]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agocmd/go: fix TestCgoContainsSpace
Hiroshi Ioka [Fri, 12 May 2017 00:04:55 +0000 (09:04 +0900)]
cmd/go: fix TestCgoContainsSpace

TestCgoContainsSpace builds a small program which mimics $CC.
Usually, $CC attempts to compile a trivial code to detect its own
supported flags (i.e. "-no-pie", which must be passed on some systems),
however the mimic didn't consider these cases.

This CL solve the issue.

Also, use the same name as $CC, it may solve other potential problems.

Fixes #20324

Change-Id: I7a00ac016a5fd0667540f2a715371f8152edc395
Reviewed-on: https://go-review.googlesource.com/43330
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix store chain in schedule pass
Keith Randall [Thu, 11 May 2017 21:46:49 +0000 (14:46 -0700)]
cmd/compile: fix store chain in schedule pass

Tuple ops are weird. They are essentially a pair of ops,
one which consumes a mem and one which generates a mem (the Select1).
The schedule pass didn't handle these quite right.

Fix the scheduler to include both parts of the paired op in
the store chain. That makes sure that loads are correctly ordered
with respect to the first of the pair.

Add a check for the ssacheck builder, that there is only one
live store at a time. I thought we already had such a check, but
apparently not...

Fixes #20335

Change-Id: I59eb3446a329100af38d22820b1ca2190ca46a78
Reviewed-on: https://go-review.googlesource.com/43294
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: restore panic deduplication
Josh Bleecher Snyder [Thu, 11 May 2017 18:41:02 +0000 (11:41 -0700)]
cmd/compile: restore panic deduplication

The switch to detailed position information broke
the removal of duplicate panics on the same line.
Restore it.

Neutral compiler performance impact:

name        old alloc/op      new alloc/op      delta
Template         38.8MB ± 0%       38.8MB ± 0%    ~     (p=0.690 n=5+5)
Unicode          28.7MB ± 0%       28.7MB ± 0%  +0.13%  (p=0.032 n=5+5)
GoTypes           109MB ± 0%        109MB ± 0%    ~     (p=1.000 n=5+5)
Compiler          457MB ± 0%        457MB ± 0%    ~     (p=0.151 n=5+5)
SSA              1.09GB ± 0%       1.10GB ± 0%  +0.17%  (p=0.008 n=5+5)
Flate            24.6MB ± 0%       24.5MB ± 0%  -0.35%  (p=0.008 n=5+5)
GoParser         30.9MB ± 0%       31.0MB ± 0%    ~     (p=0.421 n=5+5)
Reflect          73.4MB ± 0%       73.4MB ± 0%    ~     (p=0.056 n=5+5)
Tar              25.6MB ± 0%       25.5MB ± 0%  -0.61%  (p=0.008 n=5+5)
XML              40.9MB ± 0%       40.9MB ± 0%    ~     (p=0.841 n=5+5)
[Geo mean]       71.6MB            71.6MB       -0.07%

name        old allocs/op     new allocs/op     delta
Template           394k ± 0%         395k ± 1%    ~     (p=0.151 n=5+5)
Unicode            343k ± 0%         344k ± 0%  +0.38%  (p=0.032 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%    ~     (p=1.000 n=5+5)
Compiler          4.41M ± 0%        4.42M ± 0%    ~     (p=0.151 n=5+5)
SSA               9.79M ± 0%        9.79M ± 0%    ~     (p=0.690 n=5+5)
Flate              238k ± 1%         238k ± 0%    ~     (p=0.151 n=5+5)
GoParser           321k ± 0%         321k ± 1%    ~     (p=0.548 n=5+5)
Reflect            958k ± 0%         957k ± 0%    ~     (p=0.841 n=5+5)
Tar                252k ± 0%         252k ± 1%    ~     (p=0.151 n=5+5)
XML                401k ± 0%         400k ± 0%    ~     (p=1.000 n=5+5)
[Geo mean]         741k              742k       +0.08%

Reduces object files a little bit:

name        old object-bytes  new object-bytes  delta
Template           386k ± 0%         386k ± 0%  -0.04%  (p=0.008 n=5+5)
Unicode            202k ± 0%         202k ± 0%    ~     (all equal)
GoTypes           1.16M ± 0%        1.16M ± 0%  -0.04%  (p=0.008 n=5+5)
Compiler          3.91M ± 0%        3.91M ± 0%  -0.08%  (p=0.008 n=5+5)
SSA               7.91M ± 0%        7.91M ± 0%  -0.04%  (p=0.008 n=5+5)
Flate              228k ± 0%         227k ± 0%  -0.28%  (p=0.008 n=5+5)
GoParser           283k ± 0%         283k ± 0%  -0.01%  (p=0.008 n=5+5)
Reflect            952k ± 0%         951k ± 0%  -0.03%  (p=0.008 n=5+5)
Tar                188k ± 0%         187k ± 0%  -0.09%  (p=0.008 n=5+5)
XML                406k ± 0%         406k ± 0%  -0.04%  (p=0.008 n=5+5)
[Geo mean]         648k              648k       -0.06%

This was discovered in the context for the Fannkuch benchmark.
It shrinks the number of panicindex calls in that function
from 13 back to 9, their 1.8.1 level.

It shrinks the function text a bit, from 829 to 801 bytes.
It slows down execution a little, presumably due to alignment (?).

name          old time/op  new time/op  delta
Fannkuch11-8   2.68s ± 2%   2.74s ± 1%  +2.09%  (p=0.000 n=19+20)

After this CL, 1.8.1 and tip are identical:

name          old time/op  new time/op  delta
Fannkuch11-8   2.74s ± 2%   2.74s ± 1%   ~     (p=0.301 n=20+20)

Fixes #20332

Change-Id: I2aeacc3e8cf2ac1ff10f36c572a27856f4f8f7c9
Reviewed-on: https://go-review.googlesource.com/43291
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: don't use statictmps for SSA-able composite literals
Josh Bleecher Snyder [Wed, 10 May 2017 19:48:17 +0000 (12:48 -0700)]
cmd/compile: don't use statictmps for SSA-able composite literals

The writebarrier test has to change.
Now that T23 composite literals are passed to the backend,
they get SSA'd, so writes to their fields are treated separately,
so the relevant part of the first write to t23 is now a dead store.
Preserve the intent of the test by splitting it up into two functions.

Reduces code size a bit:

name        old object-bytes  new object-bytes  delta
Template           386k ± 0%         386k ± 0%    ~     (all equal)
Unicode            202k ± 0%         202k ± 0%    ~     (all equal)
GoTypes           1.16M ± 0%        1.16M ± 0%    ~     (all equal)
Compiler          3.92M ± 0%        3.91M ± 0%  -0.19%  (p=0.008 n=5+5)
SSA               7.91M ± 0%        7.91M ± 0%    ~     (all equal)
Flate              228k ± 0%         228k ± 0%  -0.05%  (p=0.008 n=5+5)
GoParser           283k ± 0%         283k ± 0%    ~     (all equal)
Reflect            952k ± 0%         952k ± 0%  -0.06%  (p=0.008 n=5+5)
Tar                188k ± 0%         188k ± 0%  -0.09%  (p=0.008 n=5+5)
XML                406k ± 0%         406k ± 0%  -0.02%  (p=0.008 n=5+5)
[Geo mean]         649k              648k       -0.04%

Fixes #18872

Change-Id: Ifeed0f71f13849732999aa731cc2bf40c0f0e32a
Reviewed-on: https://go-review.googlesource.com/43154
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: avoid checkwidth of [...] arrays
Josh Bleecher Snyder [Thu, 11 May 2017 18:09:45 +0000 (11:09 -0700)]
cmd/compile: avoid checkwidth of [...] arrays

Fixes #20333

Change-Id: I0653cc859076f146d8ea8f5bd55cb22b0b8d987f
Reviewed-on: https://go-review.googlesource.com/43290
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agonet/http: for http2, use the priority write scheduler by default
Tom Bergan [Thu, 11 May 2017 17:13:10 +0000 (10:13 -0700)]
net/http: for http2, use the priority write scheduler by default

Updates #18318

Change-Id: Ibd4ebc7708abf87eded8da9661378b5777b8a400
Reviewed-on: https://go-review.googlesource.com/43231
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/internal/obj: continue to optimize ARM's constant pool
Ben Shi [Tue, 2 May 2017 09:29:03 +0000 (09:29 +0000)]
cmd/internal/obj: continue to optimize ARM's constant pool

Both Keith's https://go-review.googlesource.com/c/41612/ and
and Ben's https://go-review.googlesource.com/c/41679/ optimized ARM's
constant pool. But neither was complete.

First, BIC was forgotten.
1. "BIC $0xff00ff00, Reg" can be optimized to
   "BIC $0xff000000, Reg
    BIC $0x0000ff00, Reg"
2. "BIC $0xffff00ff, Reg" can be optimized to
   "AND $0x0000ff00, Reg"
3. "AND $0xffff00ff, Reg" can be optimized to
   "BIC $0x0000ff00, Reg"

Second, break a non-ARMImmRot to the subtraction of two ARMImmRots was
left as TODO.
1. "ADD $0x00fffff0, Reg" can be optimized to
   "ADD $0x01000000, Reg
    SUB $0x00000010, Reg"
2. "SUB $0x00fffff0, Reg" can be optimized to
   "SUB $0x01000000, Reg
    ADD $0x00000010, Reg"

This patch fixes them and issue #19844.

The go1 benchmark shows improvements.

name                     old time/op    new time/op    delta
BinaryTree17-4              41.4s ± 1%     41.7s ± 1%  +0.54%  (p=0.000 n=50+49)
Fannkuch11-4                24.7s ± 1%     25.1s ± 0%  +1.70%  (p=0.000 n=50+49)
FmtFprintfEmpty-4           853ns ± 1%     852ns ± 1%    ~     (p=0.833 n=50+50)
FmtFprintfString-4         1.33µs ± 1%    1.33µs ± 1%    ~     (p=0.163 n=50+50)
FmtFprintfInt-4            1.40µs ± 1%    1.40µs ± 0%    ~     (p=0.293 n=50+35)
FmtFprintfIntInt-4         2.09µs ± 1%    2.08µs ± 1%  -0.39%  (p=0.000 n=50+49)
FmtFprintfPrefixedInt-4    2.43µs ± 1%    2.43µs ± 1%    ~     (p=0.552 n=50+50)
FmtFprintfFloat-4          4.57µs ± 1%    4.42µs ± 1%  -3.18%  (p=0.000 n=50+50)
FmtManyArgs-4              8.62µs ± 1%    8.52µs ± 0%  -1.08%  (p=0.000 n=50+50)
GobDecode-4                 101ms ± 1%     101ms ± 2%  +0.45%  (p=0.001 n=49+49)
GobEncode-4                90.7ms ± 1%    91.1ms ± 2%  +0.51%  (p=0.001 n=50+50)
Gzip-4                      4.23s ± 1%     4.21s ± 1%  -0.62%  (p=0.000 n=50+50)
Gunzip-4                    623ms ± 1%     619ms ± 0%  -0.63%  (p=0.000 n=50+42)
HTTPClientServer-4          721µs ± 5%     683µs ± 3%  -5.25%  (p=0.000 n=50+47)
JSONEncode-4                251ms ± 1%     253ms ± 1%  +0.54%  (p=0.000 n=49+50)
JSONDecode-4                941ms ± 1%     944ms ± 1%  +0.30%  (p=0.001 n=49+50)
Mandelbrot200-4            49.3ms ± 1%    49.3ms ± 0%    ~     (p=0.918 n=50+48)
GoParse-4                  47.1ms ± 1%    47.2ms ± 1%  +0.18%  (p=0.025 n=50+50)
RegexpMatchEasy0_32-4      1.23µs ± 1%    1.24µs ± 1%  +0.30%  (p=0.000 n=49+50)
RegexpMatchEasy0_1K-4      7.74µs ± 7%    7.76µs ± 5%    ~     (p=0.888 n=50+50)
RegexpMatchEasy1_32-4      1.32µs ± 1%    1.32µs ± 1%  +0.23%  (p=0.003 n=50+50)
RegexpMatchEasy1_1K-4      10.6µs ± 2%    10.5µs ± 3%  -1.29%  (p=0.000 n=49+50)
RegexpMatchMedium_32-4     2.19µs ± 1%    2.10µs ± 1%  -3.79%  (p=0.000 n=49+49)
RegexpMatchMedium_1K-4      544µs ± 0%     545µs ± 0%    ~     (p=0.123 n=41+50)
RegexpMatchHard_32-4       28.8µs ± 0%    28.8µs ± 1%    ~     (p=0.580 n=46+50)
RegexpMatchHard_1K-4        863µs ± 1%     865µs ± 1%  +0.31%  (p=0.027 n=47+50)
Revcomp-4                  82.2ms ± 2%    82.3ms ± 2%    ~     (p=0.894 n=48+49)
Template-4                  1.06s ± 1%     1.04s ± 1%  -1.18%  (p=0.000 n=50+49)
TimeParse-4                7.25µs ± 1%    7.35µs ± 0%  +1.48%  (p=0.000 n=50+50)
TimeFormat-4               13.3µs ± 1%    13.2µs ± 1%  -0.13%  (p=0.007 n=50+50)
[Geo mean]                  736µs          733µs       -0.37%

name                     old speed      new speed      delta
GobDecode-4              7.60MB/s ± 1%  7.56MB/s ± 2%  -0.46%  (p=0.001 n=49+49)
GobEncode-4              8.47MB/s ± 1%  8.42MB/s ± 2%  -0.50%  (p=0.001 n=50+50)
Gzip-4                   4.58MB/s ± 1%  4.61MB/s ± 1%  +0.59%  (p=0.000 n=50+50)
Gunzip-4                 31.2MB/s ± 1%  31.4MB/s ± 0%  +0.63%  (p=0.000 n=50+42)
JSONEncode-4             7.73MB/s ± 1%  7.69MB/s ± 1%  -0.53%  (p=0.000 n=49+50)
JSONDecode-4             2.06MB/s ± 1%  2.06MB/s ± 1%    ~     (p=0.052 n=44+50)
GoParse-4                1.23MB/s ± 0%  1.23MB/s ± 2%    ~     (p=0.526 n=26+50)
RegexpMatchEasy0_32-4    25.9MB/s ± 1%  25.9MB/s ± 1%  -0.30%  (p=0.000 n=49+50)
RegexpMatchEasy0_1K-4     132MB/s ± 7%   132MB/s ± 6%    ~     (p=0.885 n=50+50)
RegexpMatchEasy1_32-4    24.2MB/s ± 1%  24.1MB/s ± 1%  -0.22%  (p=0.003 n=50+50)
RegexpMatchEasy1_1K-4    96.4MB/s ± 2%  97.8MB/s ± 3%  +1.36%  (p=0.000 n=50+50)
RegexpMatchMedium_32-4    460kB/s ± 0%   476kB/s ± 1%  +3.43%  (p=0.000 n=49+50)
RegexpMatchMedium_1K-4   1.88MB/s ± 0%  1.88MB/s ± 0%    ~     (all equal)
RegexpMatchHard_32-4     1.11MB/s ± 0%  1.11MB/s ± 1%  +0.34%  (p=0.000 n=45+50)
RegexpMatchHard_1K-4     1.19MB/s ± 1%  1.18MB/s ± 1%  -0.34%  (p=0.033 n=50+50)
Revcomp-4                30.9MB/s ± 2%  30.9MB/s ± 2%    ~     (p=0.894 n=48+49)
Template-4               1.84MB/s ± 1%  1.86MB/s ± 2%  +1.19%  (p=0.000 n=48+50)
[Geo mean]               6.63MB/s       6.65MB/s       +0.26%

Fixes #19844.

