]> Cypherpunks repositories - gostls13.git/log
gostls13.git
6 years agocmd/link: add new loading package for XCOFF files
Clément Chigot [Fri, 28 Sep 2018 14:19:18 +0000 (16:19 +0200)]
cmd/link: add new loading package for XCOFF files

This commit adds the new package cmd/link/internal/loadxcoff.

It also adds a new symbol kind in cmd/link/internal/sym package, which
aims to represent TOC entries for XCOFF files.

cmd/dist is updated to add this new package and cmd/internal/xcoff during
the bootstrap.

Updates: #25893

Change-Id: I42b6578cf0ba4cc28ad4aa98122a91ab1d1bbf6e
Reviewed-on: https://go-review.googlesource.com/c/138728
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agointernal/traceparser: provide parser that uses less space and parses segments of...
Peter Weinberger [Wed, 26 Sep 2018 14:10:48 +0000 (10:10 -0400)]
internal/traceparser: provide parser that uses less space and parses segments of runtime trace files

Traceparser generally takes 20-30% less space than internal/trace. The only
user of these pakcages is cmd/trace, and the new package lets it handle some
trace files that were too large. The new parser will also convert segments
of the raw trace file (e.g. the last 10 seconds) to Events. Trace files from
go 1.8 and before are not supported.

Change-Id: If83fa183246db8f75182ccd3ba8df07673c0ebd0
Reviewed-on: https://go-review.googlesource.com/c/137635
Run-TryBot: Peter Weinberger <pjw@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/compile/internal/gc: remove dead code from stringtoarraylit
Iskander Sharipov [Wed, 25 Jul 2018 16:06:01 +0000 (19:06 +0300)]
cmd/compile/internal/gc: remove dead code from stringtoarraylit

The code path for []byte is unused.
Rename function to stringtoruneslit to reflect change in the behavior.

Note that removed code had a bug in it,
it used [0] index instead of [i] inside a loop body.

Change-Id: I58ece5d9d3835887b014446f8a7d3e7fc2fdcaa3
Reviewed-on: https://go-review.googlesource.com/c/125796
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agoruntime: use multiplication with overflow check for makeslice
Martin Möhrmann [Mon, 15 Oct 2018 22:27:42 +0000 (00:27 +0200)]
runtime: use multiplication with overflow check for makeslice

This improves performance for slices with an element size larger
than 32 bytes and removes loading a value from the maxElems
array for smaller element sizes.

name                 old time/op  new time/op  delta
MakeSlice/Byte       18.0ns ± 4%  18.0ns ± 2%     ~     (p=0.575 n=20+17)
MakeSlice/Int16      21.8ns ± 2%  21.6ns ± 1%   -0.63%  (p=0.035 n=20+19)
MakeSlice/Int        42.0ns ± 2%  41.6ns ± 1%     ~     (p=0.121 n=20+18)
MakeSlice/Ptr        62.6ns ± 2%  62.4ns ± 2%     ~     (p=0.491 n=20+18)
MakeSlice/Struct/24  57.4ns ± 3%  56.0ns ± 2%   -2.40%  (p=0.000 n=19+19)
MakeSlice/Struct/32  62.1ns ± 2%  60.6ns ± 3%   -2.43%  (p=0.000 n=20+20)
MakeSlice/Struct/40  77.3ns ± 3%  68.9ns ± 3%  -10.91%  (p=0.000 n=20+20)

Updates #21588

Change-Id: Ie12807bf8f77c0e15453413f47e3d7de771b798f
Reviewed-on: https://go-review.googlesource.com/c/142377
Run-TryBot: Martin Möhrmann <martisch@uos.de>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agoruntime: use multiplication with overflow check for growslice
Martin Möhrmann [Mon, 22 Oct 2018 18:22:55 +0000 (20:22 +0200)]
runtime: use multiplication with overflow check for growslice

This improves performance for slices with an element size larger
than 32 bytes and removes loading a value from the maxElems
array for smaller element sizes.

name                 old time/op  new time/op  delta
GrowSlice/Byte       41.4ns ± 2%  41.5ns ± 1%    ~     (p=0.366 n=10+9)
GrowSlice/Int16      51.1ns ± 2%  51.0ns ± 2%    ~     (p=0.985 n=10+10)
GrowSlice/Int        64.0ns ± 1%  64.2ns ± 1%    ~     (p=0.180 n=10+10)
GrowSlice/Ptr        90.8ns ± 1%  90.7ns ± 1%    ~     (p=0.858 n=9+10)
GrowSlice/Struct/24   108ns ± 0%   108ns ± 2%    ~     (p=0.488 n=8+9)
GrowSlice/Struct/32   118ns ± 2%   117ns ± 2%    ~     (p=0.327 n=10+10)
GrowSlice/Struct/40   159ns ± 1%   148ns ± 1%  -6.87%  (p=0.000 n=10+9)

