]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years agoruntime/cgo: save callee-saved xmm registers on windows/amd64
Shenghou Ma [Thu, 24 Mar 2016 23:12:44 +0000 (19:12 -0400)]
runtime/cgo: save callee-saved xmm registers on windows/amd64

For #14876.

Change-Id: I33947f74e8058437a784862f1f064974afc99250
Reviewed-on: https://go-review.googlesource.com/21084
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocompress/gzip: clean up zero-mtimes test
Caleb Spare [Fri, 25 Mar 2016 17:06:19 +0000 (10:06 -0700)]
compress/gzip: clean up zero-mtimes test

- Fix a typo.
- Skip this test on -short on non-builders.

Change-Id: Id102eceb59451694bf92b618e02ccee6603b6852
Reviewed-on: https://go-review.googlesource.com/21113
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agocrypto/tls: Update references to draft-ietf-tls-downgrade-scsv-00.
David Benjamin [Mon, 15 Feb 2016 16:56:18 +0000 (11:56 -0500)]
crypto/tls: Update references to draft-ietf-tls-downgrade-scsv-00.

It's RFC 7507 now.

Change-Id: Iccd6c65f9d4b1f4d17ee068dee4576a512ba8405
Reviewed-on: https://go-review.googlesource.com/21154
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoencoding/asn1: tags should be encoded in minimal form.
David Benjamin [Tue, 5 Jan 2016 00:11:02 +0000 (16:11 -0800)]
encoding/asn1: tags should be encoded in minimal form.

High tag number form may not be used for tag numbers that fit in low tag number
form.

Change-Id: I93edde0e1f86087047e0b3f2e55d6180b01e78bf
Reviewed-on: https://go-review.googlesource.com/18224
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

9 years agoruntime: fix signal handling on Solaris
Joe Sylve [Fri, 25 Mar 2016 18:51:56 +0000 (13:51 -0500)]
runtime: fix signal handling on Solaris

This fixes the problems with signal handling that were inadvertently
introduced in https://go-review.googlesource.com/21006.

Fixes #14899

Change-Id: Ia746914dcb3146a52413d32c57b089af763f0810
Reviewed-on: https://go-review.googlesource.com/21145
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoreflect: test name data is aligned
David Crawshaw [Fri, 25 Mar 2016 18:28:15 +0000 (14:28 -0400)]
reflect: test name data is aligned

For #14962.

Change-Id: I3539d882487c99dee99ac953e039b79c6b963cf9
Reviewed-on: https://go-review.googlesource.com/21150
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/go: stat the archive file only when executing the commands
Mohit Agarwal [Thu, 24 Mar 2016 17:16:10 +0000 (22:46 +0530)]
cmd/go: stat the archive file only when executing the commands

Fixes #14944

Change-Id: I73e0997cb6ebaeced1045b0ddadac893319bd78f
Reviewed-on: https://go-review.googlesource.com/21065
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

9 years agoreflect: name interface type that pins method info
David Crawshaw [Fri, 25 Mar 2016 15:03:47 +0000 (11:03 -0400)]
reflect: name interface type that pins method info

I recently added TestUnexportedMethods which uses an interface type
to pin type information for an unexported method. But as written,
the interface type is not accessible to the reflect package.

You can imagine a future compiler optimization realizing that and
removing the type information for f. In fact, cl/20901 happens to
do that.

Change-Id: I1ddb67f50cb9b5737253b58f10545f3de652c29d
Reviewed-on: https://go-review.googlesource.com/21112
Reviewed-by: Michel Lespinasse <walken@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: speed up growslice by avoiding divisions 2
Marvin Stenger [Thu, 24 Mar 2016 16:42:01 +0000 (17:42 +0100)]
runtime: speed up growslice by avoiding divisions 2

This is a follow-up of https://go-review.googlesource.com/#/c/20653/

Special case computation for slices with elements of byte size or
pointer size.

name                      old time/op  new time/op  delta
GrowSliceBytes-4          86.2ns ± 3%  75.4ns ± 2%  -12.50%  (p=0.000 n=20+20)
GrowSliceInts-4            161ns ± 3%   136ns ± 3%  -15.59%  (p=0.000 n=19+19)
GrowSlicePtr-4             239ns ± 2%   233ns ± 2%   -2.52%  (p=0.000 n=20+20)
GrowSliceStruct24Bytes-4   258ns ± 3%   256ns ± 3%     ~     (p=0.134 n=20+20)

Change-Id: Ice5fa648058fe9d7fa89dee97ca359966f671128
Reviewed-on: https://go-review.googlesource.com/21101
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomisc/ios: fix exec wrapper locking
Elias Naur [Fri, 25 Mar 2016 14:40:44 +0000 (15:40 +0100)]
misc/ios: fix exec wrapper locking

The exec wrapper lock file was opened, locked and then never used
again, assuming it would close and unlock at process exit.
However, the garbage collector could collect and run the *os.File
finalizer that closes the file prematurely, rendering the lock
ineffective.

Make the lock global so that the lock is live during the entire
execution.

(Hopefully) fix the iOS builders.

Change-Id: I62429e92042a0a49c4f1ea553fdb32b6ea53a43e
Reviewed-on: https://go-review.googlesource.com/21137
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agocmd/compile: avoid pointers in go.string.* symbols
David Crawshaw [Fri, 25 Mar 2016 02:32:38 +0000 (22:32 -0400)]
cmd/compile: avoid pointers in go.string.* symbols

When creating binaries for dynamic linking, the linker moves
read-only data symbols that contain pointers into relro sections.
It is not setup for handling a go.string symbol moving to relro.

Instead of teaching it how (because go.string symbols with pointers
are unusual anyhow), put the data in a type.. section.

Fixes the android builder.

Change-Id: Ica4722d32241643c060923517b90276ff8ac6b07
Reviewed-on: https://go-review.googlesource.com/21110
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile/ssa: generate less garbage in schedule
Josh Bleecher Snyder [Fri, 25 Mar 2016 14:33:39 +0000 (07:33 -0700)]
cmd/compile/ssa: generate less garbage in schedule

Passes toolstash -cmp.

name       old alloc/op    new alloc/op    delta
Template      58.5MB ± 0%     57.8MB ± 0%  -1.15%        (p=0.000 n=10+10)
Unicode       41.3MB ± 0%     41.2MB ± 0%  -0.17%        (p=0.000 n=10+10)
GoTypes        196MB ± 0%      193MB ± 0%  -1.26%        (p=0.000 n=10+10)
Compiler       863MB ± 0%      850MB ± 0%  -1.49%        (p=0.000 n=10+10)

name       old allocs/op   new allocs/op   delta
Template        522k ± 0%       507k ± 0%  -2.99%        (p=0.000 n=10+10)
Unicode         403k ± 0%       401k ± 0%  -0.42%        (p=0.000 n=10+10)
GoTypes        1.58M ± 0%      1.52M ± 0%  -3.61%        (p=0.000 n=10+10)
Compiler       6.47M ± 0%      6.17M ± 0%  -4.62%        (p=0.000 n=10+10)

Change-Id: Ia7a6242e8d226b41966c344d253814dcce6424a8
Reviewed-on: https://go-review.googlesource.com/21141
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years agoruntime: avoid fork/exit race in plan9
Richard Miller [Fri, 25 Mar 2016 12:50:35 +0000 (12:50 +0000)]
runtime: avoid fork/exit race in plan9

There's a race between runtime.goexitsall killing all OS processes
of a go program in order to exit, and runtime.newosproc forking a
new one.  If the new process has been created but not yet stored
its pid in m.procid, it will not be killed by goexitsall and
deadlock results.

This CL prevents the race by making the newly forked process
check whether the program is exiting.  It also prevents a
potential "shoot-out" if multiple goroutines call Exit at
the same time, which could possibly lead to two processes
killing each other and leaving the rest deadlocked.

Change-Id: I3170b4a62d2461f6b029b3d6aad70373714ed53e
Reviewed-on: https://go-review.googlesource.com/21135
Run-TryBot: David du Colombier <0intro@gmail.com>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David du Colombier <0intro@gmail.com>
9 years agoruntime: improve randomized stealing logic
Dmitry Vyukov [Fri, 18 Mar 2016 11:52:52 +0000 (12:52 +0100)]
runtime: improve randomized stealing logic

During random stealing we steal 4*GOMAXPROCS times from random procs.
One would expect that most of the time we check all procs this way,
but due to low quality PRNG we actually miss procs with frightening
probability. Below are modelling experiment results for 1e6 tries:

GOMAXPROCS = 2 : missed 1 procs 7944 times

GOMAXPROCS = 3 : missed 1 procs 101620 times
GOMAXPROCS = 3 : missed 2 procs 3571 times

GOMAXPROCS = 4 : missed 1 procs 63916 times
GOMAXPROCS = 4 : missed 2 procs 61 times
GOMAXPROCS = 4 : missed 3 procs 16 times

GOMAXPROCS = 5 : missed 1 procs 133136 times
GOMAXPROCS = 5 : missed 2 procs 1025 times
GOMAXPROCS = 5 : missed 3 procs 101 times
GOMAXPROCS = 5 : missed 4 procs 15 times

GOMAXPROCS = 8 : missed 1 procs 151765 times
GOMAXPROCS = 8 : missed 2 procs 5057 times
GOMAXPROCS = 8 : missed 3 procs 1726 times
GOMAXPROCS = 8 : missed 4 procs 68 times

GOMAXPROCS = 12 : missed 1 procs 199081 times
GOMAXPROCS = 12 : missed 2 procs 27489 times
GOMAXPROCS = 12 : missed 3 procs 3113 times
GOMAXPROCS = 12 : missed 4 procs 233 times
GOMAXPROCS = 12 : missed 5 procs 9 times

GOMAXPROCS = 16 : missed 1 procs 237477 times
GOMAXPROCS = 16 : missed 2 procs 30037 times
GOMAXPROCS = 16 : missed 3 procs 9466 times
GOMAXPROCS = 16 : missed 4 procs 1334 times
GOMAXPROCS = 16 : missed 5 procs 192 times
GOMAXPROCS = 16 : missed 6 procs 5 times
GOMAXPROCS = 16 : missed 7 procs 1 times
GOMAXPROCS = 16 : missed 8 procs 1 times

A missed proc won't lead to underutilization because we check all procs
again after dropping P. But it can lead to an unpleasant situation
when we miss a proc, drop P, check all procs, discover work, acquire P,
miss the proc again, repeat.