Change-Id: I5ad16cc0b29267bb4579aca3dcc10a0b8ade1aa4
Reviewed-on: https://go-review.googlesource.com/42430
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agoreflect: remove dead v.typ assignment
Daniel Martí [Wed, 10 May 2017 16:21:21 +0000 (18:21 +0200)]
reflect: remove dead v.typ assignment

v is not a pointer receiver, and v.typ isn't used in the lines below.
The assignment is dead. Remove it.

Keep the comment, as it refers to the whole case block and not just the
removed line.

Change-Id: Icb2d20c287d9a41bf620ebe5cdec764cd84178a7
Reviewed-on: https://go-review.googlesource.com/43134
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoos: remove extraneous 's' from TestStdPipe error message
Ian Lance Taylor [Wed, 10 May 2017 19:57:55 +0000 (12:57 -0700)]
os: remove extraneous 's' from TestStdPipe error message

Change-Id: I73674a90d88da630b09dcb75ced5d9f0d96c3c09
Reviewed-on: https://go-review.googlesource.com/43153
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: repair copy-pasted method comment
David Chase [Wed, 10 May 2017 17:50:19 +0000 (13:50 -0400)]
cmd/compile: repair copy-pasted method comment

Inadvertent copy-pasta in CL
https://go-review.googlesource.com/c/36207/

Is this allowed for 1.9 post-freeze?

Change-Id: I00dd27f9f53babeded3695346986a7a491f8a771
Reviewed-on: https://go-review.googlesource.com/43092
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: remove unused cpuid_X variables
Martin Möhrmann [Wed, 10 May 2017 17:03:48 +0000 (19:03 +0200)]
runtime: remove unused cpuid_X variables

They are not exported and not used in the compiler or standard library.

Change-Id: Ie1d210464f826742d282f12258ed1792cbd2d188
Reviewed-on: https://go-review.googlesource.com/43135
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/link: set Fieldtrack symbol type to SDATA
Josh Bleecher Snyder [Wed, 10 May 2017 19:05:20 +0000 (12:05 -0700)]
cmd/link: set Fieldtrack symbol type to SDATA

Fixes #20316

Change-Id: I2553ad9a82dfcb30f020c2a3e37b868111425a20
Reviewed-on: https://go-review.googlesource.com/43151
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: reduce debugger-worsening line number churn
David Chase [Thu, 2 Feb 2017 19:51:15 +0000 (14:51 -0500)]
cmd/compile: reduce debugger-worsening line number churn

Reuse block head or preceding instruction's line number for
register allocator's spill, fill, copy, rematerialization
instructionsl; and also for phi, and for no-src-pos
instructions.  Assembler creates same line number tables
for copy-predecessor-line and for no-src-pos,
but copy-predecessor produces better-looking assembly
language output with -S and with GOSSAFUNC, and does not
require changes to tests of existing assembly language.

Split "copyInto" into two cases, one for register allocation,
one for otherwise.  This caused the test score line change
count to increase by one, which may reflect legitimately
useful information preserved.  Without any special treatment
for copyInto, the change count increases by 21 more, from
51 to 72 (i.e., quite a lot).

There is a test; using two naive "scores" for line number
churn, the old numbering is 2x or 4x worse.

Fixes #18902.

Change-Id: I0a0a69659d30ee4e5d10116a0dd2b8c5df8457b1
Reviewed-on: https://go-review.googlesource.com/36207
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agogo/build: accept spaces in cgo directives
Hiroshi Ioka [Wed, 10 May 2017 06:35:33 +0000 (15:35 +0900)]
go/build: accept spaces in cgo directives

Fixes #7906

Change-Id: Ibcf9cd670593241921ab3c426ff7357f799ebc3e
Reviewed-on: https://go-review.googlesource.com/43072
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 agointernal/cpu: new package to detect cpu features
Martin Möhrmann [Mon, 3 Apr 2017 20:38:09 +0000 (22:38 +0200)]
internal/cpu: new package to detect cpu features

Implements detection of x86 cpu features that
are used in the go standard library.

Changes all standard library packages to use the new cpu package
instead of using runtime internal variables to check x86 cpu features.

Updates: #15403

Change-Id: I2999a10cb4d9ec4863ffbed72f4e021a1dbc4bb9
Reviewed-on: https://go-review.googlesource.com/41476
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: add generic rules to eliminate some unnecessary stores
Michael Munday [Wed, 29 Mar 2017 20:37:12 +0000 (16:37 -0400)]
cmd/compile: add generic rules to eliminate some unnecessary stores

Eliminates stores of values that have just been loaded from the same
location. Handles the common case where there are up to 3 intermediate
stores to non-overlapping struct fields.

For example the loads and stores of x.a, x.b and x.d in the following
function are now removed:

type T struct {
a, b, c, d int
}

func f(x *T) {
y := *x
y.c += 8
*x = y
}

Before this CL (s390x):

TEXT    "".f(SB)
MOVD    "".x(R15), R5
MOVD    (R5), R1
MOVD    8(R5), R2
MOVD    16(R5), R0
MOVD    24(R5), R4
ADD     $8, R0, R3
STMG    R1, R4, (R5)
RET

After this CL (s390x):

TEXT "".f(SB)
MOVD "".x(R15), R1
MOVD 16(R1), R0
ADD $8, R0, R0
MOVD R0, 16(R1)
RET

In total these rules are triggered ~5091 times during all.bash,
which is broken down as:

Intermediate stores | Triggered
--------------------+----------
0                   | 1434
1                   | 2508
2                   | 888
3                   | 261
--------------------+----------

Change-Id: Ia4721ae40146aceec1fdd3e65b0e9283770bfba5
Reviewed-on: https://go-review.googlesource.com/38793
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile/internal/ssa: fix generation of ppc64x rules
Michael Munday [Wed, 10 May 2017 15:00:03 +0000 (11:00 -0400)]
cmd/compile/internal/ssa: fix generation of ppc64x rules

The files PPC64.rules and rewritePPC64.go were out of sync due to
conflicts between CL 41630 and CL 42145 (i.e. running 'go run *.go'
in the gen directory resulted in unexpected changes).

Change-Id: I1d409656b66afeab6cb9c6df9b3dcab7859caa75
Reviewed-on: https://go-review.googlesource.com/43091
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
7 years agocmd/link: include DW_AT_producer in .debug_info
David Chase [Mon, 8 May 2017 18:37:02 +0000 (14:37 -0400)]
cmd/link: include DW_AT_producer in .debug_info

This can make life easier for Delve (and other debuggers),
and can help them with bug reports.

Sample producer field (from objdump):
<48> DW_AT_producer : Go cmd/compile devel +8a59dbf41a Mon May 8 16:02:44 2017 -0400

Change-Id: I0605843c959b53a60a25a3b870aa8755bf5d5b13
Reviewed-on: https://go-review.googlesource.com/33588
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoreflect: don't panic in ArrayOf if elem size is 0
Daniel Martí [Wed, 10 May 2017 11:53:39 +0000 (13:53 +0200)]
reflect: don't panic in ArrayOf if elem size is 0

We do a division by the elem type size to check if the array size would
be too large for the virtual address space. This is a silly check if the
size is 0, but the problem is that it means a division by zero and a
panic.

Since arrays of empty structs are valid in a regular program, make them
also work in reflect.

Use a separate, explicit test with struct{}{} to make sure the test for
a zero-sized type is not confused with the rest.

Fixes #20313.

Change-Id: I47b8b87e6541631280b79227bdea6a0f6035c9e0
Reviewed-on: https://go-review.googlesource.com/43131
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: ppc64x intrinsics for math/bits
Lynn Boger [Mon, 24 Apr 2017 19:11:39 +0000 (15:11 -0400)]
cmd/compile: ppc64x intrinsics for math/bits

This adds math/bits intrinsics for OnesCount, Len, TrailingZeros on
ppc64x.

benchmark                       old ns/op     new ns/op     delta
BenchmarkLeadingZeros-16        4.26          1.71          -59.86%
BenchmarkLeadingZeros16-16      3.04          1.83          -39.80%
BenchmarkLeadingZeros32-16      3.31          1.82          -45.02%
BenchmarkLeadingZeros64-16      3.69          1.71          -53.66%
BenchmarkTrailingZeros-16       2.55          1.62          -36.47%
BenchmarkTrailingZeros32-16     2.55          1.77          -30.59%
BenchmarkTrailingZeros64-16     2.78          1.62          -41.73%
BenchmarkOnesCount-16           3.19          0.93          -70.85%
BenchmarkOnesCount32-16         2.55          1.18          -53.73%
BenchmarkOnesCount64-16         3.22          0.93          -71.12%

Update #18616

I also made a change to bits_test.go because when debugging some failures
the output was not quite providing the right argument information.

Change-Id: Ia58d31d1777cf4582a4505f85b11a1202ca07d3e
Reviewed-on: https://go-review.googlesource.com/41630
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoreflect: fix String of new array types
Daniel Martí [Wed, 10 May 2017 11:10:46 +0000 (13:10 +0200)]
reflect: fix String of new array types

When constructing a new type for an array type in ArrayOf, we don't
reset tflag to 0. All the other methods in the package, such as SliceOf,
do this already. This results in the new array type having weird issues
when being printed, such as having tflagExtraStar set when it shouldn't.

That flag removes the first char to get rid of '*', but when used
incorrectly in this case it eats the '[' character leading to broken
strings like "3]int".

This was fixed in 56752eb2 for issue #16722, but ArrayOf was missed.

Also make the XM test struct have a non-zero size as that leads to a
division by zero panic in ArrayOf.

Fixes #20311.

Change-Id: I18f1027fdbe9f71767201e7424269c3ceeb23eb5
Reviewed-on: https://go-review.googlesource.com/43130
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile/internal/gc: rename signatlist to signatset
Marvin Stenger [Wed, 26 Apr 2017 23:48:26 +0000 (01:48 +0200)]
cmd/compile/internal/gc: rename signatlist to signatset

Also change type from map[*types.Type]bool to map[*types.Type]struct{}.
This is basically a clean-up.

Change-Id: I167583eff0fa1070a7522647219476033b52b840
Reviewed-on: https://go-review.googlesource.com/41859
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: use a buffered channel for the function queue
Josh Bleecher Snyder [Wed, 10 May 2017 00:36:42 +0000 (17:36 -0700)]
cmd/compile: use a buffered channel for the function queue

Updates #20307

With -c=2:

name        old time/op       new time/op       delta
Template          140ms ± 3%        139ms ± 4%  -1.06%  (p=0.003 n=50+50)
Unicode          81.1ms ± 4%       81.9ms ± 4%  +0.96%  (p=0.006 n=50+49)
GoTypes           375ms ± 3%        374ms ± 3%    ~     (p=0.094 n=48+48)
Compiler          1.69s ± 2%        1.68s ± 2%  -0.41%  (p=0.004 n=49+48)
SSA               3.05s ± 1%        3.05s ± 2%    ~     (p=0.953 n=47+49)
Flate            86.3ms ± 2%       85.9ms ± 2%  -0.49%  (p=0.011 n=49+48)
GoParser         99.5ms ± 3%       99.3ms ± 3%    ~     (p=0.394 n=48+49)
Reflect           262ms ± 3%        261ms ± 3%    ~     (p=0.354 n=47+49)
Tar              81.4ms ± 3%       79.7ms ± 4%  -1.98%  (p=0.000 n=47+50)
XML               133ms ± 3%        133ms ± 3%    ~     (p=0.992 n=50+49)
[Geo mean]        236ms             235ms       -0.36%

name        old user-time/op  new user-time/op  delta
Template          249ms ± 5%        242ms ± 7%  -2.61%  (p=0.000 n=48+50)
Unicode           111ms ± 4%        111ms ± 6%    ~     (p=0.407 n=46+47)
GoTypes           753ms ± 2%        748ms ± 3%  -0.65%  (p=0.010 n=48+50)
Compiler          3.28s ± 2%        3.27s ± 2%  -0.40%  (p=0.026 n=49+47)
SSA               7.03s ± 2%        7.01s ± 3%    ~     (p=0.154 n=45+50)
Flate             154ms ± 3%        154ms ± 3%    ~     (p=0.306 n=49+49)
GoParser          180ms ± 4%        179ms ± 4%    ~     (p=0.148 n=48+48)
Reflect           427ms ± 2%        428ms ± 3%    ~     (p=0.502 n=46+49)
Tar               142ms ± 5%        135ms ± 9%  -4.83%  (p=0.000 n=46+50)
XML               247ms ± 3%        247ms ± 4%    ~     (p=0.921 n=49+49)
[Geo mean]        426ms             422ms       -0.92%

Change-Id: I4746234439ddb9a7e5840fc783b8857da6a4a680
Reviewed-on: https://go-review.googlesource.com/43110
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: allow OpVarXXX calls to be duplicated in writebarrier blocks
Josh Bleecher Snyder [Thu, 27 Apr 2017 18:56:46 +0000 (11:56 -0700)]
cmd/compile: allow OpVarXXX calls to be duplicated in writebarrier blocks

OpVarXXX Values don't generate instructions,
so there's no reason not to duplicate them,
and duplicating them generates better code
(fewer branches).