Updates #21588

Change-Id: I443b82972d379b1befa791f9ee468b3adc6bb760
Reviewed-on: https://go-review.googlesource.com/c/143798
Run-TryBot: Martin Möhrmann <martisch@uos.de>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoruntime: use multiplication with overflow check for makemap
Martin Möhrmann [Mon, 22 Oct 2018 18:45:04 +0000 (20:45 +0200)]
runtime: use multiplication with overflow check for makemap

This improves performance for maps with a bucket size
(key+value*8 bytes) larger than 32 bytes and removes loading
a value from the maxElems array for smaller bucket sizes.

name                old time/op  new time/op  delta
MakeMap/[Byte]Byte  93.5ns ± 1%  91.8ns ± 1%  -1.83%  (p=0.000 n=10+10)
MakeMap/[Int]Int     134ns ± 1%   127ns ± 2%  -5.61%  (p=0.000 n=9+10)

Updates #21588

Change-Id: I53f77186769c4bd0f2b90f3c6c17df643b060e39
Reviewed-on: https://go-review.googlesource.com/c/143797
Run-TryBot: Martin Möhrmann <martisch@uos.de>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agointernal/syscall/unix, net: improve interface_aix.go
Clément Chigot [Thu, 11 Oct 2018 12:35:29 +0000 (14:35 +0200)]
internal/syscall/unix, net: improve interface_aix.go

This commit improves the interface_aix.go file, based on feedbacks about
CL 138724.

To retrieve MTU, ioctl is needed. It's implemented inside
internal/syscall/unix.

Change-Id: Ic583d26b93935a32a5f1eb5a2170b86e80a4a85e
Reviewed-on: https://go-review.googlesource.com/c/142157
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile: use proved bounds to remove signed division fix-ups
ChrisALiles [Mon, 6 Aug 2018 09:50:38 +0000 (19:50 +1000)]
cmd/compile: use proved bounds to remove signed division fix-ups

prove is able to find 94 occurrences in std cmd where a divisor
can't have the value -1. The change removes
the extraneous fix-up code for these cases.

Fixes #25239

Change-Id: Ic184de971f47cc57c702eb72805b8e291c14035d
Reviewed-on: https://go-review.googlesource.com/c/130215
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/dist: copy cmd/internal/xcoff to bootstrap directory
Ian Lance Taylor [Tue, 23 Oct 2018 00:14:36 +0000 (17:14 -0700)]
cmd/dist: copy cmd/internal/xcoff to bootstrap directory

Fixes build.

Change-Id: Id3d78ae9e4669371b4ebf29f69217d0809ffb829
Reviewed-on: https://go-review.googlesource.com/c/143917
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotext/template: removed truncation of context in error message
Alexandre Maari [Mon, 15 Oct 2018 17:23:09 +0000 (13:23 -0400)]
text/template: removed truncation of context in error message

Fixes #27930

Change-Id: I31ad3fdb74d74152268c59ae4c651cc4c8c1716d
Reviewed-on: https://go-review.googlesource.com/c/142217
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/go: update private copy of goosList
Ian Lance Taylor [Mon, 22 Oct 2018 19:56:01 +0000 (12:56 -0700)]
cmd/go: update private copy of goosList

This copies the change to goosList in CL 138115 to the private copy in
cmd/go.

The change introducing the private copy was apparently not made with
Gerrit, but can be seen at
https://github.com/golang/vgo/commit/08359e782fb601567c57f56beb540841c2416d92.
That change says "This is adapted from code in go/build and the rest
of cmd/go. At some point, we should deduplicate them."

Doing another copy for now, rather than something more complex
involving cmd/dist, pending that deduplication.

Change-Id: I9b6e1f63a3a68c002b60a9a97aa367c5cc7801c9
Reviewed-on: https://go-review.googlesource.com/c/143759
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/cgo: add AIX operating system
Clément Chigot [Fri, 28 Sep 2018 15:15:49 +0000 (17:15 +0200)]
cmd/cgo: add AIX operating system

This commit adds AIX operating system to cmd/cgo package for ppc64
architecture.

It doesn't fully adapt cgo tool to AIX. But it allows to use
go tool cgo -godefs which is really usefull for others packages.

Update: #25893

Change-Id: I38e289cf0122d143ba100986d08229b51b03ddfc
Reviewed-on: https://go-review.googlesource.com/c/138731
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoencoding/gob: delete out of memory test
Rob Pike [Mon, 22 Oct 2018 20:01:35 +0000 (07:01 +1100)]
encoding/gob: delete out of memory test

Now that the library allows much larger data, it can kill
machines with less memory.