Improve stealing logic to cover all procs.
Also don't enter spinning mode and try to steal when there is nobody around.

Change-Id: Ibb6b122cc7fb836991bad7d0639b77c807aab4c2
Reviewed-on: https://go-review.googlesource.com/20836
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
9 years agocompress/gzip: skip mtime test when GOROOT doesn't exist
Elias Naur [Fri, 25 Mar 2016 10:14:29 +0000 (11:14 +0100)]
compress/gzip: skip mtime test when GOROOT doesn't exist

Fixes the iOS builders

Change-Id: I5097ca19048381ffb5a4c5ea038b7c4aa18ee4b7
Reviewed-on: https://go-review.googlesource.com/21132
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/go: remove double space in template
Marcel van Lohuizen [Fri, 25 Mar 2016 09:16:40 +0000 (10:16 +0100)]
cmd/go: remove double space in template

Change-Id: I6113145baa727b9fd103765f74dc5d7af86dfdf8
Reviewed-on: https://go-review.googlesource.com/21131
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agotesting: probe with N=1
Marcel van Lohuizen [Wed, 23 Mar 2016 20:24:22 +0000 (21:24 +0100)]
testing: probe with N=1

Change control flow to probe with N=1. This calls benchFunc
the same number of times as the old implementation in the
absence of subbenchmarks.

To be compatible with existing tools, benchmarking only
prints a line for "leaf" benchmarks. This means, though, that
the name of a benchmark can only be printed after the first
iteration.

Issue #14863

Change-Id: Ic7b9b89b058f8ebb5287755f24f9e47df8c9537c
Reviewed-on: https://go-review.googlesource.com/21043
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/go: update alldocs.go
Marcel van Lohuizen [Wed, 23 Mar 2016 16:22:09 +0000 (17:22 +0100)]
cmd/go: update alldocs.go

Also added go:generate directive.

Change-Id: Ib1f0eddc75e3c47a4d904786a29b964a35b18456
Reviewed-on: https://go-review.googlesource.com/21042
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>

9 years agoall: delete dead non-test code
Dominik Honnef [Sun, 20 Mar 2016 23:12:18 +0000 (00:12 +0100)]
all: delete dead non-test code

This change removes a lot of dead code. Some of the code has never been
used, not even when it was first commited. The rest shouldn't have
survived refactors.

This change doesn't remove unused routines helpful for debugging, nor
does it remove code that's used in commented out blocks of code that are
only unused temporarily. Furthermore, unused constants weren't removed
when they were part of a set of constants from specifications.

One noteworthy omission from this CL are about 1000 lines of unused code
in cmd/fix, 700 lines of which are the typechecker, which hasn't been
used ever since the pre-Go 1 fixes have been removed. I wasn't sure if
this code should stick around for future uses of cmd/fix or be culled as
well.

Change-Id: Ib714bc7e487edc11ad23ba1c3222d1fd02e4a549
Reviewed-on: https://go-review.googlesource.com/20926
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocrypto/hmac: simplify implementation
Dmitry Chestnykh [Tue, 22 Mar 2016 15:41:11 +0000 (16:41 +0100)]
crypto/hmac: simplify implementation

Store already padded keys instead of storing key and padding it during
Reset and Sum. This simplifies code and makes Reset-Write-Sum sequences
faster, which helps /x/crypto/pbkdf2.

HMAC benchmark:

benchmark                    old ns/op     new ns/op     delta
BenchmarkHMACSHA256_1K-4     7669          7613          -0.73%
BenchmarkHMACSHA256_32-4     1880          1737          -7.61%

benchmark                    old MB/s     new MB/s     speedup
BenchmarkHMACSHA256_1K-4     133.52       134.50       1.01x
BenchmarkHMACSHA256_32-4     17.02        18.41        1.08x

PBKDF2 benchmark:

benchmark                       old ns/op     new ns/op     delta
BenchmarkPBKDF2HMACSHA256-4     1943196       1807699       -6.97%

Change-Id: I6697028370c226715ab477b0844951a83eb3488c
Reviewed-on: https://go-review.googlesource.com/21024
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
9 years agoreflect: add method StructTag.Lookup
John Jeffery [Sun, 20 Mar 2016 10:24:12 +0000 (20:24 +1000)]
reflect: add method StructTag.Lookup

The Lookup method provides a way to extract a tag value, while
determining whether the tag key exists in the struct field's tag.

Fixes #14883

Change-Id: I7460cb68f0ca1aaa025935050b9e182efcb64db3
Reviewed-on: https://go-review.googlesource.com/20864
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoall: zero mtimes in testdata gz files
Caleb Spare [Thu, 24 Mar 2016 00:33:32 +0000 (17:33 -0700)]
all: zero mtimes in testdata gz files

Fixes #14937.

Change-Id: Iea11a32230d44d5a43f8aec812d25f13bce85895
Reviewed-on: https://go-review.googlesource.com/21038
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoencoding/json: add (*Encoder).Indent
Caleb Spare [Thu, 24 Mar 2016 06:14:35 +0000 (23:14 -0700)]
encoding/json: add (*Encoder).Indent

Fixes #6492.

Change-Id: Ibc633c43a6d134bb140addb59780a5758b35a5c5
Reviewed-on: https://go-review.googlesource.com/21057
Run-TryBot: Caleb Spare <cespare@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/go: remove limits on parallel tasks in the Go command for iOS
Elias Naur [Thu, 24 Mar 2016 19:45:54 +0000 (20:45 +0100)]
cmd/go: remove limits on parallel tasks in the Go command for iOS

Ther darwin/arm{,64} exec wrapper now limits the number of concurrent
executions to 1, so remove the higher level parallel task limit from
the Go command.

Change-Id: Id84f65c3908305bde0452b3c8db6df8c5a8881bb
Reviewed-on: https://go-review.googlesource.com/21100
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agomisc/cgo/testcarchive: fix build
Elias Naur [Fri, 25 Mar 2016 00:27:34 +0000 (01:27 +0100)]
misc/cgo/testcarchive: fix build

I failed to rebase (and re-test) CL 21102 before submit, which meant
that two extra tests sneaked into testcarchive that still referenced
runtime.GOOS and runtime.GOARCH.

Convert the new tests.

While we're here, make sure pending tasks are flushed before running
the host tests. If not, the "##### misc/cgo/testcarchive" banner
and "PASS" won't show up in the all.bash output.

Change-Id: I41fc4ec9515f9a193fa052f7c31fac452153c897
Reviewed-on: https://go-review.googlesource.com/21106
Run-TryBot: Elias Naur <elias.naur@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile, runtime: new static name encoding
David Crawshaw [Mon, 21 Mar 2016 17:21:55 +0000 (13:21 -0400)]
cmd/compile, runtime: new static name encoding

Create a byte encoding designed for static Go names.

It is intended to be a compact representation of a name
and optional tag data that can be turned into a Go string
without allocating, and describes whether or not it is
exported without unicode table.

The encoding is described in reflect/type.go:

// The first byte is a bit field containing:
//
// 1<<0 the name is exported
// 1<<1 tag data follows the name
// 1<<2 pkgPath *string follow the name and tag
//
// The next two bytes are the data length:
//
//  l := uint16(data[1])<<8 | uint16(data[2])
//
// Bytes [3:3+l] are the string data.
//
// If tag data follows then bytes 3+l and 3+l+1 are the tag length,
// with the data following.
//
// If the import path follows, then ptrSize bytes at the end of
// the data form a *string. The import path is only set for concrete
// methods that are defined in a different package than their type.

Shrinks binary sizes:

cmd/go: 164KB (1.6%)
jujud:  1.0MB (1.5%)

For #6853.

Change-Id: I46b6591015b17936a443c9efb5009de8dfe8b609
Reviewed-on: https://go-review.googlesource.com/20968
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agomisc/cgo/testcarchive: re-enable c-archive test on iOS and Android
Elias Naur [Thu, 24 Mar 2016 20:47:02 +0000 (21:47 +0100)]
misc/cgo/testcarchive: re-enable c-archive test on iOS and Android

The c-archive test were recently converted from shell script to Go.
Unfortunately, it also lost the ability to target iOS and Android
that lack C compilers and require exec wrappers.

Compile the c-archive test for the host and run it with the target
GOOS/GOARCH environment. Change the test to rely on go env GOOS
and go env GOARCH instead of runtime.GOOS and runtime.GOARCH.