This requires changing the start/end accounting
to correctly handle the case in which we have run
of Values beginning with an OpVarXXX, e.g.
OpVarDef, OpZeroWB, OpMoveWB.
In that case, the sequence of values should begin
at the OpZeroWB, not the OpVarDef.

This also lays the groundwork for experimenting
with allowing duplication of some scalar stores.

Shrinks function text sizes a tiny amount:

name        old object-bytes  new object-bytes  delta
Template           381k ± 0%         381k ± 0%  -0.01%  (p=0.008 n=5+5)
Unicode            203k ± 0%         203k ± 0%  -0.04%  (p=0.008 n=5+5)
GoTypes           1.17M ± 0%        1.17M ± 0%  -0.01%  (p=0.008 n=5+5)
SSA               8.24M ± 0%        8.24M ± 0%  -0.00%  (p=0.008 n=5+5)
Flate              230k ± 0%         230k ± 0%    ~     (all equal)
GoParser           286k ± 0%         286k ± 0%    ~     (all equal)
Reflect           1.00M ± 0%        1.00M ± 0%    ~     (all equal)
Tar                189k ± 0%         189k ± 0%    ~     (all equal)
XML                415k ± 0%         415k ± 0%  -0.01%  (p=0.008 n=5+5)

Updates #19838

Change-Id: Ic5ef30855919f1468066eba08ae5c4bd9a01db27
Reviewed-on: https://go-review.googlesource.com/42011
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/internal/obj, cmd/link: fix st_other field on PPC64
Ian Lance Taylor [Tue, 9 May 2017 21:34:16 +0000 (14:34 -0700)]
cmd/internal/obj, cmd/link: fix st_other field on PPC64

In PPC64 ELF files, the st_other field indicates the number of
prologue instructions between the global and local entry points.
We add the instructions in the compiler and assembler if -shared is used.
We were assuming that the instructions were present when building a
c-archive or PIE or doing dynamic linking, on the assumption that those
are the cases where the go tool would be building with -shared.
That assumption fails when using some other tool, such as Bazel,
that does not necessarily use -shared in exactly the same way.

This CL records in the object file whether a symbol was compiled
with -shared (this will be the same for all symbols in a given compilation)
and uses that information when setting the st_other field.

Fixes #20290.

Change-Id: Ib2b77e16aef38824871102e3c244fcf04a86c6ea
Reviewed-on: https://go-review.googlesource.com/43051
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: ignore types when considering tuple select for CSE
Todd Neal [Mon, 24 Apr 2017 15:20:09 +0000 (11:20 -0400)]
cmd/compile: ignore types when considering tuple select for CSE

Fixes #20097

Change-Id: I3c9626ccc8cd0c46a7081ea8650b2ff07a5d4fcd
Reviewed-on: https://go-review.googlesource.com/41505
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: change ssa.Type into *types.Type
Josh Bleecher Snyder [Fri, 28 Apr 2017 21:12:28 +0000 (14:12 -0700)]
cmd/compile: change ssa.Type into *types.Type

When package ssa was created, Type was in package gc.
To avoid circular dependencies, we used an interface (ssa.Type)
to represent type information in SSA.

In the Go 1.9 cycle, gri extricated the Type type from package gc.
As a result, we can now use it in package ssa.
Now, instead of package types depending on package ssa,
it is the other way.
This is a more sensible dependency tree,
and helps compiler performance a bit.

Though this is a big CL, most of the changes are
mechanical and uninteresting.

Interesting bits:

* Add new singleton globals to package types for the special
  SSA types Memory, Void, Invalid, Flags, and Int128.
* Add two new Types, TSSA for the special types,
  and TTUPLE, for SSA tuple types.
  ssa.MakeTuple is now types.NewTuple.
* Move type comparison result constants CMPlt, CMPeq, and CMPgt
  to package types.
* We had picked the name "types" in our rules for the handy
  list of types provided by ssa.Config. That conflicted with
  the types package name, so change it to "typ".
* Update the type comparison routine to handle tuples and special
  types inline.
* Teach gc/fmt.go how to print special types.
* We can now eliminate ElemTypes in favor of just Elem,
  and probably also some other duplicated Type methods
  designed to return ssa.Type instead of *types.Type.
* The ssa tests were using their own dummy types,
  and they were not particularly careful about types in general.
  Of necessity, this CL switches them to use *types.Type;
  it does not make them more type-accurate.
  Unfortunately, using types.Type means initializing a bit
  of the types universe.
  This is prime for refactoring and improvement.

This shrinks ssa.Value; it now fits in a smaller size class
on 64 bit systems. This doesn't have a giant impact,
though, since most Values are preallocated in a chunk.

name        old alloc/op      new alloc/op      delta
Template         37.9MB ± 0%       37.7MB ± 0%  -0.57%  (p=0.000 n=10+8)
Unicode          28.9MB ± 0%       28.7MB ± 0%  -0.52%  (p=0.000 n=10+10)
GoTypes           110MB ± 0%        109MB ± 0%  -0.88%  (p=0.000 n=10+10)
Flate            24.7MB ± 0%       24.6MB ± 0%  -0.66%  (p=0.000 n=10+10)
GoParser         31.1MB ± 0%       30.9MB ± 0%  -0.61%  (p=0.000 n=10+9)
Reflect          73.9MB ± 0%       73.4MB ± 0%  -0.62%  (p=0.000 n=10+8)
Tar              25.8MB ± 0%       25.6MB ± 0%  -0.77%  (p=0.000 n=9+10)
XML              41.2MB ± 0%       40.9MB ± 0%  -0.80%  (p=0.000 n=10+10)
[Geo mean]       40.5MB            40.3MB       -0.68%

name        old allocs/op     new allocs/op     delta
Template           385k ± 0%         386k ± 0%    ~     (p=0.356 n=10+9)
Unicode            343k ± 1%         344k ± 0%    ~     (p=0.481 n=10+10)
GoTypes           1.16M ± 0%        1.16M ± 0%  -0.16%  (p=0.004 n=10+10)
Flate              238k ± 1%         238k ± 1%    ~     (p=0.853 n=10+10)
GoParser           320k ± 0%         320k ± 0%    ~     (p=0.720 n=10+9)
Reflect            957k ± 0%         957k ± 0%    ~     (p=0.460 n=10+8)
Tar                252k ± 0%         252k ± 0%    ~     (p=0.133 n=9+10)
XML                400k ± 0%         400k ± 0%    ~     (p=0.796 n=10+10)
[Geo mean]         428k              428k       -0.01%

Removing all the interface calls helps non-trivially with CPU, though.

name        old time/op       new time/op       delta
Template          178ms ± 4%        173ms ± 3%  -2.90%  (p=0.000 n=94+96)
Unicode          85.0ms ± 4%       83.9ms ± 4%  -1.23%  (p=0.000 n=96+96)
GoTypes           543ms ± 3%        528ms ± 3%  -2.73%  (p=0.000 n=98+96)
Flate             116ms ± 3%        113ms ± 4%  -2.34%  (p=0.000 n=96+99)
GoParser          144ms ± 3%        140ms ± 4%  -2.80%  (p=0.000 n=99+97)
Reflect           344ms ± 3%        334ms ± 4%  -3.02%  (p=0.000 n=100+99)
Tar               106ms ± 5%        103ms ± 4%  -3.30%  (p=0.000 n=98+94)
XML               198ms ± 5%        192ms ± 4%  -2.88%  (p=0.000 n=92+95)
[Geo mean]        178ms             173ms       -2.65%

name        old user-time/op  new user-time/op  delta
Template          229ms ± 5%        224ms ± 5%  -2.36%  (p=0.000 n=95+99)
Unicode           107ms ± 6%        106ms ± 5%  -1.13%  (p=0.001 n=93+95)
GoTypes           696ms ± 4%        679ms ± 4%  -2.45%  (p=0.000 n=97+99)
Flate             137ms ± 4%        134ms ± 5%  -2.66%  (p=0.000 n=99+96)
GoParser          176ms ± 5%        172ms ± 8%  -2.27%  (p=0.000 n=98+100)
Reflect           430ms ± 6%        411ms ± 5%  -4.46%  (p=0.000 n=100+92)
Tar               128ms ±13%        123ms ±13%  -4.21%  (p=0.000 n=100+100)
XML               239ms ± 6%        233ms ± 6%  -2.50%  (p=0.000 n=95+97)
[Geo mean]        220ms             213ms       -2.76%

Change-Id: I15c7d6268347f8358e75066dfdbd77db24e8d0c1
Reviewed-on: https://go-review.googlesource.com/42145
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: add boolean simplification rules
Josh Bleecher Snyder [Tue, 2 May 2017 23:43:27 +0000 (16:43 -0700)]
cmd/compile: add boolean simplification rules

These collectively fire a few hundred times during make.bash,
mostly rewriting XOR SETNE -> SETEQ.

Fixes #17905.

Change-Id: Ic5eb241ee93ed67099da3de11f59e4df9fab64a3
Reviewed-on: https://go-review.googlesource.com/42491
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile/internal/ssa: mark boolean instructions commutative
Marvin Stenger [Wed, 3 May 2017 11:33:14 +0000 (13:33 +0200)]
cmd/compile/internal/ssa: mark boolean instructions commutative

Mark AndB, OrB, EqB, and NeqB as commutative.

Change-Id: Ife7cfcb9780cc5dd669617cb52339ab336667da4
Reviewed-on: https://go-review.googlesource.com/42515
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: make builds reproducible in presence of **byte and **int8
Josh Bleecher Snyder [Sun, 7 May 2017 06:19:41 +0000 (23:19 -0700)]
cmd/compile: make builds reproducible in presence of **byte and **int8

CL 39915 introduced sorting of signats by ShortString
for reproducible builds. But ShortString treats types
byte and uint8 identically; same for rune and uint32.
CL 39915 attempted to compensate for this by only
adding the underlying type (uint8) to signats in addsignat.

This only works for byte and uint8. For e.g. *byte and *uint,
both get added, and their sort order is random,
leading to non-reproducible builds.

One fix would be to add yet another type printing mode
that doesn't eliminate byte and rune, and use it
for sorting signats. But the formatting routines
are complicated enough as it is.

Instead, just sort first by ShortString and then by String.
We can't just use String, because ShortString makes distinctions
that String doesn't. ShortString is really preferred here;
String is serving only as a backstop for handling of bytes and runes.

The long series of types in the test helps increase the odds of
failure, allowing a smaller number of iterations in the test.
On my machine, a full test takes 700ms.

Passes toolstash-check.

Updates #19961
Fixes #20272

name        old alloc/op      new alloc/op      delta
Template         37.9MB ± 0%       37.9MB ± 0%  +0.12%  (p=0.032 n=5+5)
Unicode          28.9MB ± 0%       28.9MB ± 0%    ~     (p=0.841 n=5+5)
GoTypes           110MB ± 0%        110MB ± 0%    ~     (p=0.841 n=5+5)
Compiler          463MB ± 0%        463MB ± 0%    ~     (p=0.056 n=5+5)
SSA              1.11GB ± 0%       1.11GB ± 0%  +0.02%  (p=0.016 n=5+5)
Flate            24.7MB ± 0%       24.8MB ± 0%  +0.14%  (p=0.032 n=5+5)
GoParser         31.1MB ± 0%       31.1MB ± 0%    ~     (p=0.421 n=5+5)
Reflect          73.9MB ± 0%       73.9MB ± 0%    ~     (p=1.000 n=5+5)
Tar              25.8MB ± 0%       25.8MB ± 0%  +0.15%  (p=0.016 n=5+5)
XML              41.2MB ± 0%       41.2MB ± 0%    ~     (p=0.310 n=5+5)
[Geo mean]       72.0MB            72.0MB       +0.07%

name        old allocs/op     new allocs/op     delta
Template           384k ± 0%         385k ± 1%    ~     (p=0.056 n=5+5)
Unicode            343k ± 0%         344k ± 0%    ~     (p=0.548 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%    ~     (p=0.421 n=5+5)
Compiler          4.43M ± 0%        4.44M ± 0%  +0.26%  (p=0.032 n=5+5)
SSA               9.86M ± 0%        9.87M ± 0%  +0.10%  (p=0.032 n=5+5)
Flate              237k ± 1%         238k ± 0%  +0.49%  (p=0.032 n=5+5)
GoParser           319k ± 1%         320k ± 1%    ~     (p=0.151 n=5+5)
Reflect            957k ± 0%         957k ± 0%    ~     (p=1.000 n=5+5)
Tar                251k ± 0%         252k ± 1%  +0.49%  (p=0.016 n=5+5)
XML                399k ± 0%         401k ± 1%    ~     (p=0.310 n=5+5)
[Geo mean]         739k              741k       +0.26%

Change-Id: Ic27995a8d374d012b8aca14546b1df9d28d30df7
Reviewed-on: https://go-review.googlesource.com/42955
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: make "imported and not used" errors deterministic
Josh Bleecher Snyder [Tue, 9 May 2017 15:22:43 +0000 (08:22 -0700)]
cmd/compile: make "imported and not used" errors deterministic

If there were more unused imports than
the maximum default number of errors to report,
the set of reported imports was non-deterministic.

Fix by accumulating and sorting them prior to output.