Fixes #28321

Change-Id: I98e1a5fdf812fd75adfb22bf01542423de405fe2
Reviewed-on: https://go-review.googlesource.com/c/143817
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
6 years agogo/types: report error for invalid use of ... in parameter lists
Robert Griesemer [Mon, 22 Oct 2018 21:07:36 +0000 (14:07 -0700)]
go/types: report error for invalid use of ... in parameter lists

The parser accepts ...T types in parameter lists whereever a type
is permitted; this matches the syntax and allows for more tolerant
parsing and error recovery.

go/types on the other hand assumed that the parser would report
those errors and assumed any outstanding such errors would be due
to otherwise manipulated ASTs leading to invalid ASTs.

go/types further assumed that a parameter list (a, b, c ...int)
was permitted (a couple of tests used such parameter lists).

With this CL, go/types now correctly refuses invalid parameter lists.

Fixes #28281.

Change-Id: Ib788255f7b7819fdb972c7801bb153a53ce2ddf7
Reviewed-on: https://go-review.googlesource.com/c/143857
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/go: add AIX operating system
Clément Chigot [Fri, 28 Sep 2018 15:21:57 +0000 (17:21 +0200)]
cmd/go: add AIX operating system

This commit adds AIX operating system to cmd/go package for ppc64
architecture.

Updates: #25893

Change-Id: I2605d10a7833fa2eb197f6db4a52d5919cf93614
Reviewed-on: https://go-review.googlesource.com/c/138732
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: use the new log functions
Genevieve Luyt [Mon, 22 Oct 2018 17:39:13 +0000 (13:39 -0400)]
doc: use the new log functions

The log interface was changed in https://golang.org/cl/2419042.

Change-Id: I3eaddd8a5cfcae961db16555fb1b0ce6770b6334
Reviewed-on: https://go-review.googlesource.com/c/143777
Reviewed-by: Rob Pike <r@golang.org>
6 years agocmd/compile: s/eqtype/types.Identical/ (fix build)
Robert Griesemer [Mon, 22 Oct 2018 18:21:38 +0000 (11:21 -0700)]
cmd/compile: s/eqtype/types.Identical/ (fix build)

TBR=khr

Change-Id: Ia5a08f1acd9f37c466829754990733330264f546
Reviewed-on: https://go-review.googlesource.com/c/143758
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agogo/types: copy embedded methods unchanged when completing interfaces
Robert Griesemer [Mon, 22 Oct 2018 17:54:38 +0000 (10:54 -0700)]
go/types: copy embedded methods unchanged when completing interfaces

The existing code adjusted the receivers of embedded interface methods
to match the embedding interface type. That required cloning (shallow
copying) the embedded methods and destroyed their object identity in
the process. Don't do this anymore. The consequence to clients is that
they might see different methods of an interface having different
receiver types; they are always the type of the interface that explicitly
declared the method (which is what one usually would want, anyway).

Fixes #28282.

Change-Id: I2e6f1497f46affdf7510547a64601de3787367db
Reviewed-on: https://go-review.googlesource.com/c/143757
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/compile,cmd/link: remove statictmp variables from symbol table.
Alessandro Arzilli [Wed, 17 Oct 2018 07:36:06 +0000 (09:36 +0200)]
cmd/compile,cmd/link: remove statictmp variables from symbol table.

Removes statictmp variables from debug_info and the final symbol table.

Fixes #27800

Change-Id: I302c59a04bc3f460e7085fef241f937bbf30421d
Reviewed-on: https://go-review.googlesource.com/c/142577
Run-TryBot: Alessandro Arzilli <alessandro.arzilli@gmail.com>
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
6 years agocmd/compile: in append(f()), type convert appended items
Keith Randall [Tue, 16 Oct 2018 21:31:49 +0000 (14:31 -0700)]
cmd/compile: in append(f()), type convert appended items

The second and subsequent return values from f() need to be
converted to the element type of the first return value from f()
(which must be a slice).

Fixes #22327

Change-Id: I5c0a424812c82c1b95b6d124c5626cfc4408bdb6
Reviewed-on: https://go-review.googlesource.com/c/142718
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agocmd/internal/obj/arm64: reclassify 32-bit/64-bit constants
fanzha02 [Fri, 22 Jun 2018 12:08:47 +0000 (12:08 +0000)]
cmd/internal/obj/arm64: reclassify 32-bit/64-bit constants

Current assembler saves constants in Offset which type is int64,
causing 32-bit constants have a incorrect class. This CL reclassifies
constants when opcodes are 32-bit variant, like MOVW, ANDW and
ADDW, etc. Besides, this CL encodes some constants of ADDCON class
as MOVs instructions.