Fixes #8345

Change-Id: I290ace2f7e96b87c55d99492feb7d660140dcb32
Reviewed-on: https://go-review.googlesource.com/21102
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agovendor: move golang.org/x/net/http2/hpack back to vendor
Brad Fitzpatrick [Thu, 24 Mar 2016 05:49:47 +0000 (16:49 +1100)]
vendor: move golang.org/x/net/http2/hpack back to vendor

Updates #14047

Change-Id: I7e314e2c7e3e8da18ab023729740fbc9ea3f661e
Reviewed-on: https://go-review.googlesource.com/21063
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/compile: simplify shifts when the counter fits 6 bits.
Alexandru Moșoi [Thu, 24 Mar 2016 12:40:11 +0000 (13:40 +0100)]
cmd/compile: simplify shifts when the counter fits 6 bits.

In f the extra & 63 is redundant because SHRQ already
looks at the bottom 6 bits only. This is a trick on AMD64
to get rid of CMPQ/SBBQ/ANDQ if one knows that the shift
counter is small.

func f(x uint64, s uint) uint64 {
        return x >> (s & 63)
}

Change-Id: I4861c902168dabec9a6a14a85750246dde94fc08
Reviewed-on: https://go-review.googlesource.com/21073
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: fold more of CMPQ and ANDQ
Alexandru Moșoi [Thu, 24 Mar 2016 07:48:41 +0000 (08:48 +0100)]
cmd/compile: fold more of CMPQ and ANDQ

g used to produce CMPQ/SBBQ/ANDQ, but f didn't even though
s&15 is at most s&63.

func f(x uint64, s uint) uint64 {
        return x >> (s & 63)
}
func g(x uint64, s uint) uint64 {
        return x >> (s & 15)
}

Change-Id: Iab4a1a6e10b471dead9f1203e9d894677cf07bb2
Reviewed-on: https://go-review.googlesource.com/21048
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: improve last ditch signal forwarding for Unix libraries
Joe Sylve [Wed, 23 Mar 2016 03:11:42 +0000 (22:11 -0500)]
runtime: improve last ditch signal forwarding for Unix libraries

The current runtime attempts to forward signals generated by non-Go
code to the original signal handler.  If it can't call the original
handler directly, it currently attempts to re-raise the signal after
resetting the handler.  In this case, the original context is lost.

This fix prevents that problem by simply returning from the go signal
handler after resetting the original handler.  It only does this when
the original handler is the system default handler, which in all cases
is known to not recover.  The signal is not reset, so it is retriggered
and the original handler takes over with the proper context.

Fixes #14899

Change-Id: Ib1c19dfa4b50d9732d7a453de3784c8141e1cbb3
Reviewed-on: https://go-review.googlesource.com/21006
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoos: skip TestStatStdin on Android
Elias Naur [Thu, 24 Mar 2016 18:55:40 +0000 (19:55 +0100)]
os: skip TestStatStdin on Android

Android doesn't (generally) have /bin/sh.

Change-Id: I343817c342e3473d09c85155761682b5ddb043e4
Reviewed-on: https://go-review.googlesource.com/21075
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: fix inconsistency in slice.go
Marvin Stenger [Thu, 24 Mar 2016 00:55:42 +0000 (01:55 +0100)]
runtime: fix inconsistency in slice.go

Fixes #14938.

Additionally some simplifications along the way.

Change-Id: I2c5fb7e32dcc6fab68fff36a49cb72e715756abe
Reviewed-on: https://go-review.googlesource.com/21046
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

9 years agomisc/ios: serialize iOS execution
Elias Naur [Thu, 24 Mar 2016 15:03:07 +0000 (16:03 +0100)]
misc/ios: serialize iOS execution

The iOS exec wrapper use complicated machinery to run a iOS binary
on a device.
Running several binaries concurrently doesn't work (reliably), which
can break tests running concurrently. For my setup, the
runtime:cpu124 and sync_cpu tests can't run reliably without one of them
crashing.

Add a file lock to the exec wrapper to serialize execution.

Fixes #14318 (for me)

Change-Id: I023610e014b327f8d66f1d2fd2e54dd0e56f2be0
Reviewed-on: https://go-review.googlesource.com/21074
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agocmd/compile: avoid allocating a Nodes for readonly method receivers
Josh Bleecher Snyder [Wed, 23 Mar 2016 22:27:23 +0000 (15:27 -0700)]
cmd/compile: avoid allocating a Nodes for readonly method receivers

We were allocating a Nodes for common method calls
that did not modify the Nodes.
Though there is no clear wall time impact,
this significantly reduces the number of allocations,
so it seems worth doing.

Passes toolstash -cmp.

name       old alloc/op    new alloc/op    delta
Template      59.0MB ± 0%     58.6MB ± 0%   -0.81%        (p=0.000 n=25+25)
Unicode       41.4MB ± 0%     41.3MB ± 0%   -0.18%        (p=0.000 n=25+25)
GoTypes        198MB ± 0%      197MB ± 0%   -0.80%        (p=0.000 n=24+25)
Compiler       875MB ± 0%      865MB ± 0%   -1.09%        (p=0.000 n=25+25)

name       old allocs/op   new allocs/op   delta
Template        581k ± 0%       520k ± 0%  -10.42%        (p=0.000 n=25+25)
Unicode         413k ± 0%       403k ± 0%   -2.30%        (p=0.000 n=25+25)
GoTypes        1.78M ± 0%      1.58M ± 0%  -11.18%        (p=0.000 n=25+25)
Compiler       7.66M ± 0%      6.47M ± 0%  -15.51%        (p=0.000 n=25+25)

Change-Id: I012a9f4b333821bdf61b4f2bdff4ce5c3b5d3057
Reviewed-on: https://go-review.googlesource.com/21056
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/dist: skip testcarchive test in Android and iOS
Elias Naur [Thu, 24 Mar 2016 09:16:46 +0000 (10:16 +0100)]
cmd/dist: skip testcarchive test in Android and iOS

CL 20892 converted the misc/cgo/testcarchive test to Go.
Unfortunately, dist does not (yet) support tests running off the host
so the testcarchive is disabled for now.

For #14318

Change-Id: Iab3d0a7b5309187a603b48f22a7fa736f089f89d
Reviewed-on: https://go-review.googlesource.com/21070
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agomisc/ios: deflake tests on darwin/arm{,arm64}
Elias Naur [Wed, 23 Mar 2016 15:17:44 +0000 (16:17 +0100)]
misc/ios: deflake tests on darwin/arm{,arm64}

A retry mechanism is in place to combat the inherent flakiness of
launching iOS test binaries. Before it covered just the starting of
lldb; expand it to cover the setup steps as well. Note that the
running of the binary itself is (still) not retried, to avoid
covering over genuine bugs.

On my test device (iPhone 5S, iOS 9.3) starting lldb can take longer
than 10 seconds, so increase the timeout for that.
Furthermore, some basic steps such as setting breakpoints in lldb
can take longer than the 1 second timeout. Increase that timeout
as well, to 2 seconds.

Finally, improve the error message for when ios-deploy is not
installed.

For #14318

Change-Id: Iba41d1bd9d023575b9454cb577b08f8cae081c2a
Reviewed-on: https://go-review.googlesource.com/21072
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agomisc/ios: pass through SIGCONT in the exec wrapper
Elias Naur [Thu, 24 Mar 2016 09:40:34 +0000 (10:40 +0100)]
misc/ios: pass through SIGCONT in the exec wrapper

Instruct lldb to pass through SIGCONT unhindered when running iOS
tests. Fixes the TestSIGCONT test in os/signal.

For #14318

Change-Id: I669264208cc3d6ecae9fbc8790e0b753a93a5e04
Reviewed-on: https://go-review.googlesource.com/21071
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agoruntime/cgo: block signals to the iOS mach exception handler
Elias Naur [Thu, 24 Mar 2016 07:39:54 +0000 (08:39 +0100)]
runtime/cgo: block signals to the iOS mach exception handler

For darwin/arm{,64} a non-Go thread is created to convert
EXC_BAD_ACCESS to panics. However, the Go signal handler refuse to
handle signals that would otherwise be ignored if they arrive at
non-Go threads.

Block all (posix) signals to that thread, making sure that
no unexpected signals arrive to it. At least one test, TestStop in
os/signal, depends on signals not arriving on any non-Go threads.

For #14318

Change-Id: I901467fb53bdadb0d03b0f1a537116c7f4754423
Reviewed-on: https://go-review.googlesource.com/21047
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agocmd/compile: minor cleanups
Dave Cheney [Thu, 24 Mar 2016 09:57:53 +0000 (20:57 +1100)]
cmd/compile: minor cleanups

Some minor scoping cleanups found by a very old version of grind.

Change-Id: I1d373817586445fc87e38305929097b652696fdd
Reviewed-on: https://go-review.googlesource.com/21064
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoimage/png: ignore trailing IDAT chunks
Tilman Dilo [Wed, 23 Mar 2016 21:38:52 +0000 (22:38 +0100)]
image/png: ignore trailing IDAT chunks

Ignore superfluous trailing IDAT chunks which were not consumed when decoding
the image. This change fixes decoding of valid images in which a zero-length
IDAT chunk appears after the actual image data. It also prevents decoding of
trailing garbage IDAT chunks or maliciously embedded additional images.

Fixes #14936

Change-Id: I8c76cfa9a03496d9576f72bed2db109271f97c5e
Reviewed-on: https://go-review.googlesource.com/21045
Reviewed-by: Nigel Tao <nigeltao@golang.org>
9 years agoos: fix Stdin.Stat() on windows
Yasuhiro Matsumoto [Fri, 18 Mar 2016 04:00:26 +0000 (13:00 +0900)]
os: fix Stdin.Stat() on windows

If name is /dev/{stdin,stdout,stderr}, return fileInfo.