Fixes #20298

Change-Id: Ib3d5a15fd7dc40009523fcdc1b93ddc62a1b05f2
Reviewed-on: https://go-review.googlesource.com/42954
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/internal/obj/arm64, cmd/compile: improve offset folding on ARM64
Cherry Zhang [Fri, 28 Apr 2017 22:02:00 +0000 (18:02 -0400)]
cmd/internal/obj/arm64, cmd/compile: improve offset folding on ARM64

ARM64 assembler backend only accepts loads and stores with small
or aligned offset. The compiler therefore can only fold small or
aligned offsets into loads and stores. For locals and args, their
offsets to SP are not known until very late, and the compiler
makes conservative decision not folding some of them. However,
in most cases, the offset is indeed small or aligned, and can
be folded into load and store (but actually not).

This CL adds support of loads and stores with large and unaligned
offsets. When the offset doesn't fit into the instruction, it
uses two instructions and (for very large offset) the constant
pool. This way, the compiler doesn't need to be conservative,
and can simply fold the offset.

To make it work, the assembler's optab matching rules need to be
changed. Before, MOVD accepts C_UAUTO32K which matches multiple
of 8 between 0 and 32K, and also C_UAUTO16K, which may not be
multiple of 8 and does not fit into MOVD instruction. The
assembler errors in the latter case. This change makes it only
matches multiple of 8 (or offsets within ±256, which also fits
in instruction), and uses the large-or-unaligned-offset rule
for things doesn't fit (without error). Other sized move rules
are changed similarly.

Class C_UAUTO64K and C_UOREG64K are removed, as they are never
used.

In shared library, load/store of global is rewritten to using
GOT and temp register, which conflicts with the use of temp
register for assembling large offset. So the folding is disabled
for globals in shared library mode.

Reduce cmd/go binary size by 2%.

name                     old time/op    new time/op    delta
BinaryTree17-8              8.67s ± 0%     8.61s ± 0%   -0.60%  (p=0.000 n=9+10)
Fannkuch11-8                6.24s ± 0%     6.19s ± 0%   -0.83%  (p=0.000 n=10+9)
FmtFprintfEmpty-8           116ns ± 0%     116ns ± 0%     ~     (all equal)
FmtFprintfString-8          196ns ± 0%     192ns ± 0%   -1.89%  (p=0.000 n=10+10)
FmtFprintfInt-8             199ns ± 0%     198ns ± 0%   -0.35%  (p=0.001 n=9+10)
FmtFprintfIntInt-8          294ns ± 0%     293ns ± 0%   -0.34%  (p=0.000 n=8+8)
FmtFprintfPrefixedInt-8     318ns ± 1%     318ns ± 1%     ~     (p=1.000 n=10+10)
FmtFprintfFloat-8           537ns ± 0%     531ns ± 0%   -1.17%  (p=0.000 n=9+10)
FmtManyArgs-8              1.19µs ± 1%    1.18µs ± 1%   -1.41%  (p=0.001 n=10+10)
GobDecode-8                17.2ms ± 1%    17.3ms ± 2%     ~     (p=0.165 n=10+10)
GobEncode-8                14.7ms ± 1%    14.7ms ± 2%     ~     (p=0.631 n=10+10)
Gzip-8                      837ms ± 0%     836ms ± 0%   -0.14%  (p=0.006 n=9+10)
Gunzip-8                    141ms ± 0%     139ms ± 0%   -1.24%  (p=0.000 n=9+10)
HTTPClientServer-8          256µs ± 1%     253µs ± 1%   -1.35%  (p=0.000 n=10+10)
JSONEncode-8               40.1ms ± 1%    41.3ms ± 1%   +3.06%  (p=0.000 n=10+9)
JSONDecode-8                157ms ± 1%     156ms ± 1%   -0.83%  (p=0.001 n=9+8)
Mandelbrot200-8            8.94ms ± 0%    8.94ms ± 0%   +0.02%  (p=0.000 n=9+9)
GoParse-8                  8.69ms ± 0%    8.54ms ± 1%   -1.69%  (p=0.000 n=8+10)
RegexpMatchEasy0_32-8       227ns ± 1%     228ns ± 1%   +0.48%  (p=0.016 n=10+9)
RegexpMatchEasy0_1K-8      1.92µs ± 0%    1.63µs ± 0%  -15.08%  (p=0.000 n=10+9)
RegexpMatchEasy1_32-8       256ns ± 0%     251ns ± 0%   -2.19%  (p=0.000 n=10+9)
RegexpMatchEasy1_1K-8      2.38µs ± 0%    2.09µs ± 0%  -12.49%  (p=0.000 n=10+9)
RegexpMatchMedium_32-8      352ns ± 0%     354ns ± 0%   +0.39%  (p=0.002 n=10+9)
RegexpMatchMedium_1K-8      106µs ± 0%     106µs ± 0%   -0.05%  (p=0.005 n=10+9)
RegexpMatchHard_32-8       5.92µs ± 0%    5.89µs ± 0%   -0.40%  (p=0.000 n=9+8)
RegexpMatchHard_1K-8        180µs ± 0%     179µs ± 0%   -0.14%  (p=0.000 n=10+9)
Revcomp-8                   1.20s ± 0%     1.13s ± 0%   -6.29%  (p=0.000 n=9+8)
Template-8                  159ms ± 1%     154ms ± 1%   -3.14%  (p=0.000 n=9+10)
TimeParse-8                 800ns ± 3%     769ns ± 1%   -3.91%  (p=0.000 n=10+10)
TimeFormat-8                826ns ± 2%     817ns ± 2%   -1.04%  (p=0.050 n=10+10)
[Geo mean]                  145µs          143µs        -1.79%

Change-Id: I5fc42087cee9b54ea414f8ef6d6d020b80eb5985
Reviewed-on: https://go-review.googlesource.com/42172
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/go: enable concurrent backend compilation by default
Josh Bleecher Snyder [Wed, 26 Apr 2017 21:05:06 +0000 (14:05 -0700)]
cmd/go: enable concurrent backend compilation by default

It can be disabled by setting the environment variable
GO19CONCURRENTCOMPILATION=0, or with -gcflags=-c=1.

Fixes #15756.

Change-Id: I7acbf16330512b62ee14ecbab1f46b53ec5a67b6
Reviewed-on: https://go-review.googlesource.com/41820
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/go: add support for concurrent backend compilation
Josh Bleecher Snyder [Thu, 20 Apr 2017 23:41:42 +0000 (16:41 -0700)]
cmd/go: add support for concurrent backend compilation

It is disabled by default.
It can be enabled by setting the environment variable
GO19CONCURRENTCOMPILATION=1.

Benchmarking results are presented in a grid.
Columns are different values of c (compiler backend concurrency);
rows are different values of p (process concurrency).

'go build -a std cmd', a 4 core raspberry pi 3:

            c=1        c=2        c=4
StdCmd/p=1  504s ± 2%  413s ± 4%  367s ± 3%
StdCmd/p=2  314s ± 3%  266s ± 4%  267s ± 4%
StdCmd/p=4  254s ± 5%  241s ± 5%  238s ± 6%

'go build -a std cmd', an 8 core darwin/amd64 laptop:

            c=1         c=2         c=4         c=6         c=8
StdCmd/p=1  40.4s ± 7%  31.0s ± 1%  27.3s ± 1%  27.8s ± 0%  27.7s ± 0%
StdCmd/p=2  21.9s ± 1%  17.9s ± 1%  16.9s ± 1%  17.0s ± 1%  17.2s ± 0%
StdCmd/p=4  17.4s ± 2%  14.5s ± 2%  13.3s ± 2%  13.5s ± 2%  13.6s ± 2%
StdCmd/p=6  16.9s ± 1%  14.2s ± 2%  13.1s ± 2%  13.2s ± 2%  13.3s ± 2%
StdCmd/p=8  16.7s ± 2%  14.2s ± 2%  13.2s ± 3%  13.2s ± 2%  13.4s ± 2%

'go build -a std cmd', a 96 core arm64 server:

             c=1         c=2         c=4         c=6         c=8         c=16        c=32        c=64        c=96
StdCmd/p=1    173s ± 1%   133s ± 1%   114s ± 1%   109s ± 1%   106s ± 0%   106s ± 1%   107s ± 1%   110s ± 1%   113s ± 1%
StdCmd/p=2   94.2s ± 2%  71.5s ± 1%  61.7s ± 1%  58.7s ± 1%  57.5s ± 2%  56.9s ± 1%  58.0s ± 1%  59.6s ± 1%  61.0s ± 1%
StdCmd/p=4   74.1s ± 2%  53.5s ± 1%  43.7s ± 2%  40.5s ± 1%  39.2s ± 2%  38.9s ± 2%  39.5s ± 3%  40.3s ± 2%  40.8s ± 1%
StdCmd/p=6   69.3s ± 1%  50.2s ± 2%  40.3s ± 2%  37.3s ± 3%  36.0s ± 3%  35.3s ± 2%  36.0s ± 2%  36.8s ± 2%  37.5s ± 2%
StdCmd/p=8   66.1s ± 2%  47.7s ± 2%  38.6s ± 2%  35.7s ± 2%  34.4s ± 1%  33.6s ± 2%  34.2s ± 2%  34.6s ± 1%  35.0s ± 1%
StdCmd/p=16  63.4s ± 2%  45.3s ± 2%  36.3s ± 2%  33.3s ± 2%  32.0s ± 3%  31.6s ± 2%  32.1s ± 2%  32.5s ± 2%  32.7s ± 2%
StdCmd/p=32  62.2s ± 1%  44.2s ± 2%  35.3s ± 2%  32.4s ± 2%  31.2s ± 2%  30.9s ± 2%  31.1s ± 2%  31.7s ± 2%  32.0s ± 2%
StdCmd/p=64  62.2s ± 1%  44.3s ± 2%  35.4s ± 2%  32.4s ± 2%  31.2s ± 2%  30.9s ± 2%  31.2s ± 2%  31.8s ± 3%  32.2s ± 3%
StdCmd/p=96  62.2s ± 2%  44.4s ± 2%  35.3s ± 2%  32.3s ± 2%  31.1s ± 2%  30.9s ± 3%  31.3s ± 2%  31.7s ± 1%  32.1s ± 2%

benchjuju, an 8 core darwin/amd64 laptop:

               c=1         c=2         c=4         c=6         c=8
BuildJuju/p=1  55.3s ± 0%  46.3s ± 0%  41.9s ± 0%  41.4s ± 1%  41.3s ± 0%
BuildJuju/p=2  33.7s ± 1%  28.4s ± 1%  26.7s ± 1%  26.6s ± 1%  26.8s ± 1%
BuildJuju/p=4  24.7s ± 1%  22.3s ± 1%  21.4s ± 1%  21.7s ± 1%  21.8s ± 1%
BuildJuju/p=6  20.6s ± 1%  19.3s ± 2%  19.4s ± 1%  19.7s ± 1%  19.9s ± 1%
BuildJuju/p=8  20.6s ± 2%  19.5s ± 2%  19.3s ± 2%  19.6s ± 1%  19.8s ± 2%

Updates #15756

Change-Id: I8a56e88953071a05eee764002024c54cd888a56c
Reviewed-on: https://go-review.googlesource.com/41819
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agospec: clarify unsafe.Pointer conversions
Robert Griesemer [Fri, 28 Apr 2017 17:32:31 +0000 (10:32 -0700)]
spec: clarify unsafe.Pointer conversions

A pointer type of underlying type unsafe.Pointer can be used in
unsafe conversions. Document unfortunate status quo.

Fixes #19306.

Change-Id: I28172508a200561f8df366bbf2c2807ef3b48c97
Reviewed-on: https://go-review.googlesource.com/42132
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agogo/token: remove excess parenthesis in NoPos.IsValid() documentation
Ibrahim AshShohail [Tue, 9 May 2017 14:54:26 +0000 (17:54 +0300)]
go/token: remove excess parenthesis in NoPos.IsValid() documentation

Fixes #20294

Change-Id: I32ac862fe00180210a04103cc94c4d9fef5d1b6c
Reviewed-on: https://go-review.googlesource.com/42992
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime/pprof: deflake TestGoroutineCounts
Austin Clements [Tue, 9 May 2017 13:42:16 +0000 (09:42 -0400)]
runtime/pprof: deflake TestGoroutineCounts

TestGoroutineCounts currently depends on timing to get 100 goroutines
to a known blocking point before taking a profile. This fails
frequently, with different goroutines captured at different stacks.
The test is disabled on openbsd because it was too flaky, but in fact
it flakes on all platforms.

Fix this by using Gosched instead of timing. This is both much more
reliable and makes the test run faster.

Fixes #15156.

Change-Id: Ia6e894196d717655b8fb4ee96df53f6cc8bc5f1f
Reviewed-on: https://go-review.googlesource.com/42953
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/go: put user flags after code generation flag
Ian Lance Taylor [Tue, 9 May 2017 13:40:04 +0000 (06:40 -0700)]
cmd/go: put user flags after code generation flag

This permits the user to override the code generation flag when they
know better. This is always a good policy for all flags automatically
inserted by the build system.

Doing this now so that I can write a test for #20290.

Update #20290

Change-Id: I5c6708a277238d571b8d037993a5a59e2a442e98
Reviewed-on: https://go-review.googlesource.com/42952
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agonet: fix ExampleParseCIDR IPv4 prefix length
Rob Phoenix [Tue, 9 May 2017 09:26:08 +0000 (10:26 +0100)]
net: fix ExampleParseCIDR IPv4 prefix length

Issue #15228 describes that reserved address blocks should be used for
documentation purposes. This change updates the prefix length so the
IPv4 address adheres to this.

