go-review.googlesource.com/56210
Run-TryBot: David du Colombier <0intro@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime/cgo: defeat inlining in x_cgo_yield
Bryan C. Mills [Mon, 7 Aug 2017 23:30:03 +0000 (19:30 -0400)]
runtime/cgo: defeat inlining in x_cgo_yield

We use a call to strncpy to work around a TSAN bug (wherein TSAN only
delivers asynchronous signals when the thread receiving the signal
calls a libc function). Unfortunately, GCC 7 inlines the call,
avoiding the TSAN libc trap entirely.

Per Ian's suggestion, use global variables as strncpy arguments: that
way, the compiler can't make any assumptions about the concrete values
and can't inline the call away.

fixes #21196

Change-Id: Ie95f1feaf9af1a8056f924f49c29cfc8515385d7
Reviewed-on: https://go-review.googlesource.com/55872
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agofmt: add Stringer example
Blain Smith [Mon, 17 Jul 2017 15:42:25 +0000 (09:42 -0600)]
fmt: add Stringer example

Change-Id: I901f995f8aedee47c48252745816e53192d4b7e4
Reviewed-on: https://go-review.googlesource.com/49090
Reviewed-by: Sam Whited <sam@samwhited.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Sam Whited <sam@samwhited.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomisc/nacl: add cmd/vendor/golang.org/x/arch/arm64/arm64asm testdata
Cherry Zhang [Tue, 15 Aug 2017 13:50:44 +0000 (09:50 -0400)]
misc/nacl: add cmd/vendor/golang.org/x/arch/arm64/arm64asm testdata

This should fix NaCl build failure for CL 49530.

Change-Id: Id9a54f0c81b1b5db5b5efb12a2ad6509c4ab42b3
Reviewed-on: https://go-review.googlesource.com/55770
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

7 years agocmd/vendor/golang.org/x/arch: pull updates from x repo
Wei Xiao [Tue, 18 Jul 2017 04:25:37 +0000 (12:25 +0800)]
cmd/vendor/golang.org/x/arch: pull updates from x repo

Vendor from golang.org/x/arch (commit f185940).

Implements #19157

Updates #12840
Updates #20762
Updates #20897
Updates #20096
Updates #20766
Updates #20752
Updates #20096
Updates #19142

Change-Id: Idefb8ba2c355dc07f3b9e8dcf5f00173256a0f0f
Reviewed-on: https://go-review.googlesource.com/49530
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: combine x*n + y*n into (x+y)*n
Alberto Donizetti [Mon, 14 Aug 2017 09:44:09 +0000 (11:44 +0200)]
cmd/compile: combine x*n + y*n into (x+y)*n

There are a few cases where this can be useful. Apart from the obvious
(and silly)

  100*n + 200*n

where we generate one IMUL instead of two, consider:

  15*n + 31*n

Currently, the compiler strength-reduces both imuls, generating:

    0x0000 00000 MOVQ "".n+8(SP), AX
0x0005 00005  MOVQ AX, CX
0x0008 00008  SHLQ $4, AX
0x000c 00012  SUBQ CX, AX
0x000f 00015  MOVQ CX, DX
0x0012 00018  SHLQ $5, CX
0x0016 00022  SUBQ DX, CX
0x0019 00025  ADDQ CX, AX
0x001c 00028  MOVQ AX, "".~r1+16(SP)
0x0021 00033  RET

But combining the imuls is both faster and shorter:

0x0000 00000 MOVQ "".n+8(SP), AX
0x0005 00005  IMULQ $46, AX
0x0009 00009 MOVQ AX, "".~r1+16(SP)
0x000e 00014  RET

even without strength-reduction.