Fixes #14853.

Change-Id: Ibf7d1ae7b9f3dc43f6ed7c905ea2c5102e1971cc
Reviewed-on: https://go-review.googlesource.com/20845
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
9 years agonet/http: redirect if the URL path is a dir & doesn't end in a slash
Mohit Agarwal [Thu, 3 Mar 2016 18:23:39 +0000 (23:53 +0530)]
net/http: redirect if the URL path is a dir & doesn't end in a slash

Fixes #13996

Change-Id: I9b2c7fba0705900aca9a70bc6a2687667a9a976c
Reviewed-on: https://go-review.googlesource.com/20128
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile/internal/gc: remove remaining Nod(OXXX, ...)
Dave Cheney [Wed, 23 Mar 2016 05:01:15 +0000 (16:01 +1100)]
cmd/compile/internal/gc: remove remaining Nod(OXXX, ...)

Remove almost all the remaining Nod(OXXX, ... ) uses. The performance
change is due entirely to the changes to func temp(*Type). The other
cleanups have no effect, as expected.

I'll address the remaining Nod(OXXX, ...) uses in a followup CL as they
are very sensitive to change.

lucky(~/go/src/cmd/compile) % benchstat /tmp/{old,new}.txt
name      old time/op    new time/op    delta
Template     391ms ± 6%     385ms ± 6%    ~     (p=0.127 n=19+20)
GoTypes      1.27s ± 2%     1.27s ± 2%    ~     (p=0.172 n=19+19)
Compiler     6.17s ± 2%     6.15s ± 2%    ~     (p=0.647 n=19+20)

name      old alloc/op   new alloc/op   delta
Template    63.7MB ± 0%    63.4MB ± 0%  -0.35%  (p=0.000 n=16+20)
GoTypes      219MB ± 0%     218MB ± 0%  -0.38%  (p=0.000 n=20+20)
Compiler     980MB ± 0%     976MB ± 0%  -0.38%  (p=0.000 n=20+20)

name      old allocs/op  new allocs/op  delta
Template      586k ± 0%      584k ± 0%  -0.30%  (p=0.000 n=20+20)
GoTypes      1.80M ± 0%     1.79M ± 0%  -0.31%  (p=0.000 n=20+20)
Compiler     7.74M ± 0%     7.71M ± 0%  -0.34%  (p=0.000 n=20+20)

Change-Id: Ie21a5443c33a23ce30f987bdddec9fe350365d35
Reviewed-on: https://go-review.googlesource.com/21017
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: remove redundant parameter from finishcompare
Josh Bleecher Snyder [Wed, 23 Mar 2016 23:43:17 +0000 (16:43 -0700)]
cmd/compile: remove redundant parameter from finishcompare

This is follow-up 3 of 3 to CL 20959.

Passes toolstash -cmp.

Change-Id: I06efded21bbc970cbefa10e8f2cac1ebc6942e1b
Reviewed-on: https://go-review.googlesource.com/21054
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: reduce stutter
Josh Bleecher Snyder [Wed, 23 Mar 2016 15:51:38 +0000 (08:51 -0700)]
cmd/compile: reduce stutter

This is follow-up 1 of 3 to CL 20959.

Passes toolstash -cmp.

Change-Id: I9bddf7d88333fa4755e03ff8a034a35bd01b7855
Reviewed-on: https://go-review.googlesource.com/21052
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http/httptest: add NewRequest helper for ease of testing handlers
Brad Fitzpatrick [Wed, 23 Mar 2016 05:25:27 +0000 (16:25 +1100)]
net/http/httptest: add NewRequest helper for ease of testing handlers

Fixes #14199

Change-Id: Ic9284023b663de3db1ca7b7b1e96eeab82ec0944
Reviewed-on: https://go-review.googlesource.com/21016
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/compile: prettify loop iterations
Marvin Stenger [Wed, 23 Mar 2016 15:35:50 +0000 (16:35 +0100)]
cmd/compile: prettify loop iterations

This commit replaces some of

for i := len(x) - 1; i >= 0; i-- {...}

style loops, which do not rely on reverse iteration order.

Change-Id: I5542834286562da058200c06e7a173b13760e54d
Reviewed-on: https://go-review.googlesource.com/21044
Reviewed-by: Keith Randall <khr@golang.org>
9 years agocmd/compile: reduce some SSA garbage
Brad Fitzpatrick [Sun, 20 Mar 2016 20:20:01 +0000 (07:20 +1100)]
cmd/compile: reduce some SSA garbage

It's pretty hard to get reliable CPU numbers, even with 50 runs on an
otherwise-idle physical Linux machine, but the garbage reduction
numbers are nice. To get useful time/op numbers, I modified
compilebench to report user CPU time instead of wall time:

name       old time/op     new time/op     delta
Template       547ms ± 6%      557ms ± 5%   +1.80%        (p=0.001 n=49+49)
Unicode        360ms ± 9%      365ms ± 6%     ~           (p=0.094 n=50+45)
GoTypes        1.84s ± 3%      1.82s ± 3%   -1.50%        (p=0.000 n=50+49)
Compiler       9.19s ± 2%      9.02s ± 2%   -1.87%        (p=0.000 n=45+50)

name       old alloc/op    new alloc/op    delta
Template      63.3MB ± 0%     59.1MB ± 0%   -6.72%        (p=0.000 n=50+50)
Unicode       43.1MB ± 0%     42.9MB ± 0%   -0.47%        (p=0.000 n=50+49)
GoTypes        220MB ± 0%      200MB ± 0%   -9.00%        (p=0.000 n=50+50)
Compiler      1.00GB ± 0%     0.89GB ± 0%  -10.09%        (p=0.000 n=50+49)

name       old allocs/op   new allocs/op   delta
Template        681k ± 0%       680k ± 0%   -0.16%        (p=0.000 n=50+48)
Unicode         541k ± 0%       541k ± 0%   -0.02%        (p=0.011 n=48+50)
GoTypes        2.08M ± 0%      2.08M ± 0%   -0.19%        (p=0.000 n=48+50)
Compiler       9.24M ± 0%      9.23M ± 0%   -0.11%        (p=0.000 n=50+50)

Change-Id: I1fac4ebf85a1783e3289c3ffb1ed365442837643
Reviewed-on: https://go-review.googlesource.com/20995
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agocmd/compile: small Mpint method simplifications
Matthew Dempsky [Wed, 23 Mar 2016 18:07:20 +0000 (11:07 -0700)]
cmd/compile: small Mpint method simplifications

Get rid of (*Mpint).Add's "quiet" parameter: it's always set to 0.

Inline (*Mpint).shift into (*Mpint).Lsh and (*Mpint).Rsh. There's no
need for a common shift method that can handle both left or right
shifts based on sign when the higher level abstractions only ever do
one or the other.

Change-Id: Icd3b082413f9193961b6835279e0bd4b6a6a6621
Reviewed-on: https://go-review.googlesource.com/21050
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/compile: start on ARM port
Keith Randall [Tue, 22 Mar 2016 05:57:26 +0000 (22:57 -0700)]
cmd/compile: start on ARM port

Start working on arm port.  Gets close to correct
code for fibonacci:
    func fib(n int) int {
        if n < 2 {
            return n
        }
        return fib(n-1) + fib(n-2)
    }

Still a lot to do, but this is a good starting point.

Cleaned up some arch-specific dependencies in regalloc.

Change-Id: I4301c6c31a8402168e50dcfee8bcf7aee73ea9d5
Reviewed-on: https://go-review.googlesource.com/21000
Reviewed-by: David Chase <drchase@google.com>
9 years agocmd/link, reflect: remove some method type data
David Crawshaw [Wed, 23 Mar 2016 01:25:40 +0000 (21:25 -0400)]
cmd/link, reflect: remove some method type data

Remove reflect type information for unexported methods that do not
satisfy any interface in the program.

Ideally the unexported method would not appear in the method list at
all, but that is tricky because the slice is built by the compiler.

Reduces binary size:

cmd/go: 81KB (0.8%)
jujud: 258KB (0.4%)

For #6853.

Change-Id: I25ef8df6907e9ac03b18689d584ea46e7d773043
Reviewed-on: https://go-review.googlesource.com/21033
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: don't simplify nilchecks in loops
Alexandru Moșoi [Wed, 23 Mar 2016 10:11:34 +0000 (11:11 +0100)]
cmd/compile: don't simplify nilchecks in loops

khr: Lifting the nil check out of the loop altogether is an admirable
goal, and this rewrite is one step on the way. But without lifting it
out of the loop, the rewrite is just hurting us.

Fixes #14917

Change-Id: Idb917f37d89f50f8e046d5ebd7c092b1e0eb0633
Reviewed-on: https://go-review.googlesource.com/21040
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agobytes: Equal perf improvements on ppc64le/ppc64
Lynn Boger [Mon, 7 Mar 2016 17:59:16 +0000 (11:59 -0600)]
bytes: Equal perf improvements on ppc64le/ppc64

The existing implementation for Equal and similar
functions in the bytes package operate on one byte at
at time.  This performs poorly on ppc64/ppc64le especially
when the byte buffers are large.  This change improves
those functions by loading and comparing double words where
possible.  The common code has been moved to a function
that can be shared by the other functions in this
file which perform the same type of comparison.
Further optimizations are done for the case where
>= 32 bytes are being compared.  The new function
memeqbody is used by memeq_varlen, Equal, and eqstring.