Change-Id: I237d9cce1a71f4fd95f927ec894ce53fa806047f
Reviewed-on: https://go-review.googlesource.com/42991
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/go: run tests that require symlinks
Alex Brainman [Mon, 8 May 2017 06:40:49 +0000 (16:40 +1000)]
cmd/go: run tests that require symlinks

Change-Id: I19a724ea4eb1ba0ff558721650c89a949e53b7c7
Reviewed-on: https://go-review.googlesource.com/42895
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoos: avoid calulating fileStat.path until it is needed
Alex Brainman [Mon, 8 May 2017 02:49:04 +0000 (12:49 +1000)]
os: avoid calulating fileStat.path until it is needed

This CL improves

on my Windows 7

name         old time/op    new time/op    delta
Readdirname    58.1µs ± 1%    58.1µs ± 0%     ~     (p=0.817 n=8+8)
Readdir        58.0µs ± 3%    57.8µs ± 0%     ~     (p=0.944 n=9+8)

name         old alloc/op   new alloc/op   delta
Readdirname    3.03kB ± 0%    2.84kB ± 0%   -6.33%  (p=0.000 n=10+10)
Readdir        3.00kB ± 0%    2.81kB ± 0%   -6.40%  (p=0.000 n=10+10)

name         old allocs/op  new allocs/op  delta
Readdirname      34.0 ± 0%      30.0 ± 0%  -11.76%  (p=0.000 n=10+10)
Readdir          33.0 ± 0%      29.0 ± 0%  -12.12%  (p=0.000 n=10+10)

on my Windows XP

name           old time/op    new time/op    delta
Readdirname-2    85.5µs ± 0%    84.0µs ± 0%   -1.83%  (p=0.000 n=10+10)
Readdir-2        84.6µs ± 0%    83.5µs ± 0%   -1.31%  (p=0.000 n=10+9)

name           old alloc/op   new alloc/op   delta
Readdirname-2    6.52kB ± 0%    5.66kB ± 0%  -13.25%  (p=0.000 n=10+10)
Readdir-2        6.39kB ± 0%    5.53kB ± 0%  -13.52%  (p=0.000 n=10+10)

name           old allocs/op  new allocs/op  delta
Readdirname-2      78.0 ± 0%      66.0 ± 0%  -15.38%  (p=0.000 n=10+10)
Readdir-2          77.0 ± 0%      65.0 ± 0%  -15.58%  (p=0.000 n=10+10)

Change-Id: I5d698eca86b8e94a46b6cfbd5947898b7b3fbdbd
Reviewed-on: https://go-review.googlesource.com/42894
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocontainer/heap: optimization when selecting smaller child
ltnwgl [Fri, 24 Mar 2017 03:55:22 +0000 (11:55 +0800)]
container/heap: optimization when selecting smaller child

In down(), if two children are equal, we can choose either one.
Inspired by https://codereview.appspot.com/6613064/

Change-Id: Iaad4ca5e2f5111bf3abb87f606584e7d274c620b
Reviewed-on: https://go-review.googlesource.com/38612
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agonet: add examples for IPv4, ParseCIDR & IPv4Mask
Rob Phoenix [Mon, 8 May 2017 21:08:42 +0000 (22:08 +0100)]
net: add examples for IPv4, ParseCIDR & IPv4Mask

Further examples to support the net package.

See issue #5757

Change-Id: I839fd97a468c8d9195e8f4a0ee886ba50ca3f382
Reviewed-on: https://go-review.googlesource.com/42912
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: better errors for float constants with large exponents
Robert Griesemer [Mon, 8 May 2017 20:32:18 +0000 (13:32 -0700)]
cmd/compile: better errors for float constants with large exponents

Also: Removed misleading comment.

Fixes #20232.

Change-Id: I0b141b1360ac53267b7ebfcec7a2e2a238f3f46c
Reviewed-on: https://go-review.googlesource.com/42930
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agomath: use SIMD to accelerate additional scalar math functions on s390x
Bill O'Farrell [Fri, 24 Mar 2017 20:43:02 +0000 (16:43 -0400)]
math: use SIMD to accelerate additional scalar math functions on s390x

As necessary, math functions were structured to use stubs, so that they can
be accelerated with assembly on any platform.

Technique used was minimax polynomial approximation using tables of
polynomial coefficients, with argument range reduction.

Benchmark         New     Old     Speedup
BenchmarkAcos     12.2    47.5    3.89
BenchmarkAcosh    18.5    56.2    3.04
BenchmarkAsin     13.1    40.6    3.10
BenchmarkAsinh    19.4    62.8    3.24
BenchmarkAtan     10.1    23      2.28
BenchmarkAtanh    19.1    53.2    2.79
BenchmarkAtan2    16.5    33.9    2.05
BenchmarkCbrt     14.8    58      3.92
BenchmarkErf      10.8    20.1    1.86
BenchmarkErfc     11.2    23.5    2.10
BenchmarkExp      8.77    53.8    6.13
BenchmarkExpm1    10.1    38.3    3.79
BenchmarkLog      13.1    40.1    3.06
BenchmarkLog1p    12.7    38.3    3.02
BenchmarkPowInt   31.7    40.5    1.28
BenchmarkPowFrac  33.1    141     4.26
BenchmarkTan      11.5    30      2.61

Accuracy was tested against a high precision
reference function to determine maximum error.
Note: ulperr is error in "units in the last place"

       max
      ulperr
Acos  1.15
Acosh 1.07
Asin  2.22
Asinh 1.72
Atan  1.41
Atanh 3.00
Atan2 1.45
Cbrt  1.18
Erf   1.29
Erfc  4.82
Exp   1.00
Expm1 2.26
Log   0.94
Log1p 2.39
Tan   3.14

Pow will have 99.99% correctly rounded results with reasonable inputs
producing numeric (non Inf or NaN) results

Change-Id: I850e8cf7b70426e8b54ec49d74acd4cddc8c6cb2
Reviewed-on: https://go-review.googlesource.com/38585
Reviewed-by: Michael Munday <munday@ca.ibm.com>
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agobytes: skip inline test by default
Marvin Stenger [Mon, 8 May 2017 11:18:15 +0000 (13:18 +0200)]
bytes: skip inline test by default

The test "TestTryGrowByResliceInlined" introduced in c08ac36 broke the
noopt builder as it fails when inlining is disabled.
Since there are currently no other options at hand for checking
inlined-ness other than looking at emited symbols of the compilation,
we for now skip the problem causing test by default and only run
it on one specific builder ("linux-amd64").
Also see CL 42813, which introduced the test and contains comments
suggesting this temporary solution.

Change-Id: I3978ab0831da04876cf873d78959f821c459282b
Reviewed-on: https://go-review.googlesource.com/42820
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agointernal/poll: remove allocation in windows FD.Writev
Alex Brainman [Sun, 30 Apr 2017 08:21:16 +0000 (18:21 +1000)]
internal/poll: remove allocation in windows FD.Writev

Use closure parameter instead of external variable to
remove 1 allocation.

I tried to add test, but it is difficult to add something simple
and not flake here. I did test this with:

diff --git a/src/net/writev_test.go b/src/net/writev_test.go
index 4c05be4..e417d68 100644
--- a/src/net/writev_test.go
+++ b/src/net/writev_test.go
@@ -99,6 +99,15 @@ func TestBuffers_WriteTo(t *testing.T) {
  }
 }

+func TestBuffers_WriteToAllocs(t *testing.T) {
+ allocs := testing.AllocsPerRun(10, func() {
+ testBuffer_writeTo(t, 10, false)
+ })
+ if allocs > 0 {
+ t.Fatalf("got %v; want 0", allocs)
+ }
+}
+
 func testBuffer_writeTo(t *testing.T, chunks int, useCopy bool) {
  oldHook := poll.TestHookDidWritev
  defer func() { poll.TestHookDidWritev = oldHook }()

It makes allocation count go down by 1 after the fix.

Before:

C:\>u:\test -test.v -test.run=WriteToAllocs
=== RUN   TestBuffers_WriteToAllocs
--- FAIL: TestBuffers_WriteToAllocs (0.05s)
        writev_test.go:107: got 66; want 0
FAIL

and after:

C:\>u:\test -test.v -test.run=WriteToAllocs
=== RUN   TestBuffers_WriteToAllocs
--- FAIL: TestBuffers_WriteToAllocs (0.04s)
        writev_test.go:107: got 65; want 0
FAIL

Thanks to @MichaelMonashev for report and the fix.

Fixes #19222

Change-Id: I0f73cd9e2c8bbaa0653083f81f3ccb83b5ea84e1
Reviewed-on: https://go-review.googlesource.com/42893
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/link/internal/ld: don't link with -no_pie on darwin/arm64
Elias Naur [Sun, 7 May 2017 17:42:11 +0000 (19:42 +0200)]
cmd/link/internal/ld: don't link with -no_pie on darwin/arm64

Ever since CL 33301 linking darwin/arm64 excutables has resulted in
warnings like:

ld: warning: -no_pie ignored for arm64

Remove -no_pie on darwin/arm64.

Change-Id: I9f7685351fa8cce29795283e1a24fc7a6753d698
Reviewed-on: https://go-review.googlesource.com/42815
Run-TryBot: Elias Naur <elias.naur@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoos, cmd/link: fix typos
Kevin Burke [Sun, 7 May 2017 18:04:24 +0000 (11:04 -0700)]
os, cmd/link: fix typos

Also switch "stating" to "statting" to describe applying os.Stat to
a resource; the former is more confusable than the latter.

Change-Id: I9d8e3506bd383f8f1479c05948c03b8c633dc4af
Reviewed-on: https://go-review.googlesource.com/42855
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agobytes: optimize Buffer's Write, WriteString, WriteByte, and WriteRune
Marvin Stenger [Sun, 7 May 2017 08:43:17 +0000 (10:43 +0200)]
bytes: optimize Buffer's Write, WriteString, WriteByte, and WriteRune

In the common case, the grow method only needs to reslice the internal
buffer. Making another function call to grow can be expensive when Write
is called very often with small pieces of data (like a byte or rune).
Thus, we add a tryGrowByReslice method that is inlineable so that we can
avoid an extra call in most cases.

name                       old time/op    new time/op    delta
WriteByte-4                  35.5µs ± 0%    17.4µs ± 1%   -51.03%  (p=0.000 n=19+20)
WriteRune-4                  55.7µs ± 1%    38.7µs ± 1%   -30.56%  (p=0.000 n=18+19)
BufferNotEmptyWriteRead-4     304µs ± 5%     283µs ± 3%    -6.86%  (p=0.000 n=19+17)
BufferFullSmallReads-4       87.0µs ± 5%    66.8µs ± 2%   -23.26%  (p=0.000 n=17+17)

name                       old speed      new speed      delta
WriteByte-4                 115MB/s ± 0%   235MB/s ± 1%  +104.19%  (p=0.000 n=19+20)
WriteRune-4                 221MB/s ± 1%   318MB/s ± 1%   +44.01%  (p=0.000 n=18+19)

Fixes #17857

Change-Id: I08dfb10a1c7e001817729dbfcc951bda12fe8814
Reviewed-on: https://go-review.googlesource.com/42813
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/asm: enable MOVSD in the encoding end-to-end test
Damien Lespiau [Sun, 7 May 2017 14:58:03 +0000 (15:58 +0100)]
cmd/asm: enable MOVSD in the encoding end-to-end test

MOVSD is properly handled but its encoding test wasn't enabled. Enable
it.

For reference this was found with a little tool I wrote [1] to explore
which instructions are missing or not tested in the go obj package and
assembler:

"which SSE2 instructions aren't tested? And don't list instructions
which can take MMX operands"

$ x86db-gogen list --extension SSE2 --not-tested --not-mmx
CLFLUSH mem           [m:  np 0f ae /7] WILLAMETTE,SSE2
MOVSD   xmmreg,xmmreg [rm: f2 0f 10 /r] WILLAMETTE,SSE2
MOVSD   xmmreg,xmmreg [mr: f2 0f 11 /r] WILLAMETTE,SSE2
MOVSD   mem64,xmmreg  [mr: f2 0f 11 /r] WILLAMETTE,SSE2
MOVSD   xmmreg,mem64  [rm: f2 0f 10 /r] WILLAMETTE,SSE2

(CLFLUSH was introduced with SSE2, but has its own CPUID bit)

[1] https://github.com/dlespiau/x86db

Change-Id: Ic3af3028cb8d4f02e53fdebb9b30fb311f4ee454
Reviewed-on: https://go-review.googlesource.com/42814
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoos: reimplement windows os.Stat
Alex Brainman [Thu, 27 Apr 2017 07:43:33 +0000 (17:43 +1000)]
os: reimplement windows os.Stat

Currently windows Stat uses combination of Lstat and Readlink to
walk symlinks until it reaches file or directory. Windows Readlink
is implemented via Windows DeviceIoControl(FSCTL_GET_REPARSE_POINT, ...)
call, but that call does not work on network shares or inside of
Docker container (see issues #18555 ad #19922 for details).

But Raymond Chen suggests different approach:
https://blogs.msdn.microsoft.com/oldnewthing/20100212-00/?p=14963/
- he suggests to use Windows I/O manager to dereferences the
symbolic link.

This appears to work for all normal symlinks, but also for network
shares and inside of Docker container.

This CL implements described procedure.

I also had to adjust TestStatSymlinkLoop, because the test is
expecting Stat to return syscall.ELOOP for symlink with a loop.
But new Stat returns Windows error of ERROR_CANT_RESOLVE_FILENAME
= 1921 instead. I could map ERROR_CANT_RESOLVE_FILENAME into
syscall.ELOOP, but I suspect the former is broader than later.
And ERROR_CANT_RESOLVE_FILENAME message text of "The name of
the file cannot be resolved by the system." sounds fine to me.

Fixes #10935
Fixes #18555
Fixes #19922

Change-Id: I979636064cdbdb9c7c840cf8ae73fe2c24499879
Reviewed-on: https://go-review.googlesource.com/41834
Reviewed-by: Harshavardhana <hrshvardhana@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/link/internal/ld: fix external linking on windows
Egon Elbre [Thu, 4 May 2017 13:14:59 +0000 (16:14 +0300)]
cmd/link/internal/ld: fix external linking on windows

gcc on windows writes invalid location for .debug_gdb_scripts
which causes the executable loading to fail.

Fixes #20183

Change-Id: I5134013bc926b44a55b528f66ab79555855d1f4d
Reviewed-on: https://go-review.googlesource.com/42651
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/asm: fix operand order of ARM's MULA instruction
Ben Shi [Fri, 28 Apr 2017 10:55:41 +0000 (10:55 +0000)]
cmd/asm: fix operand order of ARM's MULA instruction

As discussion in issue #19141, the addend should be the third
argument of MULA. This patch fixes it in both the front end
and the back end of the assembler. And also tests are added to
the encoding test.