Moreover, consider:

  5*n + 7*(n+1) + 11*(n+2)

We already have a rule that rewrites 7(n+1) into 7n+7, so the
generated code (without imuls merging) looks like this:

0x0000 00000  MOVQ "".n+8(SP), AX
0x0005 00005  LEAQ (AX)(AX*4), CX
0x0009 00009  MOVQ AX, DX
0x000c 00012  NEGQ AX
0x000f 00015  LEAQ (AX)(DX*8), AX
0x0013 00019  ADDQ CX, AX
0x0016 00022  LEAQ (DX)(CX*2), CX
0x001a 00026  LEAQ 29(AX)(CX*1), AX
0x001f 00031  MOVQ AX, "".~r1+16(SP)

But with imuls merging, the 5n, 7n and 11n factors get merged, and the
generated code looks like this:

0x0000 00000  MOVQ "".n+8(SP), AX
0x0005 00005  IMULQ $23, AX
0x0009 00009  ADDQ $29, AX
0x000d 00013  MOVQ AX, "".~r1+16(SP)
0x0012 00018  RET

Which is both faster and shorter; that's also the exact same code that
clang and the intel c compiler generate for the above expression.

Change-Id: Ib4d5503f05d2f2efe31a1be14e2fe6cac33730a9
Reviewed-on: https://go-review.googlesource.com/55143
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/link: fix bad dwarf for sudog<T>
Keith Randall [Thu, 20 Jul 2017 02:52:29 +0000 (19:52 -0700)]
cmd/link: fix bad dwarf for sudog<T>

The DWARF entries for type-specific sudog entries used the
channel value type instead of a pointer-to-value type for the elem field.

Fixes #21094

R=go1.10

Change-Id: I3f63a5664f42b571f729931309f2c9f6f38ab031
Reviewed-on: https://go-review.googlesource.com/50170
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile/internal/ssa: use sse to zero on amd64
Ilya Tocar [Wed, 9 Aug 2017 19:50:58 +0000 (14:50 -0500)]
cmd/compile/internal/ssa: use sse to zero on amd64

Use 16-byte stores instead of 8-byte stores to zero small blocks.
Also switch to duffzero for 65+ bytes only, because for each
duffzero call we also save/restore BP, so call requires 4 instructions
and replacing it with 4 sse stores doesn't cause code-bloat.
Also switch duffzero to use leaq, instead of addq to avoid clobbering flags.

ClearFat8-6     0.54ns ± 0%  0.54ns ± 0%     ~     (all equal)
ClearFat12-6    1.07ns ± 0%  1.07ns ± 0%     ~     (all equal)
ClearFat16-6    1.07ns ± 0%  0.69ns ± 0%  -35.51%  (p=0.001 n=8+9)
ClearFat24-6    1.61ns ± 1%  1.07ns ± 0%  -33.33%  (p=0.000 n=10+10)
ClearFat32-6    2.14ns ± 0%  1.07ns ± 0%  -50.00%  (p=0.001 n=8+9)
ClearFat40-6    2.67ns ± 1%  1.61ns ± 0%  -39.72%  (p=0.000 n=10+8)
ClearFat48-6    3.75ns ± 0%  2.68ns ± 0%  -28.59%  (p=0.000 n=9+9)
ClearFat56-6    4.29ns ± 0%  3.22ns ± 0%  -25.10%  (p=0.000 n=9+9)
ClearFat64-6    4.30ns ± 0%  3.22ns ± 0%  -25.15%  (p=0.000 n=8+8)
ClearFat128-6   7.50ns ± 1%  7.51ns ± 0%     ~     (p=0.767 n=10+9)
ClearFat256-6   13.9ns ± 1%  13.9ns ± 1%     ~     (p=0.257 n=10+10)
ClearFat512-6   26.8ns ± 0%  26.8ns ± 0%     ~     (p=0.467 n=8+8)
ClearFat1024-6  52.5ns ± 0%  52.5ns ± 0%     ~     (p=1.000 n=8+8)

Also shaves ~20kb from go tool:

go_old 10384994
go_new 10364514 [-20480 bytes]

section differences
global text (code) = -20585 bytes (-0.532047%)
read-only data = -302 bytes (-0.018101%)
Total difference -20887 bytes (-0.348731%)

Change-Id: I15854e87544545c1af24775df895e38e16e12694
Reviewed-on: https://go-review.googlesource.com/54410
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agogo/importer: make source importer more tolerant in presence of errors
griesemer [Tue, 15 Aug 2017 12:44:26 +0000 (14:44 +0200)]
go/importer: make source importer more tolerant in presence of errors

If the source importer only encounters "soft" type checking errors
it can safely return the type-checked package because it will be
completely set up. This makes the source importer slightly more
robust in the presence of errors.