When running the bytes test with -test.bench=Equal

benchmark                     old MB/s     new MB/s     speedup
BenchmarkEqual1               164.83       129.49       0.79x
BenchmarkEqual6               563.51       445.47       0.79x
BenchmarkEqual9               656.15       1099.00      1.67x
BenchmarkEqual15              591.93       1024.30      1.73x
BenchmarkEqual16              613.25       1914.12      3.12x
BenchmarkEqual20              682.37       1687.04      2.47x
BenchmarkEqual32              807.96       3843.29      4.76x
BenchmarkEqual4K              1076.25      23280.51     21.63x
BenchmarkEqual4M              1079.30      13120.14     12.16x
BenchmarkEqual64M             1073.28      10876.92     10.13x

It was determined that the degradation in the smaller byte tests
were due to unfavorable code alignment of the single byte loop.

Fixes #14368

Change-Id: I0dd87382c28887c70f4fbe80877a8ba03c31d7cd
Reviewed-on: https://go-review.googlesource.com/20249
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/link: Clean up Pcln struct
Shahar Kohanim [Wed, 23 Mar 2016 12:02:36 +0000 (14:02 +0200)]
cmd/link: Clean up Pcln struct

Removes unnecessary fields from Pcln.

Change-Id: I175049ca749b510eedaf65162355bc4d7a93315e
Reviewed-on: https://go-review.googlesource.com/21041
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agocompress/flate: rework matching algorithm
Klaus Post [Mon, 21 Mar 2016 09:21:55 +0000 (10:21 +0100)]
compress/flate: rework matching algorithm

This changes how matching is done in deflate algorithm.

The major change is that we do not look for matches that are only
3 bytes in length, matches must be 4 bytes at least.
Contrary to what you would expect this actually improves the
compresion ratio, since 3 literal bytes will often be shorter
than a match after huffman encoding.
This varies a bit by source, but is most often the case when the
source is "easy" to compress.

Second of all, a "stronger" hash is used. The hash is similar to
the hashing function used by Snappy.

Overall, the speed impact is biggest on higher compression levels.
I intend to replace the "speed" compression level, which can be
seen in CL 21021.

The built-in benchmark using "digits" is slower at level 1.
I see this as an exception, since "digits" is a special type
of data, where you have low entropy (numbers 0->9), but no
significant matches. Again, CL 20021 fixes that case.

NewWriterDict is also made considerably faster, by not running data
through the entire encoder. This is not reflected by the benchmark.

Overall, the speed impact is biggest on higher compression levels.
I intend to replace the "speed" compression level.

COMPARED to tip/master:
name                       old time/op    new time/op     delta
EncodeDigitsSpeed1e4-4        401µs ± 1%      345µs ± 2%   -13.95%
EncodeDigitsSpeed1e5-4       3.19ms ± 1%     4.27ms ± 3%   +33.96%
EncodeDigitsSpeed1e6-4       27.7ms ± 4%     43.8ms ± 3%   +58.00%
EncodeDigitsDefault1e4-4      641µs ± 0%      403µs ± 1%   -37.15%
EncodeDigitsDefault1e5-4     13.8ms ± 1%      6.4ms ± 3%   -53.73%
EncodeDigitsDefault1e6-4      162ms ± 1%       64ms ± 2%   -60.51%
EncodeDigitsCompress1e4-4     627µs ± 1%      405µs ± 2%   -35.45%
EncodeDigitsCompress1e5-4    13.9ms ± 0%      6.3ms ± 2%   -54.46%
EncodeDigitsCompress1e6-4     159ms ± 1%       64ms ± 0%   -59.91%
EncodeTwainSpeed1e4-4         433µs ± 4%      331µs ± 1%   -23.53%
EncodeTwainSpeed1e5-4        2.82ms ± 1%     3.08ms ± 0%    +9.10%
EncodeTwainSpeed1e6-4        28.1ms ± 2%     28.8ms ± 0%    +2.82%
EncodeTwainDefault1e4-4       695µs ± 4%      474µs ± 1%   -31.78%
EncodeTwainDefault1e5-4      11.8ms ± 0%      7.4ms ± 0%   -37.31%
EncodeTwainDefault1e6-4       128ms ± 0%       75ms ± 0%   -40.93%
EncodeTwainCompress1e4-4      719µs ± 3%      480µs ± 0%   -33.27%
EncodeTwainCompress1e5-4     15.0ms ± 3%      8.2ms ± 2%   -45.55%
EncodeTwainCompress1e6-4      170ms ± 0%       85ms ± 1%   -49.99%

name                       old speed      new speed       delta
EncodeDigitsSpeed1e4-4     25.0MB/s ± 1%   29.0MB/s ± 2%   +16.24%
EncodeDigitsSpeed1e5-4     31.4MB/s ± 1%   23.4MB/s ± 3%   -25.34%
EncodeDigitsSpeed1e6-4     36.1MB/s ± 4%   22.8MB/s ± 3%   -36.74%
EncodeDigitsDefault1e4-4   15.6MB/s ± 0%   24.8MB/s ± 1%   +59.11%
EncodeDigitsDefault1e5-4   7.27MB/s ± 1%  15.72MB/s ± 3%  +116.23%
EncodeDigitsDefault1e6-4   6.16MB/s ± 0%  15.60MB/s ± 2%  +153.25%
EncodeDigitsCompress1e4-4  15.9MB/s ± 1%   24.7MB/s ± 2%   +54.97%
EncodeDigitsCompress1e5-4  7.19MB/s ± 0%  15.78MB/s ± 2%  +119.62%
EncodeDigitsCompress1e6-4  6.27MB/s ± 1%  15.65MB/s ± 0%  +149.52%
EncodeTwainSpeed1e4-4      23.1MB/s ± 4%   30.2MB/s ± 1%   +30.68%
EncodeTwainSpeed1e5-4      35.4MB/s ± 1%   32.5MB/s ± 0%    -8.34%
EncodeTwainSpeed1e6-4      35.6MB/s ± 2%   34.7MB/s ± 0%    -2.77%
EncodeTwainDefault1e4-4    14.4MB/s ± 4%   21.1MB/s ± 1%   +46.48%
EncodeTwainDefault1e5-4    8.49MB/s ± 0%  13.55MB/s ± 0%   +59.50%
EncodeTwainDefault1e6-4    7.83MB/s ± 0%  13.25MB/s ± 0%   +69.19%
EncodeTwainCompress1e4-4   13.9MB/s ± 3%   20.8MB/s ± 0%   +49.83%
EncodeTwainCompress1e5-4   6.65MB/s ± 3%  12.20MB/s ± 2%   +83.51%
EncodeTwainCompress1e6-4   5.88MB/s ± 0%  11.76MB/s ± 1%  +100.06%

Change-Id: I724e33c1dd3e3a6a1b0a68e094baa959352baf32
Reviewed-on: https://go-review.googlesource.com/20929
Run-TryBot: Nigel Tao <nigeltao@golang.org>
Reviewed-by: Nigel Tao <nigeltao@golang.org>
9 years agocmd/compile/internal/ssa: avoid string conversion in zcse
Dave Cheney [Fri, 18 Mar 2016 02:21:53 +0000 (13:21 +1100)]
cmd/compile/internal/ssa: avoid string conversion in zcse

Some ssa.Type implementations fell through to gc.Tconv which generated
garbage to produce a string form of the Type.

name      old time/op    new time/op    delta
Template     405ms ± 7%     401ms ± 6%    ~     (p=0.478 n=20+20)
GoTypes      1.32s ± 1%     1.30s ± 2%  -1.27%  (p=0.000 n=19+20)
Compiler     6.07s ± 2%     6.03s ± 2%    ~     (p=0.121 n=20+20)

name      old alloc/op   new alloc/op   delta
Template    63.9MB ± 0%    63.7MB ± 0%  -0.21%  (p=0.000 n=19+20)
GoTypes      220MB ± 0%     219MB ± 0%  -0.21%  (p=0.000 n=20+20)
Compiler     966MB ± 0%     965MB ± 0%  -0.11%  (p=0.000 n=20+20)

name      old allocs/op  new allocs/op  delta
Template      708k ± 0%      701k ± 0%  -0.99%  (p=0.000 n=20+20)
GoTypes      2.20M ± 0%     2.17M ± 0%  -1.43%  (p=0.000 n=17+20)
Compiler     9.45M ± 0%     9.36M ± 0%  -0.91%  (p=0.000 n=20+20)

Change-Id: I5fcc30e0f76a823d1c301d4980b583d716a75ce3
Reviewed-on: https://go-review.googlesource.com/20844
Reviewed-by: Keith Randall <khr@golang.org>
9 years agocmd/compile/internal/gc: remove redundant Nod(OXXX, ...) pattern
Dave Cheney [Wed, 23 Mar 2016 02:27:49 +0000 (13:27 +1100)]
cmd/compile/internal/gc: remove redundant Nod(OXXX, ...) pattern

The pattern

    n := Nod(OXXX, nil, nil)
    Nodconst(n, ...)

was a leftover from the C days where n must be heap allocated.