Fixes #19141

Change-Id: Idbc6f338b8fdfcad97a135f27a98c5b375b27d43
Reviewed-on: https://go-review.googlesource.com/42028
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agoimage/gif: don't panic on large or nil-containing color.Palettes.
Nigel Tao [Fri, 5 May 2017 06:44:42 +0000 (16:44 +1000)]
image/gif: don't panic on large or nil-containing color.Palettes.

Fixes #20249

Change-Id: I5dceaef31de96345f8e6c155e12775dc4cc31bfb
Reviewed-on: https://go-review.googlesource.com/42790
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agogo/types: remove invalid documentation and assertion on package names
Robert Griesemer [Fri, 5 May 2017 20:57:22 +0000 (13:57 -0700)]
go/types: remove invalid documentation and assertion on package names

NewPackage required through documentation that the package name not
be blank (which wasn't true since each time we check a new package
we create one with a blank name (api.go:350). NewPackage also asserted
that a package name not be "_". While it is invalid for a package name
to be "_", one could conceivably create a package named "_" through
export data manipulation. Furthermore, it is ok to import a package
with package path "_" as long as the package itself is not named "_".

- removed misleading documentation
- removed unnecessary assertion
- added safety checks when we actually do the import

Fixes #20231.

Change-Id: I1eb1ab7b5e3130283db715374770cf05d749d159
Reviewed-on: https://go-review.googlesource.com/42852
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
8 years agogo/importer: report import path if package is not found
Robert Griesemer [Fri, 5 May 2017 19:47:53 +0000 (12:47 -0700)]
go/importer: report import path if package is not found

Fixes #20230.

Change-Id: I2e9b9e9d2540eb66c8411ac7910962933bc2c0e9
Reviewed-on: https://go-review.googlesource.com/42870
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: add Func.SetNilCheckDisabled
Josh Bleecher Snyder [Thu, 4 May 2017 14:55:00 +0000 (07:55 -0700)]
cmd/compile: add Func.SetNilCheckDisabled

Generated hash and eq routines don't need nil checks.
Prior to this CL, this was accomplished by
temporarily incrementing the global variable disable_checknil.
However, that increment lasted only the lifetime of the
call to funccompile. After CL 41503, funccompile may
do nothing but enqueue the function for compilation,
resulting in nil checks being generated.

Fix this by adding an explicit flag to a function
indicating whether nil checks should be disabled
for that function.

While we're here, allow concurrent compilation
with the -w and -W flags, since that was needed
to investigate this issue.

Fixes #20242

Change-Id: Ib9140c22c49e9a09e62fa3cf350f5d3eff18e2bd
Reviewed-on: https://go-review.googlesource.com/42591
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/internal/obj/ppc64, cmd/link/internal/ppc64: Change function alignment to 16
Carlos Eduardo Seo [Mon, 6 Feb 2017 17:33:43 +0000 (15:33 -0200)]
cmd/internal/obj/ppc64, cmd/link/internal/ppc64: Change function alignment to 16

The Power processor manual states that "Branches not from the last instruction
of an aligned quadword and not to the first instruction of an aligned quadword
cause inefficiencies in the IBuffer". This changes the function alignment from 8
to 16 bytes to comply with that.

Fixes #18963

Change-Id: Ibce9bf8302110a86c6ab05948569af9ffdfcf4bb
Reviewed-on: https://go-review.googlesource.com/36390
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>

8 years agohtml/template: allow safe usage of predefined escapers in pipelines
Samuel Tan [Mon, 17 Apr 2017 23:10:54 +0000 (16:10 -0700)]
html/template: allow safe usage of predefined escapers in pipelines

Allow the predefined escapers "html", "urlquery", and "js" to be used
in pipelines when they have no potential to affect the correctness or
safety of the escaped pipeline output. Specifically:
- "urlquery" may be used if it is the last command in the pipeline.
- "html" may be used if it is the last command in the pipeline, and
  the pipeline does not occur in an unquoted HTML attribute value
  context.
- "js" may be used in any pipeline, since it does not affect the
  merging of contextual escapers.

This change will loosens the restrictions on predefined escapers
introduced in golang.org/cl/37880, which will hopefully ease the
upgrade path for existing template users.

This change brings back the escaper-merging logic, and associated
unit tests, that were removed in golang.org/cl/37880. However, a
few notable changes have been made:
- "_html_template_nospaceescaper" is no longer considered
  equivalent to "html", since the former escapes spaces, while
  the latter does not (see #19345). This change should not silently
  break any templates, since pipelines where this substituion will
  happen will already trigger an explicit error.
- An "_eval_args_" internal directive has been added to
  handle pipelines containing a single explicit call to a
  predefined escaper, e.g. {{html .X}} (see #19353).

Also, the HTMLEscape function called by the predefined
text/template "html" function now escapes the NULL character as
well. This effectively makes it as secure as the internal
html/template HTML escapers (see #19345). While this change is
backward-incompatible, it will only affect illegitimate uses
of this escaper, since the NULL character is always illegal in
valid HTML.

Fixes #19952

Change-Id: I9b5570a80a3ea284b53901e6a1f842fc59b33d3a
Reviewed-on: https://go-review.googlesource.com/40936
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agotemplate: warn about interleaved nature of writes
Dieter Plaetinck [Fri, 10 Feb 2017 10:54:58 +0000 (11:54 +0100)]
template: warn about interleaved nature of writes

Execute incurs separate writes for each "step", e.g. each
variable that needs to be printed, and the final newline.
While it is correct to state that templates can be executed
concurrently, there is a more subtle nuance that is easily missed:
when writing to the same writer, the writes from concurrent execute
calls can be interleaved, leading to unexpected output.

Change-Id: I0abbd7960d8a8d15e109a8a3eeff3b43b852bbbf
Reviewed-on: https://go-review.googlesource.com/37444
Reviewed-by: Rob Pike <r@golang.org>
8 years agocmd/link: stop passing unused read_only_relocs
David Crawshaw [Thu, 4 May 2017 21:36:20 +0000 (17:36 -0400)]
cmd/link: stop passing unused read_only_relocs

The external darwin linker has been printing:

ld: warning: -read_only_relocs cannot be used with x86_64

for a long time. Now that it is printed by CL 33301, we may as
well get rid of it.

Fixes #20246

Change-Id: I1147cf1ff197fdfda228a1349f13627bcf9fc72f
Reviewed-on: https://go-review.googlesource.com/42730
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Todd Neal <todd@tneal.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/link: do not read .bss sections in ldpe
Alex Brainman [Thu, 4 May 2017 06:49:23 +0000 (16:49 +1000)]
cmd/link: do not read .bss sections in ldpe

For .bss section symbol ldelf does not set P (raw symbol data).
Make ldpe do the same.

Change-Id: Ib3d558456f505ee568d0972465fa9b08b5794a87
Reviewed-on: https://go-review.googlesource.com/42631
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agogo/constant: avoid generating rats for large negative exponents
Josh Bleecher Snyder [Thu, 4 May 2017 15:41:49 +0000 (08:41 -0700)]
go/constant: avoid generating rats for large negative exponents

Fixes #20228

Change-Id: I1893ae3e192da01f9befe5469b2a32e534a691ba
Reviewed-on: https://go-review.googlesource.com/42592
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/compile: suppress duplicate type errors
Josh Bleecher Snyder [Thu, 4 May 2017 18:43:41 +0000 (11:43 -0700)]
cmd/compile: suppress duplicate type errors

If we've already complained about a type T,
don't complain again about further expressions
involving it.

Fixes #20245 and hopefully all of its ilk.

Change-Id: Ic0abe8235d52e8a7ac40e3615aea8f3a54fd7cec
Reviewed-on: https://go-review.googlesource.com/42690
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/link: set ELF CPIC bit in mips64 objects
Ian Lance Taylor [Thu, 4 May 2017 20:13:24 +0000 (16:13 -0400)]
cmd/link: set ELF CPIC bit in mips64 objects

We already set it for mips32 objects. The native ELF linker warns when
linking PIC objects with non-PIC objects. Our objects are PIC, but we
were not marking them as such.

Fixes #20243.

Change-Id: Ifab131200b263e4c72cf81f7b131a65ac02a13a9
Reviewed-on: https://go-review.googlesource.com/42710
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/go, cmd/compile: match tool versions
David Crawshaw [Thu, 4 May 2017 17:58:27 +0000 (13:58 -0400)]
cmd/go, cmd/compile: match tool versions

This change passes runtime.Version from the go tool to the compiler.
If the versions do not match, the compilation fails.
The result is a go tool from one GOROOT will complain loudly if it
is invoked with a different GOROOT value.

Only release versions are checked, so that when developing Go
you can still use "go install cmd/go" and "go install cmd/compile"
separately.

Fixes #19064

Change-Id: I17e184d07d3c1092b1d9af53ba55ed3ecf67791d
Reviewed-on: https://go-review.googlesource.com/42595
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agodoc: fix minor formatting issues in contribute.html
Adam Bender [Thu, 4 May 2017 01:57:40 +0000 (18:57 -0700)]
doc: fix minor formatting issues in contribute.html

- remove <code> from <pre>
- replace `` with <code></code>

Change-Id: I46f0aec8b7645e2ac8cb53bca73aed55441acd65
Reviewed-on: https://go-review.googlesource.com/42612
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoall: use time.Time.Equal instead of == in some tests
Brad Fitzpatrick [Thu, 4 May 2017 15:05:33 +0000 (15:05 +0000)]
all: use time.Time.Equal instead of == in some tests

(Found by making time.Time uncomparable and rerunning std tests
locally.)

Change-Id: I4fa6fb0ba7334965362387e2f6541c17a27ac3aa
Reviewed-on: https://go-review.googlesource.com/42616
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/go: use os.Executable to find GOROOT
David Crawshaw [Wed, 3 May 2017 18:46:28 +0000 (14:46 -0400)]
cmd/go: use os.Executable to find GOROOT

Before this change, building a GOROOT using make.bash, and then
moving the entire to a new path confused the go tool. Correct
operation of the go tool under these conditions required either
running make.bash again (not always possible if the new location
was owned by a different system user) or setting the GOROOT
environment variable. Setting GOROOT is unfortunate and
discouraged, as it makes it too easy to use the go tool from
one GOROOT and the compiler from another GOROOT.

With this change, the go tool finds its GOROOT relative to its
own location, using os.Executable. It checks it is in a GOROOT
by searching for the GOROOT/pkg/tool directory, to avoid two
plausible situations:

ln -s $GOROOT/bin/go /usr/local/bin/go

and

PATH=$HOME/bin:$PATH
GOPATH=$HOME
ln -s $GOROOT/bin/go $HOME/bin/go

Additionally, if the current executable path is not in a GOROOT,
the tool will follow any symlinks for the executable and check
to see if its original path is a GOROOT.

Fixes #18678

Change-Id: I151d7d449d213164f98193cc176b616849e6332c
Reviewed-on: https://go-review.googlesource.com/42533
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocrypto: revise wording about cryptographically broken algorithms
Brad Fitzpatrick [Thu, 4 May 2017 02:49:23 +0000 (02:49 +0000)]
crypto: revise wording about cryptographically broken algorithms

Updates text from https://golang.org/cl/42511

Updates #14395

Change-Id: I711100525e074ab360e577520280c37645db1c95
Reviewed-on: https://go-review.googlesource.com/42614
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Rob Pike <r@golang.org>
8 years agospec: remove an unnecessary semicolon from code example
Brad Fitzpatrick [Thu, 4 May 2017 02:39:56 +0000 (02:39 +0000)]
spec: remove an unnecessary semicolon from code example

Change-Id: Ie4c92da0e3cbb97d3d7e03c7d15196c34f58a2cd
Reviewed-on: https://go-review.googlesource.com/42613
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/compile: prevent panic while formatting func(...T) with unknown T
Josh Bleecher Snyder [Wed, 3 May 2017 23:37:03 +0000 (16:37 -0700)]
cmd/compile: prevent panic while formatting func(...T) with unknown T

Compile:

package p

var f = func(...A)

Before this CL:

x.go:3:13: type %!v(PANIC=runtime error: invalid memory address or nil pointer dereference) is not an expression
x.go:3:17: undefined: A

After this CL:

x.go:3:13: type func(...<T>) is not an expression
x.go:3:17: undefined: A

Found with go-fuzz.

Fixes #20233

Change-Id: Ibb232b3954c4091071440eba48b44c4022a8083f
Reviewed-on: https://go-review.googlesource.com/42610
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocrypto: document insecure algorithms
Andreas Auernhammer [Tue, 2 May 2017 23:08:31 +0000 (01:08 +0200)]
crypto: document insecure algorithms

This change explicitly documents that DES, MD5, RC4 and SHA-1 are
insecure / broken - at all or at least within a commonly used scenario.

Fixes #14395

Change-Id: Id1d543c85d67968ba64ed7495313501953c3ef3a
Reviewed-on: https://go-review.googlesource.com/42511
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agonet/http: update bundled http2
Brad Fitzpatrick [Wed, 3 May 2017 12:24:04 +0000 (12:24 +0000)]
net/http: update bundled http2

Updates bundled http2 to x/net/http2 git rev feeb485 for:

    http2: add all bad ciphers, use package constants
    https://golang.org/cl/42510

Updates #20213

Change-Id: I851453e3785e6b126db7a5c5eec2ebbbf61358ae
Reviewed-on: https://go-review.googlesource.com/42494
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Savintsev <dsavints@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocrypto/x509: load certs from env vars + extra locations
Steven Hartland [Thu, 2 Feb 2017 00:34:06 +0000 (00:34 +0000)]
crypto/x509: load certs from env vars + extra locations

Add the ability to override the default file and directory from
which certificates are loaded by setting the OpenSSL compatible
environment variables: SSL_CERT_FILE, SSL_CERT_DIR.

If the variables are set the default locations are not checked.

Added new default file "/usr/local/etc/ssl/cert.pem" for FreeBSD.

Certificates in the first valid location found for both file and
directory are added, instead of only the first file location if
a valid one was found, which is consistent with OpenSSL.