Fixes #20855.

Change-Id: I5af9ccdb30eee6bca7a0fab872f6057bde521bf3
Reviewed-on: https://go-review.googlesource.com/55730
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agoreflect: remove useless parameter from newName
Daniel Martí [Thu, 10 Aug 2017 02:17:59 +0000 (11:17 +0900)]
reflect: remove useless parameter from newName

pkgPath always received the empty string. Worse yet, it panicked if it
received anything else. This has been the case ever since newName was
introduced in early 2016.

Change-Id: I5f164305bd30c34455ef35e776c7616f303b37e4
Reviewed-on: https://go-review.googlesource.com/54331
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agogo/internal/gcimporter: fix typo: cmd/compiler → cmd/compile
Michael Stapelberg [Wed, 16 Aug 2017 12:33:36 +0000 (05:33 -0700)]
go/internal/gcimporter: fix typo: cmd/compiler → cmd/compile

Change-Id: I087980d30308353c4a450636122f7e87c8310090
Reviewed-on: https://go-review.googlesource.com/56090
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomath/big: recognize z.Mul(x, x) as squaring of x
Brian Kessler [Sat, 29 Jul 2017 05:29:30 +0000 (22:29 -0700)]
math/big: recognize z.Mul(x, x) as squaring of x

updates #13745

Multiprecision squaring can be done in a straightforward manner
with about half the multiplications of a basic multiplication
due to the symmetry of the operands.  This change implements
basic squaring for nat types and uses it for Int multiplication
when the same variable is supplied to both arguments of
z.Mul(x, x). This has some overhead to allocate a temporary
variable to hold the cross products, shift them to double and
add them to the diagonal terms.  There is a speed benefit in
the intermediate range when the overhead is neglible and the
asymptotic performance of karatsuba multiplication has not been
reached.

basicSqrThreshold = 20
karatsubaSqrThreshold = 400

Were set by running calibrate_test.go to measure timing differences
between the algorithms.  Benchmarks for squaring:

name           old time/op  new time/op  delta
IntSqr/1-4     51.5ns ±25%  25.1ns ± 7%  -51.38%  (p=0.008 n=5+5)
IntSqr/2-4     79.1ns ± 4%  72.4ns ± 2%   -8.47%  (p=0.008 n=5+5)
IntSqr/3-4      102ns ± 4%    97ns ± 5%     ~     (p=0.056 n=5+5)
IntSqr/5-4      161ns ± 4%   163ns ± 7%     ~     (p=0.952 n=5+5)
IntSqr/8-4      277ns ± 5%   267ns ± 6%     ~     (p=0.087 n=5+5)
IntSqr/10-4     358ns ± 3%   360ns ± 4%     ~     (p=0.730 n=5+5)
IntSqr/20-4    1.07µs ± 3%  1.01µs ± 6%     ~     (p=0.056 n=5+5)
IntSqr/30-4    2.36µs ± 4%  1.72µs ± 2%  -27.03%  (p=0.008 n=5+5)
IntSqr/50-4    5.19µs ± 3%  3.88µs ± 4%  -25.37%  (p=0.008 n=5+5)
IntSqr/80-4    11.3µs ± 4%   8.6µs ± 3%  -23.78%  (p=0.008 n=5+5)
IntSqr/100-4   16.2µs ± 4%  12.8µs ± 3%  -21.49%  (p=0.008 n=5+5)
IntSqr/200-4   50.1µs ± 5%  44.7µs ± 3%  -10.65%  (p=0.008 n=5+5)
IntSqr/300-4    105µs ±11%    95µs ± 3%   -9.50%  (p=0.008 n=5+5)
IntSqr/500-4    231µs ± 5%   227µs ± 2%     ~     (p=0.310 n=5+5)
IntSqr/800-4    496µs ± 9%   459µs ± 3%   -7.40%  (p=0.016 n=5+5)
IntSqr/1000-4   700µs ± 3%   710µs ± 5%     ~     (p=0.841 n=5+5)