No change in benchmarks, none expected as n escapes anyway.

name      old time/op    new time/op    delta
Template     391ms ± 6%     388ms ± 5%    ~     (p=0.659 n=20+20)
GoTypes      1.27s ± 1%     1.27s ± 2%    ~     (p=0.828 n=18+20)
Compiler     6.16s ± 2%     6.15s ± 1%    ~     (p=0.947 n=20+20)

name      old alloc/op   new alloc/op   delta
Template    63.7MB ± 0%    63.7MB ± 0%    ~     (p=0.414 n=20+20)
GoTypes      219MB ± 0%     219MB ± 0%    ~     (p=0.904 n=20+20)
Compiler     980MB ± 0%     980MB ± 0%  +0.00%  (p=0.007 n=20+19)

name      old allocs/op  new allocs/op  delta
Template      586k ± 0%      586k ± 0%    ~     (p=0.564 n=19+20)
GoTypes      1.80M ± 0%     1.80M ± 0%    ~     (p=0.718 n=20+20)
Compiler     7.74M ± 0%     7.74M ± 0%    ~     (p=0.358 n=20+20)

The reuse of nc in multiple overlapping scopes in walk.go is the worst.

Change-Id: I4ed6a63f7ffbfff68124ad609f6e3a68d95cbbba
Reviewed-on: https://go-review.googlesource.com/21015
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/vet: check lock copy in function calls and return statements
Aliaksandr Valialkin [Fri, 18 Mar 2016 09:25:04 +0000 (11:25 +0200)]
cmd/vet: check lock copy in function calls and return statements

Fixes #14529

Change-Id: I6ed059d279ba0fe12d76416859659f28d61781d2
Reviewed-on: https://go-review.googlesource.com/20832
Run-TryBot: Rob Pike <r@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
9 years agofmt: cleanup and optimize doPrintf for simple formats
Martin Möhrmann [Sun, 20 Mar 2016 14:24:50 +0000 (15:24 +0100)]
fmt: cleanup and optimize doPrintf for simple formats

Make a fast path for format strings that do not use
precision or width specifications or argument indices.

Only check and enforce the restriction to not pad left with zeros
in code paths that change either f.minus or f.zero.

Consolidate the if chains at the end of the main doPrintf loop
into a switch statement. Move error printing into extra
functions to reduce size of this switch statement.

name                             old time/op  new time/op  delta
SprintfPadding-2                  234ns ± 1%   233ns ± 1%   -0.54%  (p=0.010 n=19+19)
SprintfEmpty-2                   37.0ns ± 3%  39.1ns ±14%     ~     (p=0.501 n=17+20)
SprintfString-2                   112ns ± 1%   101ns ± 1%   -9.21%  (p=0.000 n=19+20)
SprintfTruncateString-2           139ns ± 1%   139ns ± 0%   +0.57%  (p=0.000 n=19+19)
SprintfQuoteString-2              402ns ± 0%   392ns ± 0%   -2.35%  (p=0.000 n=19+20)
SprintfInt-2                      114ns ± 1%   102ns ± 2%  -10.92%  (p=0.000 n=20+20)
SprintfIntInt-2                   177ns ± 2%   155ns ± 2%  -12.67%  (p=0.000 n=18+18)
SprintfPrefixedInt-2              260ns ± 3%   249ns ± 3%   -4.55%  (p=0.000 n=20+20)
SprintfFloat-2                    190ns ± 1%   178ns ± 2%   -6.54%  (p=0.000 n=20+20)
SprintfComplex-2                  533ns ± 1%   517ns ± 3%   -2.95%  (p=0.000 n=20+20)
SprintfBoolean-2                  102ns ± 1%    93ns ± 2%   -9.30%  (p=0.000 n=20+20)
SprintfHexString-2                176ns ± 0%   168ns ± 2%   -4.49%  (p=0.000 n=16+19)
SprintfHexBytes-2                 181ns ± 1%   174ns ± 2%   -4.27%  (p=0.000 n=20+20)
SprintfBytes-2                    326ns ± 1%   311ns ± 1%   -4.51%  (p=0.000 n=20+20)
ManyArgs-2                        540ns ± 2%   497ns ± 1%   -8.08%  (p=0.000 n=18+16)
FprintInt-2                       150ns ± 0%   149ns ± 0%   -0.33%  (p=0.000 n=20+18)
FprintfBytes-2                    185ns ± 0%   165ns ± 0%  -10.98%  (p=0.000 n=20+18)
FprintIntNoAlloc-2                113ns ± 0%   112ns ± 0%   -0.88%  (p=0.000 n=20+20)

Change-Id: I9ada8faa1f46aa67ea116a94ab3f4ad3e405c8fe
Reviewed-on: https://go-review.googlesource.com/20919
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
9 years agodatabase/sql/driver: remove string exclusion
Tamir Duberstein [Fri, 12 Feb 2016 05:37:52 +0000 (00:37 -0500)]
database/sql/driver: remove string exclusion

The exclusion of string from IsScanValue prevents driver authors from
writing their drivers in such a way that would allow users to
distinguish between strings and byte arrays returned from a database.
Such drivers are possible today, but require their authors to deviate
from the guidance provided by the standard library.

This exclusion has been in place since the birth of this package in
https://github.com/golang/go/commit/357f2cb1a385f4d1418e48856f9abe0cce,
but the fakedb implementation shipped in the same commit violates the
exclusion!

Strictly speaking this is a breaking change, but it increases the set
of permissible Scan types, and should not cause breakage in practice.

No test changes are necessary because fakedb already exercises this.

Fixes #6497.

Change-Id: I69dbd3a59d90464bcae8c852d7ec6c97bfd120f8
Reviewed-on: https://go-review.googlesource.com/19439
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomisc/cgo/testcarchive: rewrite test from bash to Go
Ian Lance Taylor [Wed, 23 Mar 2016 01:17:43 +0000 (18:17 -0700)]
misc/cgo/testcarchive: rewrite test from bash to Go

This is to support https://golang.org/cl/18057, which is going to add
Windows support to this directory.  Better to write the test in Go then
to have both test.bash and test.bat.

Update #13494.

Change-Id: I4af7004416309e885049ee60b9470926282f210d
Reviewed-on: https://go-review.googlesource.com/20892
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/dist: disable misc/cgo/fortran test on dragonfly
Mikio Hara [Wed, 23 Mar 2016 00:14:58 +0000 (09:14 +0900)]
cmd/dist: disable misc/cgo/fortran test on dragonfly

Updates #14544.

Change-Id: I24ab8e6f9ad9d290a672216fc2f50f78c3ed8812
Reviewed-on: https://go-review.googlesource.com/21014
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile: MOVBload and MOVBQZXload are the same op
Keith Randall [Tue, 22 Mar 2016 23:22:21 +0000 (16:22 -0700)]
cmd/compile: MOVBload and MOVBQZXload are the same op

No need to have both ops when they do the same thing.
Just declare MOVBload to zero extend and we can get rid
of MOVBQZXload.  Same for W and L.

Kind of a followon cleanup for https://go-review.googlesource.com/c/19506/
Should enable an easier fix for #14920

Change-Id: I7cfac909a8ba387f433a6ae75c050740ebb34d42
Reviewed-on: https://go-review.googlesource.com/21004
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agosyscall: correct spelling/typos in comment
Michael Munday [Tue, 22 Mar 2016 19:08:09 +0000 (15:08 -0400)]
syscall: correct spelling/typos in comment

Change-Id: Ib44c6b1ce07aa8fb67033cf21e177a90fd4005dc
Reviewed-on: https://go-review.googlesource.com/21002
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/link: use stdlib sort in dodata
Shahar Kohanim [Wed, 9 Mar 2016 14:23:25 +0000 (16:23 +0200)]
cmd/link: use stdlib sort in dodata

Speeds up linking cmd/go by 1.7%

name       old s/op   new s/op   delta
LinkCmdGo  0.58 ± 4%  0.57 ± 5%  -1.74%  (p=0.000 n=96+97)

Change-Id: I7844cf4e2eeac260318de2b6ddf52ce07a6e00f5
Reviewed-on: https://go-review.googlesource.com/20915
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agocmd/compile: remove dots from register names
Keith Randall [Tue, 22 Mar 2016 16:43:28 +0000 (09:43 -0700)]
cmd/compile: remove dots from register names

They are kind of useless and are cluttering up
https://go-review.googlesource.com/c/21000/

Change-Id: Iafdec75ada11c7ebdc40540d251fdc514bb00d3d
Reviewed-on: https://go-review.googlesource.com/21001
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/compile: update vendored math/big to latest version
Robert Griesemer [Fri, 18 Mar 2016 20:38:52 +0000 (13:38 -0700)]
cmd/compile: update vendored math/big to latest version

This makes the rounding bug fix in math/big for issue 14651 available
to the compiler.

- changes to cmd/compile/internal/big fully automatic via script
- added test case for issue
- updated old test case with correct test data

Fixes #14651.

Change-Id: Iea37a2cd8d3a75f8c96193748b66156a987bbe40
Reviewed-on: https://go-review.googlesource.com/20818
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomath/big: much simplified and faster Float rounding
Robert Griesemer [Fri, 18 Mar 2016 19:31:37 +0000 (12:31 -0700)]
math/big: much simplified and faster Float rounding

Change-Id: Iab0add7aee51a8c72a81f51d980d22d2fd612f5c
Reviewed-on: https://go-review.googlesource.com/20817
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/compile: simplify SliceCap when it equals SliceLen
Alexandru Moșoi [Tue, 22 Mar 2016 16:34:36 +0000 (17:34 +0100)]
cmd/compile: simplify SliceCap when it equals SliceLen

Shows up occassionally, especially after p = p[:8:len(p)]

Updates #14905

Change-Id: Iab35ef2eac57817e6a10c6aaeeb84709e8021641
Reviewed-on: https://go-review.googlesource.com/21025
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agotesting: expose subtest and subbenchmark functionality
Marcel van Lohuizen [Fri, 29 Jan 2016 16:14:51 +0000 (17:14 +0100)]
testing: expose subtest and subbenchmark functionality

Fixes #12166

Change-Id: Ie62cba2c39beb5732447ba3688c93c08ef12abb5
Reviewed-on: https://go-review.googlesource.com/18898
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>

9 years agotesting: added name matcher and sanitizer
Marcel van Lohuizen [Fri, 29 Jan 2016 15:57:02 +0000 (16:57 +0100)]
testing: added name matcher and sanitizer

The matcher is responsible for sanitizing and uniquing the
test and benchmark names and thus needs to be included before the
API can be exposed.