Fixes #3905
Fixes #14022
Fixes #14311
Fixes #16920
Fixes #18813 - If user sets SSL_CERT_FILE.

Change-Id: Ia24fb7c1c2ffff4338b4cf214bd040326ce27bb0
Reviewed-on: https://go-review.googlesource.com/36093
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/link/internal/ld: always show output from external linker
Shenghou Ma [Wed, 16 Nov 2016 06:05:34 +0000 (01:05 -0500)]
cmd/link/internal/ld: always show output from external linker

Fixes #17935.

Change-Id: I49b0f6cee29ea76ed62b8faa5d6d1f51be41bf84
Reviewed-on: https://go-review.googlesource.com/33301
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: avoid panic when printing *<N>
Josh Bleecher Snyder [Wed, 3 May 2017 14:45:20 +0000 (07:45 -0700)]
cmd/compile: avoid panic when printing *<N>

When compiling the program:

package p

func _(){
*;:=
}

Before:

x.go:4:3: syntax error: unexpected semicolon, expecting expression
x.go:4:4: non-name *%!v(PANIC=runtime error: invalid memory address or nil pointer dereference) on left side of :=
x.go:5:1: syntax error: unexpected }, expecting expression

After:

x.go:4:3: syntax error: unexpected semicolon, expecting expression
x.go:4:4: non-name *<N> on left side of :=
x.go:5:1: syntax error: unexpected }, expecting expression

No test because:

(1) we don't have a good mechanism to check for the
    absence of the string "PANIC" in an error message
(2) the string "*<N>", while better, is itself ugly enough
    that I don't want to actively check for it
(3) the bug isn't very important, the kind of thing only fuzzers encounter
(4) the fix is obvious and trivial

Fixes #20220

Change-Id: I35faa986b60b671414ee999d6264b06937f250e3
Reviewed-on: https://go-review.googlesource.com/42498
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agovendor/golang_org/x/crypto/chacha20poly1305: revendor
Martin Möhrmann [Wed, 3 May 2017 08:00:38 +0000 (10:00 +0200)]
vendor/golang_org/x/crypto/chacha20poly1305: revendor

Brings in chacha20poly1305 directory from golang.org/x/crypto revision
12e9ca725de4806fbda1610fd95aacad15bd6810, adding:

CL 41862: chacha20poly1305: add runtime internal independent cpu feature detection
CL 39952: add import comment

Change-Id: Ic46ff24b081bc1c66b6317334d33180e33bfd318
Reviewed-on: https://go-review.googlesource.com/42513
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: clean redundant call of environ function
feilengcui008 [Wed, 3 May 2017 06:08:43 +0000 (14:08 +0800)]
runtime: clean redundant call of environ function

Change-Id: Ica7179d225c1fb79381f82f58ea5050ac6418b9c
Reviewed-on: https://go-review.googlesource.com/42493
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoencoding/gob: use MakeMapWithSize when decoding map
Filip Gruszczyński [Sun, 9 Apr 2017 17:44:39 +0000 (10:44 -0700)]
encoding/gob: use MakeMapWithSize when decoding map

This allows to pre-allocate the final size of the hashmap and avoid
re-allocating as we insert entries. Furthermore for the current
implementation of the hashmap it allows avoiding several rounds of
evacuating hashmap entries after each re-allocation.

DecodeComplex128Slice-8  51.9µs ± 1%  51.9µs ± 2%     ~     (p=0.797 n=30+29)
DecodeFloat64Slice-8     31.5µs ± 2%  31.6µs ± 2%     ~     (p=0.050 n=28+28)
DecodeInt32Slice-8       32.0µs ± 2%  31.9µs ± 3%     ~     (p=0.666 n=29+28)
DecodeStringSlice-8      57.7µs ± 2%  57.8µs ± 3%     ~     (p=0.780 n=27+30)
DecodeInterfaceSlice-8    498µs ± 2%   495µs ± 2%     ~     (p=0.070 n=28+29)
DecodeMap-8               300µs ± 2%   230µs ± 5%  -23.31%  (p=0.000 n=27+27)

Updates #19525

Change-Id: Ia7233da49f05bae7a86c064d9ecebca966f5f2f7
Reviewed-on: https://go-review.googlesource.com/40113
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: don't panic for bad size hint in hashmap
Filip Gruszczynski [Sat, 15 Apr 2017 22:17:29 +0000 (15:17 -0700)]
runtime: don't panic for bad size hint in hashmap

Because the hint parameter is supposed to be treated
purely as a hint, if it doesn't meet the requirements
we disregard it and continue as if there was no hint
at all.

Fixes #19926

Change-Id: I86e7f99472fad6b99ba4e2fd33e4a9e55d55115e
Reviewed-on: https://go-review.googlesource.com/40854
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agodoc/contribute: directly link to the Agreements page
Alberto Donizetti [Tue, 2 May 2017 10:38:59 +0000 (12:38 +0200)]
doc/contribute: directly link to the Agreements page

There's no Settings->Agreement path for PolyGerrit users, but if we
link directly to the page in the instructions, Gerrit will inform them
that they can access the page by switching to the old UI.

Fixes #20207

Change-Id: I0887ee854e4ac5975b5f305adb6259b81b41618f
Reviewed-on: https://go-review.googlesource.com/42412
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoapi: update next.txt
Brad Fitzpatrick [Tue, 2 May 2017 17:41:24 +0000 (17:41 +0000)]
api: update next.txt

Change-Id: Ifbfd71a9c0d447e22c369c9d1209063b2a5c657b
Reviewed-on: https://go-review.googlesource.com/42490
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/{asm,compile}: avoid zeroAuto clobbering flags on s390x
Michael Munday [Sun, 30 Apr 2017 18:25:57 +0000 (14:25 -0400)]
cmd/{asm,compile}: avoid zeroAuto clobbering flags on s390x

This CL modifies how MOV[DWHB] instructions that store a constant to
memory are assembled to avoid them clobbering the condition code
(flags). It also modifies zeroAuto to use MOVD instructions instead of
CLEAR (which is assembled as XC).

MOV[DWHB]storeconst ops also no longer clobbers flags.

Note: this CL modifies the assembler so that it can no longer handle
immediates outside the range of an int16 or offsets from SB, which
reflects what the machine instructions support. The compiler doesn't
need this capability any more and I don't think this affects any existing
assembly, but it is easy to workaround if it does.

Fixes #20187.

Change-Id: Ie54947ff38367bd6a19962bf1a6d0296a4accffb
Reviewed-on: https://go-review.googlesource.com/42179
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoexpvar: reduce sync.Map overhead for (*Map).Set
Bryan C. Mills [Sun, 30 Apr 2017 23:23:45 +0000 (19:23 -0400)]
expvar: reduce sync.Map overhead for (*Map).Set

Use Load instead of LoadOrStore in the fast path to save 1 alloc/op
for existing keys.

name                          old time/op    new time/op    delta
IntAdd                          6.39ns ± 7%    6.97ns ±19%     ~     (p=0.105 n=8+8)
IntAdd-6                        12.3ns ± 1%    12.2ns ± 1%     ~     (p=0.396 n=7+7)
IntSet                          6.41ns ± 6%    6.94ns ±21%     ~     (p=0.168 n=8+8)
IntSet-6                        12.1ns ± 3%    11.7ns ± 9%     ~     (p=0.496 n=7+8)
FloatAdd                        14.3ns ± 9%    14.7ns ± 4%     ~     (p=0.088 n=8+8)
FloatAdd-6                      36.5ns ± 1%    36.6ns ± 0%     ~     (p=0.709 n=7+6)
FloatSet                        6.59ns ± 7%    6.47ns ± 7%     ~     (p=0.397 n=8+7)
FloatSet-6                      12.2ns ± 1%    12.2ns ± 2%     ~     (p=0.748 n=7+7)
StringSet                       67.8ns ± 6%    68.7ns ± 6%     ~     (p=0.342 n=8+8)
StringSet-6                     41.8ns ± 5%    41.7ns ± 5%     ~     (p=0.979 n=8+8)
MapSet                           294ns ± 6%     234ns ± 4%  -20.35%  (p=0.000 n=8+8)
MapSet-6                        95.8ns ± 2%    89.4ns ± 3%   -6.73%  (p=0.000 n=8+8)
MapSetDifferent                 1.31µs ± 5%    1.07µs ± 4%  -18.21%  (p=0.000 n=8+8)
MapSetDifferent-6                260ns ± 8%     210ns ± 9%  -19.44%  (p=0.000 n=8+8)
MapSetString                     294ns ± 6%     236ns ± 4%  -19.92%  (p=0.000 n=8+8)
MapSetString-6                  95.6ns ± 2%    89.9ns ± 2%   -5.97%  (p=0.000 n=7+8)
MapAddSame                      1.46µs ± 3%    1.46µs ± 5%     ~     (p=0.721 n=8+8)
MapAddSame-6                     328ns ± 6%     330ns ± 4%     ~     (p=0.776 n=8+8)
MapAddDifferent                 4.89µs ± 7%    4.98µs ± 6%     ~     (p=0.505 n=8+8)
MapAddDifferent-6               1.02µs ± 3%    1.01µs ± 4%     ~     (p=0.352 n=7+8)
MapAddSameSteadyState           62.1ns ± 7%    60.8ns ± 4%     ~     (p=0.521 n=8+8)
MapAddSameSteadyState-6         38.1ns ± 3%    37.7ns ± 0%     ~     (p=0.185 n=7+6)
MapAddDifferentSteadyState       290ns ± 5%     293ns ± 4%     ~     (p=0.515 n=8+8)
MapAddDifferentSteadyState-6    63.0ns ± 7%    63.7ns ±11%     ~     (p=0.482 n=7+8)
RealworldExpvarUsage            7.39µs ± 5%    7.51µs ± 5%     ~     (p=0.382 n=8+8)
RealworldExpvarUsage-6          3.07µs ±28%    3.04µs ±43%     ~     (p=0.798 n=8+8)

name                          old alloc/op   new alloc/op   delta
IntAdd                           0.00B          0.00B          ~     (all equal)
IntAdd-6                         0.00B          0.00B          ~     (all equal)
IntSet                           0.00B          0.00B          ~     (all equal)
IntSet-6                         0.00B          0.00B          ~     (all equal)
FloatAdd                         0.00B          0.00B          ~     (all equal)
FloatAdd-6                       0.00B          0.00B          ~     (all equal)
FloatSet                         0.00B          0.00B          ~     (all equal)
FloatSet-6                       0.00B          0.00B          ~     (all equal)
StringSet                        16.0B ± 0%     16.0B ± 0%     ~     (all equal)
StringSet-6                      16.0B ± 0%     16.0B ± 0%     ~     (all equal)
MapSet                           48.0B ± 0%     32.0B ± 0%  -33.33%  (p=0.000 n=8+8)
MapSet-6                         48.0B ± 0%     32.0B ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetDifferent                   192B ± 0%      128B ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetDifferent-6                 192B ± 0%      128B ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetString                     48.0B ± 0%     32.0B ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetString-6                   48.0B ± 0%     32.0B ± 0%  -33.33%  (p=0.000 n=8+8)
MapAddSame                        480B ± 0%      480B ± 0%     ~     (all equal)
MapAddSame-6                      480B ± 0%      480B ± 0%     ~     (all equal)
MapAddDifferent                 1.09kB ± 0%    1.09kB ± 0%     ~     (all equal)
MapAddDifferent-6               1.09kB ± 0%    1.09kB ± 0%     ~     (all equal)
MapAddSameSteadyState            0.00B          0.00B          ~     (all equal)
MapAddSameSteadyState-6          0.00B          0.00B          ~     (all equal)
MapAddDifferentSteadyState       0.00B          0.00B          ~     (all equal)
MapAddDifferentSteadyState-6     0.00B          0.00B          ~     (all equal)
RealworldExpvarUsage             0.00B          0.00B          ~     (all equal)
RealworldExpvarUsage-6           0.00B          0.00B          ~     (all equal)

name                          old allocs/op  new allocs/op  delta
IntAdd                            0.00           0.00          ~     (all equal)
IntAdd-6                          0.00           0.00          ~     (all equal)
IntSet                            0.00           0.00          ~     (all equal)
IntSet-6                          0.00           0.00          ~     (all equal)
FloatAdd                          0.00           0.00          ~     (all equal)
FloatAdd-6                        0.00           0.00          ~     (all equal)
FloatSet                          0.00           0.00          ~     (all equal)
FloatSet-6                        0.00           0.00          ~     (all equal)
StringSet                         1.00 ± 0%      1.00 ± 0%     ~     (all equal)
StringSet-6                       1.00 ± 0%      1.00 ± 0%     ~     (all equal)
MapSet                            3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
MapSet-6                          3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetDifferent                   12.0 ± 0%       8.0 ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetDifferent-6                 12.0 ± 0%       8.0 ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetString                      3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
MapSetString-6                    3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
MapAddSame                        11.0 ± 0%      11.0 ± 0%     ~     (all equal)
MapAddSame-6                      11.0 ± 0%      11.0 ± 0%     ~     (all equal)
MapAddDifferent                   31.0 ± 0%      31.0 ± 0%     ~     (all equal)
MapAddDifferent-6                 31.0 ± 0%      31.0 ± 0%     ~     (all equal)
MapAddSameSteadyState             0.00           0.00          ~     (all equal)
MapAddSameSteadyState-6           0.00           0.00          ~     (all equal)
MapAddDifferentSteadyState        0.00           0.00          ~     (all equal)
MapAddDifferentSteadyState-6      0.00           0.00          ~     (all equal)
RealworldExpvarUsage              0.00           0.00          ~     (all equal)
RealworldExpvarUsage-6            0.00           0.00          ~     (all equal)

https://perf.golang.org/search?q=upload:20170501.1

Change-Id: I28fc3906473f2b7307f6d1ae05a8d9b01ef8a6f8
Reviewed-on: https://go-review.googlesource.com/42211
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agonet/http: bgErr is unused; remove it
Chris Manghane [Tue, 2 May 2017 05:34:37 +0000 (01:34 -0400)]
net/http: bgErr is unused; remove it

Fixes #20196.