Show a speed up of 10-25% in the range where basicSqr is optimal,
improved single word squaring and no significant difference when
the fallback to standard multiplication is used.

Change-Id: Iae2c82ca91cf890823f91e5c83bbe9a2c534b72b
Reviewed-on: https://go-review.googlesource.com/53638
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: make go tool suggest 'go doc cmd/<command>'
Alberto Donizetti [Tue, 15 Aug 2017 21:49:12 +0000 (23:49 +0200)]
cmd/go: make go tool suggest 'go doc cmd/<command>'

$ gotip tool -h says:

  For more about each tool command, see 'go tool command -h'.

But it's better to suggest

  go doc cmd/<command>

Fixes #18313

Change-Id: I0a36d585906a5e1879e5b7927d1b6173e97cb500
Reviewed-on: https://go-review.googlesource.com/55990
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agogo/types: document that Signature.Recv() is ignored for type identity
griesemer [Tue, 15 Aug 2017 11:46:16 +0000 (13:46 +0200)]
go/types: document that Signature.Recv() is ignored for type identity

Fixes #21367.

Change-Id: I50704c5a613abcce57b340db8992c7bcb1cb728f
Reviewed-on: https://go-review.googlesource.com/55710
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agomath/big: speed up GCD x, y calculation
Brian Kessler [Fri, 21 Jul 2017 06:31:51 +0000 (23:31 -0700)]
math/big: speed up GCD x, y calculation

The current implementation of the extended Euclidean GCD algorithm
calculates both cosequences x and y inside the division loop. This
is unneccessary since the second Bezout coefficient can be obtained
at the end of calculation via a multiplication, subtraction and a
division.  In case only one coefficient is needed, e.g. ModInverse
this calculation can be skipped entirely.  This is a standard
optimization, see e.g.

"Handbook of Elliptic and Hyperelliptic Curve Cryptography"
Cohen et al pp 191
Available at:
http://cs.ucsb.edu/~koc/ccs130h/2013/EllipticHyperelliptic-CohenFrey.pdf

Updates #15833

Change-Id: I1e0d2e63567cfed97fd955048fe6373d36f22757
Reviewed-on: https://go-review.googlesource.com/50530
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agomath: eliminate overflow in Pow(x,y) for large y
Brian Kessler [Thu, 13 Jul 2017 05:02:39 +0000 (22:02 -0700)]
math: eliminate overflow in Pow(x,y) for large y

The current implementation uses a shift and add
loop to compute the product of x's exponent xe and
the integer part of y (yi) for yi up to 1<<63.
Since xe is an 11-bit exponent, this product can be
up to 74-bits and overflow both 32 and 64-bit int.

This change checks whether the accumulated exponent
will fit in the 11-bit float exponent of the output
and breaks out of the loop early if overflow is detected.

The current handling of yi >= 1<<63 uses Exp(y * Log(x))
which incorrectly returns Nan for x<0.  In addition,
for y this large, Exp(y * Log(x)) can be enumerated
to only overflow except when x == -1 since the
boundary cases computed exactly:

Pow(NextAfter(1.0, Inf(1)), 1<<63)  == 2.72332... * 10^889
Pow(NextAfter(1.0, Inf(-1)), 1<<63) == 1.91624... * 10^-445

exceed the range of float64. So, the call can be
replaced with a simple case statement analgous to
y == Inf that correctly handles x < 0 as well.