Matching currently uses the regexp to only match the top-level
tests/benchmarks.

Support for subtest matching is for another CL.

Change-Id: I7c8464068faef7ebc179b03a7fe3d01122cc4f0b
Reviewed-on: https://go-review.googlesource.com/18897
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Marcel van Lohuizen <mpvl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: reduce use of **Node parameters
Josh Bleecher Snyder [Sun, 20 Mar 2016 15:03:31 +0000 (08:03 -0700)]
cmd/compile: reduce use of **Node parameters

Escape analysis has a hard time with tree-like
structures (see #13493 and #14858).
This is unlikely to change.
As a result, when invoking a function that accepts
a **Node parameter, we usually allocate a *Node
on the heap. This happens a whole lot.

This CL changes functions from taking a **Node
to acting more like append: It both modifies
the input and returns a replacement for it.

Because of the cascading nature of escape analysis,
in order to get the benefits, I had to modify
almost all such functions. The remaining functions
are in racewalk and the backend. I would be happy
to update them as well in a separate CL.

This CL was created by manually updating the
function signatures and the directly impacted
bits of code. The callsites were then automatically
updated using a bespoke script:
https://gist.github.com/josharian/046b1be7aceae244de39

For ease of reviewing and future understanding,
this CL is also broken down into four CLs,
mailed separately, which show the manual
and the automated changes separately.
They are CLs 20990, 20991, 20992, and 20993.

Passes toolstash -cmp.

name       old time/op     new time/op     delta
Template       335ms ± 5%      324ms ± 5%   -3.35%        (p=0.000 n=23+24)
Unicode        176ms ± 9%      165ms ± 6%   -6.12%        (p=0.000 n=23+24)
GoTypes        1.10s ± 4%      1.07s ± 2%   -2.77%        (p=0.000 n=24+24)
Compiler       5.31s ± 3%      5.15s ± 3%   -2.95%        (p=0.000 n=24+24)
MakeBash       41.6s ± 1%      41.7s ± 2%     ~           (p=0.586 n=23+23)

name       old alloc/op    new alloc/op    delta
Template      63.3MB ± 0%     62.4MB ± 0%   -1.36%        (p=0.000 n=25+23)
Unicode       42.4MB ± 0%     41.6MB ± 0%   -1.99%        (p=0.000 n=24+25)
GoTypes        220MB ± 0%      217MB ± 0%   -1.11%        (p=0.000 n=25+25)
Compiler       994MB ± 0%      973MB ± 0%   -2.08%        (p=0.000 n=24+25)

name       old allocs/op   new allocs/op   delta
Template        681k ± 0%       574k ± 0%  -15.71%        (p=0.000 n=24+25)
Unicode         518k ± 0%       413k ± 0%  -20.34%        (p=0.000 n=25+24)
GoTypes        2.08M ± 0%      1.78M ± 0%  -14.62%        (p=0.000 n=25+25)
Compiler       9.26M ± 0%      7.64M ± 0%  -17.48%        (p=0.000 n=25+25)

name       old text-bytes  new text-bytes  delta
HelloSize       578k ± 0%       578k ± 0%     ~     (all samples are equal)
CmdGoSize      6.46M ± 0%      6.46M ± 0%     ~     (all samples are equal)

name       old data-bytes  new data-bytes  delta
HelloSize       128k ± 0%       128k ± 0%     ~     (all samples are equal)
CmdGoSize       281k ± 0%       281k ± 0%     ~     (all samples are equal)

name       old exe-bytes   new exe-bytes   delta
HelloSize       921k ± 0%       921k ± 0%     ~     (all samples are equal)
CmdGoSize      9.86M ± 0%      9.86M ± 0%     ~     (all samples are equal)

Change-Id: I277d95bd56d51c166ef7f560647aeaa092f3f475
Reviewed-on: https://go-review.googlesource.com/20959
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agodebug/dwarf: add Reader.SeekPC and Data.Ranges
Ian Lance Taylor [Wed, 16 Mar 2016 21:15:54 +0000 (14:15 -0700)]
debug/dwarf: add Reader.SeekPC and Data.Ranges

These new methods help find the compilation unit to pass to the
LineReader method in order to find the line information for a PC.
The Ranges method also helps identify the specific function for a PC,
needed to determine the function name.

This uses the .debug.ranges section if necessary, and changes the object
file format packages to pass in the section contents if available.

Change-Id: I5ebc3d27faaf1a126ffb17a1e6027efdf64af836
Reviewed-on: https://go-review.googlesource.com/20769
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoencoding/json, internal/testenv: use Fatalf
Dominik Honnef [Tue, 22 Mar 2016 05:37:16 +0000 (06:37 +0100)]
encoding/json, internal/testenv: use Fatalf

Change-Id: I64dd09e76d811000a914776fdad47808e3895690
Reviewed-on: https://go-review.googlesource.com/20989
Reviewed-by: Dave Cheney <dave@cheney.net>
9 years agomisc/cgo/test: add RewindAndSetgid implementation for s390x
Michael Munday [Mon, 21 Mar 2016 02:00:06 +0000 (22:00 -0400)]
misc/cgo/test: add RewindAndSetgid implementation for s390x

Required to pass the issue9400 test.

Change-Id: I595223c403b12faade54e2e46510f8537150af39
Reviewed-on: https://go-review.googlesource.com/20940
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agointernal/syscall/unix: add randomTrap const for s390x
Michael Munday [Mon, 21 Mar 2016 01:56:12 +0000 (21:56 -0400)]
internal/syscall/unix: add randomTrap const for s390x

Change-Id: I81376f524e76db25fd52cc5bec2c80fbf618a0c5
Reviewed-on: https://go-review.googlesource.com/20877
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/internal/obj: add support for s390x
Michael Munday [Fri, 18 Mar 2016 20:30:29 +0000 (16:30 -0400)]
cmd/internal/obj: add support for s390x

Adds a new R_PCRELDBL relocation for 2-byte aligned relative
relocations on s390x. Should be removed once #14218 is
implemented.

Change-Id: I79dd2d8e746ba8cbc26c570faccfdd691e8161e8
Reviewed-on: https://go-review.googlesource.com/20941
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: change the way SSA does slice zero-cap detection
Keith Randall [Mon, 21 Mar 2016 22:24:08 +0000 (15:24 -0700)]
cmd/compile: change the way SSA does slice zero-cap detection

There is a special case for slicing s[i:j] when the resulting
slice has zero capacity, to prevent pointing to the next object
in memory.

Change this special case code from:
  rptr := rcap == 0 ? ptr : ptr+i*elemsize
to
  rptr := ptr + (rcap == 0 ? 0 : i) * elemsize

This change leads to slightly smaller generated code, replacing
a load with a register zero.

old:
0x002e 00046 (slice.go:8) CMPQ BX, $0
0x0032 00050 (slice.go:8) JEQ $0, 78
0x0034 00052 (slice.go:8) MOVQ "".a+8(FP), BP
0x0039 00057 (slice.go:8) LEAQ (BP)(CX*8), AX
0x003e 00062 ... rest of function ...

0x004e 00078 (slice.go:7) MOVQ "".a+8(FP), AX
0x0053 00083 (slice.go:8) JMP 62

new:
0x002e 00046 (slice.go:8) CMPQ BX, $0
0x0032 00050 (slice.go:8) JEQ $0, 78
0x0034 00052 (slice.go:8) MOVQ "".a+8(FP), BP
0x0039 00057 (slice.go:8) LEAQ (BP)(CX*8), AX
0x003e 00062 ... rest of function...

0x004e 00078 (slice.go:8) MOVQ $0, CX
0x0050 00080 (slice.go:8) JMP 52

Change-Id: I2a396616b0d7b090c226a47c92a7ba14b128401f
Reviewed-on: https://go-review.googlesource.com/20994
Reviewed-by: David Chase <drchase@google.com>
9 years agonet/http, internal/testenv: find go binary in PATH
David Crawshaw [Mon, 21 Mar 2016 18:41:16 +0000 (14:41 -0400)]
net/http, internal/testenv: find go binary in PATH

Fixes #14901

Change-Id: Ia32e09767374a341c9a36c5d977d47d7d1a82315
Reviewed-on: https://go-review.googlesource.com/20967
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>

9 years agocmd/dist: allow gohostarch to be s390x
Michael Munday [Mon, 21 Mar 2016 15:02:55 +0000 (11:02 -0400)]
cmd/dist: allow gohostarch to be s390x

Should let the s390x builder progress a little further.

Change-Id: I5eab5f384b0b039f8e246ba69ecfb24de08625d2
Reviewed-on: https://go-review.googlesource.com/20965
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: allow naming of subexpressions
Keith Randall [Mon, 21 Mar 2016 23:18:45 +0000 (16:18 -0700)]
cmd/compile: allow naming of subexpressions

Allow names to be used for subexpressions of match rules.
For example:

(OpA x:(OpB y)) -> ..use x here to refer to the OpB value..