Change-Id: Ib87f6e9e27a38f21f860b7150c818d77be653dd3
Reviewed-on: https://go-review.googlesource.com/42370
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/dist: spin up fewer test dir runners
Josh Bleecher Snyder [Tue, 2 May 2017 12:35:55 +0000 (05:35 -0700)]
cmd/dist: spin up fewer test dir runners

5 shards, each of which spins up NumCPU processes,
each of which is running at GOMAXPROCS=NumCPU,
is too much for one machine. It makes my laptop unusable.

It might also be in part responsible for test flakes
that require a moderately responsive system,
like #18589 (backedge scheduling) and #19276 (locklinear).

It's possible that Go should be a better neighbor in general;
that's #17969. In the meantime, fix this corner of the world.

Builders snapshot the world and run shards on different
machines, so keeping sharding high for them is good.

This is a partial reversion of CL 18199.

Fixes #20141.

Change-Id: I123cf9436f4f4da3550372896265c38117b78071
Reviewed-on: https://go-review.googlesource.com/42431
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/internal/obj: fix LSym.Type during compilation, not linking
Josh Bleecher Snyder [Sun, 16 Apr 2017 15:22:44 +0000 (08:22 -0700)]
cmd/internal/obj: fix LSym.Type during compilation, not linking

Prior to this CL, the compiler and assembler
were sloppy about the LSym.Type for LSyms
containing static data.

The linker then fixed this up, converting
Sxxx and SBSS to SDATA, and SNOPTRBSS to SNOPTRDATA
if it noticed that the symbol had associated data.

It is preferable to just get this right in cmd/compile
and cmd/asm, because it removes an unnecessary traversal
of the symbol table from the linker (see #14624).
Do this by touching up the LSym.Type fixes in
LSym.prepwrite and Link.Globl.

I have confirmed by instrumenting the linker
that the now-eliminated code paths were unreached.
And an additional check in the object file writing code
will help preserve that invariant.

There was a case in the Windows linker,
with internal linking and cgo,
where we were generating SNOPTRBSS symbols with data.
For now, convert those at the site at which they occur
into SNOPTRDATA, just like they were.

Does not pass toolstash-check,
but does generate identical linked binaries.

No compiler performance changes.

Change-Id: I77b071ab103685ff8e042cee9abb864385488872
Reviewed-on: https://go-review.googlesource.com/40864
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
8 years agomisc/android: don't let the Android exec wrapper hang indefinitely
Elias Naur [Mon, 1 May 2017 18:35:08 +0000 (20:35 +0200)]
misc/android: don't let the Android exec wrapper hang indefinitely

On Android, the exec wrapper passes on output from adb to its parent
process by passing on os.Stderr and os.Stdout to adb. If the adb
process somehow hangs, it will keep stderr and stdout will open, in turn
blocking go test from ever returning from its cmd.Wait() even though
it has killed the exec wrapper process.

Break the short circuit by introducing a wrapper between adb and the
exec wrapper, preventing os/exec.Run from passing along the raw
file descriptors for os.Stdout and os.Stderr.

(Hopefully) fixes occasional indefinite hangs on the Android builder.

Change-Id: I1188211fbde79b4a66bf93ff8e9d0091abf34560
Reviewed-on: https://go-review.googlesource.com/42271
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: use MOVL instead of MOVQ for small constants on amd64
Martin Möhrmann [Fri, 24 Mar 2017 07:13:17 +0000 (08:13 +0100)]
cmd/compile: use MOVL instead of MOVQ for small constants on amd64

The encoding of MOVL to a register is 2 bytes shorter than for MOVQ.
The upper 32bit are automatically zeroed when MOVL to a register is used.

Replaces 1657 MOVQ by MOVL in the go binary.
Reduces go binary size by 4 kilobyte.

name                   old time/op    new time/op    delta
BinaryTree17              1.93s ± 0%     1.93s ± 0%  -0.32%  (p=0.000 n=9+9)
Fannkuch11                2.66s ± 0%     2.48s ± 0%  -6.60%  (p=0.000 n=9+9)
FmtFprintfEmpty          31.8ns ± 0%    31.6ns ± 0%  -0.63%  (p=0.000 n=10+10)
FmtFprintfString         52.0ns ± 0%    51.9ns ± 0%  -0.19%  (p=0.000 n=10+10)
FmtFprintfInt            55.6ns ± 0%    54.6ns ± 0%  -1.80%  (p=0.002 n=8+10)
FmtFprintfIntInt         87.7ns ± 0%    84.8ns ± 0%  -3.31%  (p=0.000 n=9+9)
FmtFprintfPrefixedInt    98.9ns ± 0%   102.0ns ± 0%  +3.10%  (p=0.000 n=10+10)
FmtFprintfFloat           165ns ± 0%     164ns ± 0%  -0.61%  (p=0.000 n=10+10)
FmtManyArgs               368ns ± 0%     361ns ± 0%  -1.98%  (p=0.000 n=8+10)
GobDecode                4.53ms ± 0%    4.58ms ± 0%  +1.08%  (p=0.000 n=9+10)
GobEncode                3.74ms ± 0%    3.73ms ± 0%  -0.27%  (p=0.000 n=10+10)
Gzip                      164ms ± 0%     163ms ± 0%  -0.48%  (p=0.000 n=10+10)
Gunzip                   26.7ms ± 0%    26.6ms ± 0%  -0.13%  (p=0.000 n=9+10)
HTTPClientServer         30.4µs ± 1%    30.3µs ± 1%  -0.41%  (p=0.016 n=10+10)
JSONEncode               10.9ms ± 0%    11.0ms ± 0%  +0.70%  (p=0.000 n=10+10)
JSONDecode               36.8ms ± 0%    37.0ms ± 0%  +0.59%  (p=0.000 n=9+10)
Mandelbrot200            3.20ms ± 0%    3.21ms ± 0%  +0.44%  (p=0.000 n=9+10)
GoParse                  2.35ms ± 0%    2.35ms ± 0%  +0.26%  (p=0.000 n=10+9)
RegexpMatchEasy0_32      58.3ns ± 0%    58.4ns ± 0%  +0.17%  (p=0.000 n=10+10)
RegexpMatchEasy0_1K       138ns ± 0%     142ns ± 0%  +2.68%  (p=0.000 n=10+10)
RegexpMatchEasy1_32      55.1ns ± 0%    55.6ns ± 1%    ~     (p=0.104 n=10+10)
RegexpMatchEasy1_1K       242ns ± 0%     243ns ± 0%  +0.41%  (p=0.000 n=10+10)
RegexpMatchMedium_32     87.4ns ± 0%    89.9ns ± 0%  +2.86%  (p=0.000 n=10+10)
RegexpMatchMedium_1K     27.4µs ± 0%    27.4µs ± 0%  +0.15%  (p=0.000 n=10+10)
RegexpMatchHard_32       1.30µs ± 0%    1.32µs ± 1%  +1.91%  (p=0.000 n=10+10)
RegexpMatchHard_1K       39.0µs ± 0%    39.5µs ± 0%  +1.38%  (p=0.000 n=10+10)
Revcomp                   316ms ± 0%     319ms ± 0%  +1.13%  (p=0.000 n=9+8)
Template                 40.6ms ± 0%    40.6ms ± 0%    ~     (p=0.123 n=10+10)
TimeParse                 224ns ± 0%     224ns ± 0%    ~     (all equal)
TimeFormat                230ns ± 0%     225ns ± 0%  -2.17%  (p=0.000 n=10+10)

Change-Id: I32a099b65f9e6d4ad7288ed48546655c534757d8
Reviewed-on: https://go-review.googlesource.com/38630
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agoruntime: refactor cpu feature detection for 386 & amd64
Martin Möhrmann [Thu, 27 Apr 2017 06:30:27 +0000 (08:30 +0200)]
runtime: refactor cpu feature detection for 386 & amd64

Changes all cpu features to be detected and stored in bools in rt0_go.

Updates: #15403

Change-Id: I5a9961cdec789b331d09c44d86beb53833d5dc3e
Reviewed-on: https://go-review.googlesource.com/41950
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/internal/obj/x86: use LEAx rather than ADDx when calling DUFFxxxx via GOT
Michael Hudson-Doyle [Wed, 26 Apr 2017 23:56:42 +0000 (11:56 +1200)]
cmd/internal/obj/x86: use LEAx rather than ADDx when calling DUFFxxxx via GOT

DUFFZERO on 386 is not marked as clobbering flags, but rewriteToUseGot rewrote
"ADUFFZERO $offset" to "MOVL runtime.duffxxx@GOT, CX; ADDL $offset, CX; CALL CX"
which does. Luckily the fix is easier than figuring out what the problem was:
replace the ADDL $offset, CX with LEAL $offset(CX), CX.

On amd64 DUFFZERO clobbers flags, on arm, arm64 and ppc64 ADD does not clobber
flags and s390x does not use the duff functions, so I'm fairly confident this
is the only fix required.

I don't know how to write a test though.

Change-Id: I69b0958f5f45771d61db5f5ecb4ded94e8960d4d
Reviewed-on: https://go-review.googlesource.com/41821
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/internal/obj/x86: fix ANDPS encoding
Damien Lespiau [Fri, 28 Apr 2017 14:20:38 +0000 (15:20 +0100)]
cmd/internal/obj/x86: fix ANDPS encoding

ANDPS, like all others PS (Packed Single precision floats) instructions,
need Ym: they don't use the 0x66 prefix.

From the manual:

    NP 0F 54 /r        ANDPS xmm1, xmm2/m128

NP meaning, quoting the manual:

  NP - Indicates the use of 66/F2/F3 prefixes (beyond those already part
  of the instructions opcode) are not allowed with the instruction.

And indeed, the same instruction prefixed by 0x66 is ANDPD.

Updates #14069

Change-Id: If312a6f1e77113ab8c0febe66bdb1b4171e41e0a
Reviewed-on: https://go-review.googlesource.com/42090
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: avoid giant init functions due to many user inits
Josh Bleecher Snyder [Mon, 24 Apr 2017 00:31:15 +0000 (17:31 -0700)]
cmd/compile: avoid giant init functions due to many user inits

We generate code that calls each user init function one at a time.
When there are lots of user init functions,
usually due to generated code, like test/rotate* or
github.com/juju/govmomi/vim25/types,
we can end up with a giant function,
which can be slow to compile.

This CL puts in an escape valve.
When there are more than 500 functions, instead of doing:

init.0()
init.1()
// ...

we construct a static array of functions:

var fns = [...]func(){init.0, init.1, ... }

and call them in a loop.

This generates marginally bigger, marginally worse code,
so we restrict it to cases in which it might start to matter.

500 was selected as a mostly arbitrary threshold for "lots".
Each call uses two Progs, one for PCDATA and one for the call,
so at 500 calls we use ~1000 Progs.
At concurrency==8, we get a Prog cache of about
1000 Progs per worker.
So a threshold of 500 should more or less avoid
exhausting the Prog cache in most cases.

Change-Id: I276b887173ddbf65b2164ec9f9b5eb04d8c753c2
Reviewed-on: https://go-review.googlesource.com/41500
Reviewed-by: Keith Randall <khr@golang.org>
8 years agoruntime: use 64 bit calculation in overLoadFactor
Josh Bleecher Snyder [Mon, 1 May 2017 14:36:43 +0000 (07:36 -0700)]
runtime: use 64 bit calculation in overLoadFactor

overLoadFactor used a uintptr for its calculations.
When the number of potential buckets was large,
perhaps due to a coding error or corrupt/malicious user input
leading to a very large map size hint,
this led to overflow on 32 bit systems.
This overflow resulted in an infinite loop.

Prevent it by always using a 64 bit calculation.

Updates #20195

Change-Id: Iaabc710773cd5da6754f43b913478cc5562d89a2
Reviewed-on: https://go-review.googlesource.com/42185
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: add minor bit twiddling optimizations
Josh Bleecher Snyder [Tue, 25 Apr 2017 22:55:52 +0000 (15:55 -0700)]
cmd/compile: add minor bit twiddling optimizations

Noticed while adding to the bitset implementation
in cmd/compile/internal/gc.

The (Com (Const)) optimizations were already present
in the AMD64 lowered optimizations.
They trigger 118, 44, 262, and 108 times
respectively for int sizes 8, 16, 32, and 64
in a run of make.bash.

The (Or (And)) optimization is new.
It triggers 3 times for int size 8
and once for int size 64 during make.bash,
in packages internal/poll, reflect,
encoding/asn1, and go/types,
so there is a bit of natural test coverage.

Change-Id: I44072864ff88831d5ec7dce37c516d29df056e98
Reviewed-on: https://go-review.googlesource.com/41758
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: add comments to trackAllTypes
Josh Bleecher Snyder [Mon, 1 May 2017 15:06:48 +0000 (08:06 -0700)]
cmd/compile: add comments to trackAllTypes

Per gri's suggestion on CL 41623,
add a comment to trackAllTypes
about the trade-offs of enabling it.

Change-Id: Iec42b0da7933543200729003d1b2c6e0d9dcc5f0
Reviewed-on: https://go-review.googlesource.com/42186
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/asm: enable CMPPS, CMPPD, CMPSS and CMPSD encoding tests
Damien Lespiau [Mon, 1 May 2017 13:49:06 +0000 (14:49 +0100)]
cmd/asm: enable CMPPS, CMPPD, CMPSS and CMPSD encoding tests

The generated test cases had their arguments reversed, putting them back
in order makes those tests pass.

  CMPPS SRC, DEST, CC

Change-Id: Ie15021edc533d5681a6a78d10d88b665e3de9017
Reviewed-on: https://go-review.googlesource.com/42097
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoRevert "cmd/link: replace SHIDDEN bit in SymKind with a bit of Attribute"
Michael Hudson-Doyle [Mon, 1 May 2017 00:56:09 +0000 (00:56 +0000)]
Revert "cmd/link: replace SHIDDEN bit in SymKind with a bit of Attribute"

This reverts commit a69222d949bcc2c4453248ee233b63b75d79967e.

Reason for revert: broke ppc64le

Change-Id: I57d275177e90f036caf7dbade9669b8121dfa437
Reviewed-on: https://go-review.googlesource.com/42194
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>