Fixes #7394

Change-Id: I6f50dc951f3693697f9669697599860604323102
Reviewed-on: https://go-review.googlesource.com/48290
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/link: delete shNames
Alex Brainman [Mon, 5 Jun 2017 04:35:46 +0000 (14:35 +1000)]
cmd/link: delete shNames

Change-Id: Ie5d12ba4105fec17551637d066d0dffd508f74a4
Reviewed-on: https://go-review.googlesource.com/55261
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: delete addpesection
Alex Brainman [Mon, 5 Jun 2017 04:29:14 +0000 (14:29 +1000)]
cmd/link: delete addpesection

Change-Id: Iee9db172d28d4d372fa617907078a494e764bf12
Reviewed-on: https://go-review.googlesource.com/55260
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: use peSection everywhere
Alex Brainman [Mon, 5 Jun 2017 04:21:35 +0000 (14:21 +1000)]
cmd/link: use peSection everywhere

Change-Id: I4d4e8452b9b9e628f3ea8b2b727ad63ec2a1dd31
Reviewed-on: https://go-review.googlesource.com/55259
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: add peSection
Alex Brainman [Fri, 2 Jun 2017 06:53:22 +0000 (16:53 +1000)]
cmd/link: add peSection

Change-Id: Id3aeeaeaacf5f079fb2ddad579f2f209b7fc0e06
Reviewed-on: https://go-review.googlesource.com/55258
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: introduce and use peFile and peStringTable
Alex Brainman [Thu, 1 Jun 2017 07:33:15 +0000 (17:33 +1000)]
cmd/link: introduce and use peFile and peStringTable

Change-Id: Icd13b32d35cde474c9292227471f916a64af88eb
Reviewed-on: https://go-review.googlesource.com/55257
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoarchive/tar: make Writer error handling consistent
Joe Tsai [Tue, 15 Aug 2017 01:38:46 +0000 (18:38 -0700)]
archive/tar: make Writer error handling consistent

The Writer logic was not consistent about when an IO error would
persist across multiple calls on Writer's methods.

Thus, to make the error handling more consistent we always check
the persistent state of the error prior to every exported method
call, and return an error if set. Otherwise, it is the responsibility
of every exported method to persist any fatal errors that may occur.

As a simplification, we can remove the close field since that
information can be represented by simply storing ErrWriteAfterClose
in the err field.

Change-Id: I8746ca36b3739803e0373253450db69b3bd12f38
Reviewed-on: https://go-review.googlesource.com/55590
Run-TryBot: Joe Tsai <joetsai@digital-static.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoarchive/tar: add support for long binary strings in GNU format
Joe Tsai [Mon, 14 Aug 2017 23:44:15 +0000 (16:44 -0700)]
archive/tar: add support for long binary strings in GNU format

The GNU tar format defines the following type flags:
TypeGNULongName = 'L' // Next file has a long name
TypeGNULongLink = 'K' // Next file symlinks to a file w/ a long name

Anytime a string exceeds the field dedicated to store it, the GNU format
permits a fake "file" to be prepended where that file entry has a Typeflag
of 'L' or 'K' and the contents of the file is a NUL-terminated string.

Contrary to previous TODO comments,
the GNU format supports arbitrary strings (without NUL) rather UTF-8 strings.
The manual says the following:
<<<
The name, linkname, magic, uname, and gname are
null-terminated character strings
>>>
<<<
All characters in header blocks are represented
by using 8-bit characters in the local variant of ASCII.
>>>

From this description, we gather the following:
* We must forbid NULs in any GNU strings
* Any 8-bit value (other than NUL) is permitted

Since the modern world has moved to UTF-8, it is really difficult to
determine what a "local variant of ASCII" means. For this reason,
we treat strings as just an arbitrary binary string (without NUL)
and leave it to the user to determine the encoding of this string.
(Practically, it seems that UTF-8 is the typical encoding used
in GNU archives seen in the wild).

The implementation of GNU tar seems to confirm this interpretation
of the manual where it permits any arbitrary binary string to exist
within these fields so long as they do not contain the NUL character.