This gets rid of the .Args[0].Args[0]... way of naming we
used to use.

While we're here, give all subexpression matches names instead
of recomputing them with .Args[i] sequences each time they
are referenced.  Makes the generated rule code a bit smaller.

Change-Id: Ie42139f6f208933b75bd2ae8bd34e95419bc0e4e
Reviewed-on: https://go-review.googlesource.com/20997
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Todd Neal <todd@tneal.org>
9 years agocmd/compile: don't write back unchanged slice results
Keith Randall [Mon, 21 Mar 2016 17:22:03 +0000 (10:22 -0700)]
cmd/compile: don't write back unchanged slice results

Don't write back parts of a slicing operation if they
are unchanged from the source of the slice.  For example:

x.s = x.s[0:5]         // don't write back pointer or cap
x.s = x.s[:5]          // don't write back pointer or cap
x.s = x.s[:5:7]        // don't write back pointer

There is more to be done here, for example:

x.s = x.s[:len(x.s):7] // don't write back ptr or len

This CL can't handle that one yet.

Fixes #14855

Change-Id: Id1e1a4fa7f3076dc1a76924a7f1cd791b81909bb
Reviewed-on: https://go-review.googlesource.com/20954
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Keith Randall <khr@golang.org>

9 years agocmd/compile: fold IsInBounds with small index
Alexandru Moșoi [Sun, 20 Mar 2016 21:15:27 +0000 (22:15 +0100)]
cmd/compile: fold IsInBounds with small index

For the following example, but there are a few more in the stdlib:
func histogram(b []byte, h *[256]int32) {
        for _, t := range b {
                h[t]++
        }
}

Change-Id: I56615f341ae52e02ef34025588dc6d1c52122295
Reviewed-on: https://go-review.googlesource.com/20924
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: allow inlining of functions with switch statements
Todd Neal [Thu, 17 Mar 2016 02:29:17 +0000 (21:29 -0500)]
cmd/compile: allow inlining of functions with switch statements

Allow inlining of functions with switch statements as long as they don't
contain a break or type switch.

Fixes #13071

Change-Id: I057be351ea4584def1a744ee87eafa5df47a7f6d
Reviewed-on: https://go-review.googlesource.com/20824
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomath/big: fix rounding to smallest denormal for Float.Float32/64
Robert Griesemer [Fri, 18 Mar 2016 18:16:35 +0000 (11:16 -0700)]
math/big: fix rounding to smallest denormal for Float.Float32/64

Converting a big.Float value x to a float32/64 value did not correctly
round x up to the smallest denormal float32/64 if x was smaller than the
smallest denormal float32/64, but larger than 0.5 of a smallest denormal
float32/64.

Handle this case explicitly and simplify some code in the turn.

For #14651.

Change-Id: I025e24bf8f0e671581a7de0abf7c1cd7e6403a6c
Reviewed-on: https://go-review.googlesource.com/20816
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agoencoding/binary: fix bound check
Alexandru Moșoi [Mon, 21 Mar 2016 14:05:54 +0000 (15:05 +0100)]
encoding/binary: fix bound check

The inserted early bound checks cause the slice
to expand beyond the original length of the slice.

Change-Id: Ib38891605f4a9a12d3b9e2071a5f77640b083d2d
Reviewed-on: https://go-review.googlesource.com/20981
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agoruntime: use MOVSB instead of MOVSQ for unaligned moves
Keith Randall [Mon, 7 Mar 2016 00:58:30 +0000 (16:58 -0800)]
runtime: use MOVSB instead of MOVSQ for unaligned moves

MOVSB is quite a bit faster for unaligned moves.
Possibly we should use MOVSB all of the time, but Intel folks
say it might be a bit faster to use MOVSQ on some processors
(but not any I have access to at the moment).

benchmark                              old ns/op     new ns/op     delta
BenchmarkMemmove4096-8                 93.9          93.2          -0.75%
BenchmarkMemmoveUnalignedDst4096-8     256           151           -41.02%
BenchmarkMemmoveUnalignedSrc4096-8     175           90.5          -48.29%

Fixes #14630

Change-Id: I568e6d6590eb3615e6a699fb474020596be665ff
Reviewed-on: https://go-review.googlesource.com/20293
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/internal/obj: change linkgetline from C to Go func style
Josh Bleecher Snyder [Mon, 21 Mar 2016 17:47:24 +0000 (10:47 -0700)]
cmd/internal/obj: change linkgetline from C to Go func style

Passes toolstash -cmp.

Change-Id: I8725dee490778be9c1fd31990a6b27df9713c3c9
Reviewed-on: https://go-review.googlesource.com/20957
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile: remove formatting dreg
Josh Bleecher Snyder [Mon, 21 Mar 2016 17:43:32 +0000 (10:43 -0700)]
cmd/compile: remove formatting dreg

Left over from CL 20931.

Change-Id: I3b8dd9ef748bcbf70b5118da28135aaa1e5ba3a8
Reviewed-on: https://go-review.googlesource.com/20955
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/compile: special-case const comparisons against zero
Robert Griesemer [Mon, 21 Mar 2016 17:52:03 +0000 (10:52 -0700)]
cmd/compile: special-case const comparisons against zero

Constant comparisons against 0 are reasonably common.
Special-case and avoid allocating a new zero value each time.

Change-Id: I6c526c8ab30ef7f0fef59110133c764b7b90ba05
Reviewed-on: https://go-review.googlesource.com/20956
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/compile: change Mp{int,flt} functions into methods
Matthew Dempsky [Sun, 20 Mar 2016 20:55:42 +0000 (13:55 -0700)]
cmd/compile: change Mp{int,flt} functions into methods

Also give them more idiomatic Go names. Adding godocs is outside the
scope of this CL. (Besides, the method names almost all directly
parallel an underlying math/big.Int or math/big.Float method.)

CL prepared mechanically with sed (for rewriting mpint.go/mpfloat.go)
and gofmt (for rewriting call sites).

Passes toolstash -cmp.

Change-Id: Id76f4aee476ba740f48db33162463e7978c2083d
Reviewed-on: https://go-review.googlesource.com/20909
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agotesting: add test for not exceeding maximum parallism
Marcel van Lohuizen [Fri, 18 Mar 2016 18:11:19 +0000 (19:11 +0100)]
testing: add test for not exceeding maximum parallism

Fixed bug that slipped probably slipped in after rebasing and
explain why it failed on nacl/netbsd/plan9, which set default
maxparallelism to 1.

Change-Id: I4d59682fb2843d138b320334189f53fcdda5b2f6
Reviewed-on: https://go-review.googlesource.com/20980
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/cmplx: added clarifying comment
Robert Griesemer [Mon, 21 Mar 2016 16:11:45 +0000 (09:11 -0700)]
math/cmplx: added clarifying comment

Fixes #14890.

Change-Id: Ie790276b0e2ef94c92db3a777042d750269f876a
Reviewed-on: https://go-review.googlesource.com/20953
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agosyscall: change clone argument order on s390x
Michael Munday [Fri, 18 Mar 2016 22:55:26 +0000 (18:55 -0400)]
syscall: change clone argument order on s390x

The Linux ABI takes arguments in a different order on s390x.

Change-Id: Ic9cfcc22a5ea3d8ef77d4dd0b915fc266ff3e5f7
Reviewed-on: https://go-review.googlesource.com/20960
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agocmd/go: add s390x support
Michael Munday [Fri, 18 Mar 2016 21:14:23 +0000 (17:14 -0400)]
cmd/go: add s390x support

Minimum architecture of z196 required so that GCC can assemble
gcc_s390x.S in runtime/cgo.

Change-Id: I603ed2edd39f826fb8193740ece5bd11d18c3dc5
Reviewed-on: https://go-review.googlesource.com/20876
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agointernal/syscall/unix: document randomTrap
Brad Fitzpatrick [Mon, 21 Mar 2016 07:27:21 +0000 (18:27 +1100)]
internal/syscall/unix: document randomTrap

Updates #10848

Change-Id: I8353100ed01cb0e8fc19225157f5709bae388612
Reviewed-on: https://go-review.googlesource.com/20975
Reviewed-by: Rob Pike <r@golang.org>
9 years agotest: add s390x case to nosplit test
Michael Munday [Sun, 20 Mar 2016 23:41:34 +0000 (19:41 -0400)]
test: add s390x case to nosplit test

Fixes this test on s390x.

Change-Id: Ie5b70e8191169867765ec9248d827ca12c6405f4
Reviewed-on: https://go-review.googlesource.com/20964
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime/internal/sys: add s390x support
Michael Munday [Fri, 18 Mar 2016 23:17:42 +0000 (19:17 -0400)]
runtime/internal/sys: add s390x support

Change-Id: I928532b406a3457d2c5f75f4de7d46a3f795192e
Reviewed-on: https://go-review.googlesource.com/20939
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agocmd/link: remove Link.Nsymbol
Michael Hudson-Doyle [Mon, 21 Mar 2016 06:55:04 +0000 (19:55 +1300)]
cmd/link: remove Link.Nsymbol

It was just a funny way of saying len(Ctxt.Allsym) by now.

Change-Id: Iff75e73c9f7ec4ba26cfef479bbd05d7dcd172f5
Reviewed-on: https://go-review.googlesource.com/20973
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>