 $ touch `echo -e "not\x80\x81\x82\x83utf8"`
 $ gnutar -H gnu --tar -cvf gnu-not-utf8.tar $(echo -e "not\x80\x81\x82\x83utf8")

The fact that we permit arbitrary binary in GNU strings goes
hand-in-hand with the fact that GNU also permits a "base-256" encoding
of numeric fields, which is effectively two-complement binary.

Change-Id: Ic037ec6bed306d07d1312f0058594bd9b64d9880
Reviewed-on: https://go-review.googlesource.com/55573
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agonet/http: delete unnecessary string concatenation
Naoki Kanatani [Thu, 20 Jul 2017 15:35:31 +0000 (00:35 +0900)]
net/http: delete unnecessary string concatenation

In the existing implementation, if pattern is an empty string,
program calls a panic with the message which is a concatenation of
"http: invalid pattern " and pattern.
In this case, pattern is an empty, so the commit removes
this concatenation and the trailing space.

Fixes: #21102
Change-Id: I49f58b52d835311a6ac642de871eb15646e48a54
Reviewed-on: https://go-review.googlesource.com/50350
Reviewed-by: Dmitri Shuralyov <shurcool@gmail.com>
Reviewed-by: Tom Bergan <tombergan@google.com>
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotesting: don't fail all tests after racy test failure
Ian Lance Taylor [Tue, 8 Aug 2017 22:50:43 +0000 (15:50 -0700)]
testing: don't fail all tests after racy test failure

The code was adding race.Errors to t.raceErrors before checking
Failed, but Failed was using t.raceErrors+race.Errors. We don't want
to change Failed, since that would affect tests themselves, so modify
the harness to not unnecessarily change t.raceErrors.

Updates #19851
Fixes #21338
Change-Id: I7bfdf281f90e045146c92444f1370d55c45221d4
Reviewed-on: https://go-review.googlesource.com/54050
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: collapse runs of string constants in walkprint
Josh Bleecher Snyder [Mon, 22 May 2017 13:40:06 +0000 (06:40 -0700)]
cmd/compile: collapse runs of string constants in walkprint

This reduces the code footprint of code like:

println("foo=", foo, "bar=", bar)

which is fairly common in the runtime.

Prior to this change, this makes function calls to print each of:

"foo=", " ", foo, " ", "bar=", " ", bar, "\n"

After this change, this prints:

"foo= ", foo, " bar= ", bar, "\n"

This shrinks the hello world binary by 0.4%.
More importantly, this improves the instruction
density of important runtime routines.

Change-Id: I8971bdf5382fbaaf4a82bad4442f9da07c28d395
Reviewed-on: https://go-review.googlesource.com/55098
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: insert ' ' and \n at beginning of walkprint
Josh Bleecher Snyder [Sun, 21 May 2017 22:18:08 +0000 (15:18 -0700)]
cmd/compile: insert ' ' and \n at beginning of walkprint

Rather than emitting spaces and newlines for println
as we walk the expression, construct it all up front.
This enables further optimizations.

This requires using printstring instead of print in
the implementation of printsp and printnl,
on pain of infinite recursion.
That's ok; it's more efficient anyway, and just as simple.
While we're here, do it for other print routines as well.

Change-Id: I61d7df143810e00710c4d4d948d904007a7fd190
Reviewed-on: https://go-review.googlesource.com/55097
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: further simplify walkprint
Josh Bleecher Snyder [Sat, 20 May 2017 00:18:50 +0000 (17:18 -0700)]
cmd/compile: further simplify walkprint

Passes toolstash-check.

Change-Id: I6386a37a35221de8d3944253beee668927810f17
Reviewed-on: https://go-review.googlesource.com/55096
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: simplify walkprint
Josh Bleecher Snyder [Sat, 20 May 2017 00:12:57 +0000 (17:12 -0700)]
cmd/compile: simplify walkprint

Superficial simplifications: reduce variable scope,
eliminate pointless temporaries, use a switch statement.

Passes toolstash-check.

Change-Id: I6450493a0583a6ce8ec0461b66954cf1445a754f
Reviewed-on: https://go-review.googlesource.com/55095
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years ago.gitignore: include only Go project artifiacts
Hana Kim [Tue, 15 Aug 2017 19:22:28 +0000 (15:22 -0400)]
.gitignore: include only Go project artifiacts

This effectively reverts

https://golang.org/cl/53770

and adds a guide on what not to add in this file.

Update #21458

Change-Id: I7c740d492b70628b5d9f9e1622014995a3f6f8ec
Reviewed-on: https://go-review.googlesource.com/55871
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: move selectdone into g
Daniel Morsing [Wed, 2 Aug 2017 18:01:17 +0000 (19:01 +0100)]
runtime: move selectdone into g

Writing to selectdone on the stack of another goroutine meant a
pretty subtle dance between the select code and the stack copying
code. Instead move the selectdone variable into the g struct.

Change-Id: Id246aaf18077c625adef7ca2d62794afef1bdd1b
Reviewed-on: https://go-review.googlesource.com/53390
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: set itab function pointers at compile time
Keith Randall [Tue, 30 May 2017 19:59:25 +0000 (12:59 -0700)]
cmd/compile: set itab function pointers at compile time

I noticed that we don't set an itab's function pointers at compile
time. Instead, we currently do it at executable startup.

Set the function pointers at compile time instead. This shortens
startup time. It has no effect on normal binary size. Object files
will have more relocations, but that isn't a big deal.

For PIE there are additional pointers that will need to be adjusted at
load time. There are already other pointers in an itab that need to be
adjusted, so the cache line will already be paged in. There might be
some binary size overhead to mark these pointers. The "go test -c
-buildmode=pie net/http" binary is 0.18% bigger.

Update #20505

Change-Id: I267c82489915b509ff66e512fc7319b2dd79b8f7
Reviewed-on: https://go-review.googlesource.com/44341
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
7 years agoruntime: make STW duration more accurate
Austin Clements [Mon, 24 Jul 2017 20:06:10 +0000 (16:06 -0400)]
runtime: make STW duration more accurate

Currently, GC captures the start-the-world time stamp after
startTheWorldWithSema returns. This is problematic for two reasons:

1. It's possible to get preempted between startTheWorldWithSema
starting the world and calling nanotime.

2. startTheWorldWithSema does several clean-up tasks after the world
is up and running that on rare occasions can take upwards of 10ms.

Since the runtime uses the start-the-world time stamp to compute the
STW duration, both of these can significantly inflate the reported STW
duration.

Fix this by having startTheWorldWithSema itself call nanotime once the
world is started.

Change-Id: I114630234fb73c9dabae50a2ef1884661f2459db
Reviewed-on: https://go-review.googlesource.com/55410
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoencoding/asn1: handle application tag in Marshal
Hiroshi Ioka [Wed, 24 May 2017 22:48:08 +0000 (07:48 +0900)]
encoding/asn1: handle application tag in Marshal

Fixes #20488

Change-Id: Iae963b612aea3d9e814b08f655e2eb019ece256e
Reviewed-on: https://go-review.googlesource.com/44110
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>

7 years agocrypto/tls: disallow handshake messages fragmented across CCS
Filippo Valsorda [Wed, 18 Jan 2017 16:53:35 +0000 (16:53 +0000)]
crypto/tls: disallow handshake messages fragmented across CCS

Detected by BoGo test FragmentAcrossChangeCipherSpec-Server-Packed.

Change-Id: I9a76697b9cdeb010642766041971de5c7e533481
Reviewed-on: https://go-review.googlesource.com/48811
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>

7 years agocrypto/x509: add PublicKeyAlgorithm.String()
Martins Sipenko [Fri, 26 May 2017 11:35:29 +0000 (14:35 +0300)]
crypto/x509: add PublicKeyAlgorithm.String()

Change-Id: I77d9c77875519d77bac49cc8870c2e0c4563fe55
Reviewed-on: https://go-review.googlesource.com/44313
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>

7 years agocrypto/tls: add BenchmarkHandshakeServer
Filippo Valsorda [Fri, 2 Jun 2017 19:33:50 +0000 (12:33 -0700)]
crypto/tls: add BenchmarkHandshakeServer

name                                       time/op
HandshakeServer/RSA-4                      1.10ms ± 0%
HandshakeServer/ECDHE-P256-RSA-4           1.23ms ± 1%
HandshakeServer/ECDHE-P256-ECDSA-P256-4     178µs ± 1%
HandshakeServer/ECDHE-X25519-ECDSA-P256-4   180µs ± 2%
HandshakeServer/ECDHE-P521-ECDSA-P521-4    19.8ms ± 1%

Change-Id: I6b2c79392995d259cfdfc5199be44cc7cc40e155
Reviewed-on: https://go-review.googlesource.com/44730
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>

7 years agocrypto/tls: don't check whether an ec point is on a curve twice
Andreas Auernhammer [Fri, 26 May 2017 09:33:49 +0000 (11:33 +0200)]
crypto/tls: don't check whether an ec point is on a curve twice

The processClientKeyExchange and processServerKeyExchange functions unmarshal an
encoded EC point and explicitly check whether the point is on the curve. The explicit
check can be omitted because elliptic.Unmarshal fails if the point is not on the curve
and the returned error would always be the same.

Fixes #20496

Change-Id: I5231a655eace79acee2737dd036a0c255ed42dbb
Reviewed-on: https://go-review.googlesource.com/44311
Reviewed-by: Adam Langley <agl@golang.org>
Reviewed-by: Avelino <t@avelino.xxx>
Run-TryBot: Adam Langley <agl@golang.org>

7 years agocrypto/x509: add MarshalPKCS8PrivateKey
James Hartig [Fri, 14 Apr 2017 04:53:09 +0000 (00:53 -0400)]
crypto/x509: add MarshalPKCS8PrivateKey

asn1.NullRawValue was used according to RFC 3279 2.2.1. Without this tag,
the output didn't match openssl.

Fixes #19972

Change-Id: Ia52ddb810888837f913dbd65c4e1328f6c8084bb
Reviewed-on: https://go-review.googlesource.com/40730
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

7 years agoruntime: fix getclosureptr doc
Austin Clements [Tue, 15 Aug 2017 15:32:02 +0000 (11:32 -0400)]
runtime: fix getclosureptr doc

Change-Id: I1b42fca2107b06e6fc95728f7bf3d08d005c4cb4
Reviewed-on: https://go-review.googlesource.com/55810
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agosyscall: really use utimensat for UtimesNano on Solaris
Tobias Klauser [Mon, 14 Aug 2017 07:47:58 +0000 (09:47 +0200)]
syscall: really use utimensat for UtimesNano on Solaris

golang.org/cl/55130 added utimensat for Solaris but didn't use it in
UtimesNano (despite indicating otherwise in the commit message). Fix
this by also using utimensat for UtimesNano on Solaris.

Because all versions of Solaris suppported by Go support utimensat,
there is no need for the fallback logic and utimensat can be called
unconditionally.

This issue was pointed out by Shawn Walker-Salas.

Updates #16480

Change-Id: I114338113a6da3cfcb8bca950674bdc8f5a7a9e5
Reviewed-on: https://go-review.googlesource.com/55141
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years ago.gitignore: exclude /.idea/
Hana (Hyang-Ah) Kim [Fri, 4 Aug 2017 21:38:46 +0000 (17:38 -0400)]
.gitignore: exclude /.idea/

Some IDEs (e.g. intellij IDE based ones) create the .idea folder
to store project specific settings. This is irrelevant to Go project
that does not assume any specific IDEs, but interferes with git.

Change-Id: I0c93d9a3f7edff095fbe0c7a53b06c92b391c970
Reviewed-on: https://go-review.googlesource.com/53770
Reviewed-by: Ian Lance Taylor <iant@golang.org>