This CL changes the assembler behavior as follows.

1. go assembler ADDW $MOVCON, Rn, Rd
   previous version: MOVD $MOVCON, Rtmp; ADDW Rtmp, Rn, Rd
   current version: MOVW $MOVCON, Rtmp; ADDW Rtmp, Rn, Rd

2. go assembly MOVW $0xaaaaffff, R1
   previous version: treats $0xaaaaffff as VCON, encodes it as MOVW 0x994, R1 (loads it from pool).
   current version: treats $0xaaaaffff as MOVCON, and encodes it into MOVW instructions.

3. go assembly MOVD $0x210000, R1
   previous version: treats $0x210000 as ADDCON, loads it from pool
   current version: treats $0x210000 as MOVCON, and encodes it into MOVD instructions.

Add the test cases.

1. Binary size before/after.
binary                          size change
pkg/linux_arm64                 -1.534KB
pkg/tool/linux_arm64            -0.718KB
go                              -0.32KB
gofmt                           no change

2. go1 benchmark result.
name                     old time/op    new time/op    delta
BinaryTree17-8              6.26s ± 1%     6.28s ± 1%     ~     (p=0.105 n=10+10)
Fannkuch11-8                5.40s ± 0%     5.39s ± 0%   -0.29%  (p=0.028 n=9+10)
FmtFprintfEmpty-8          94.5ns ± 0%    95.0ns ± 0%   +0.51%  (p=0.000 n=10+9)
FmtFprintfString-8          163ns ± 1%     159ns ± 1%   -2.06%  (p=0.000 n=10+9)
FmtFprintfInt-8             200ns ± 1%     196ns ± 1%   -1.99%  (p=0.000 n=9+10)
FmtFprintfIntInt-8          292ns ± 3%     284ns ± 1%   -2.87%  (p=0.001 n=10+9)
FmtFprintfPrefixedInt-8     422ns ± 1%     420ns ± 1%   -0.59%  (p=0.015 n=10+10)
FmtFprintfFloat-8           458ns ± 0%     463ns ± 1%   +1.19%  (p=0.000 n=9+10)
FmtManyArgs-8              1.37µs ± 1%    1.35µs ± 1%   -1.85%  (p=0.000 n=10+10)
GobDecode-8                15.5ms ± 1%    15.3ms ± 1%   -1.82%  (p=0.000 n=10+10)
GobEncode-8                11.7ms ± 5%    11.7ms ± 2%     ~     (p=0.549 n=10+9)
Gzip-8                      622ms ± 0%     624ms ± 0%   +0.23%  (p=0.000 n=10+9)
Gunzip-8                   73.6ms ± 0%    73.8ms ± 1%     ~     (p=0.077 n=9+9)
HTTPClientServer-8          115µs ± 1%     115µs ± 1%     ~     (p=0.796 n=10+10)
JSONEncode-8               31.1ms ± 2%    28.7ms ± 1%   -7.98%  (p=0.000 n=10+9)
JSONDecode-8                145ms ± 0%     145ms ± 1%     ~     (p=0.447 n=9+10)
Mandelbrot200-8            9.67ms ± 0%    9.60ms ± 0%   -0.76%  (p=0.000 n=9+9)
GoParse-8                  7.56ms ± 1%    7.58ms ± 0%   +0.21%  (p=0.035 n=10+9)
RegexpMatchEasy0_32-8       208ns ±10%     222ns ± 0%     ~     (p=0.531 n=10+6)
RegexpMatchEasy0_1K-8       699ns ± 4%     694ns ± 4%     ~     (p=0.868 n=10+10)
RegexpMatchEasy1_32-8       186ns ± 8%     190ns ±12%     ~     (p=0.955 n=10+10)
RegexpMatchEasy1_1K-8      1.13µs ± 1%    1.05µs ± 2%   -6.64%  (p=0.000 n=10+10)
RegexpMatchMedium_32-8      316ns ± 7%     288ns ± 1%   -8.68%  (p=0.000 n=10+7)
RegexpMatchMedium_1K-8     90.2µs ± 0%    85.5µs ± 2%   -5.19%  (p=0.000 n=10+10)
RegexpMatchHard_32-8       5.53µs ± 0%    3.90µs ± 0%  -29.52%  (p=0.000 n=10+10)
RegexpMatchHard_1K-8        119µs ± 0%     124µs ± 0%   +4.29%  (p=0.000 n=9+10)
Revcomp-8                   1.07s ± 0%     1.07s ± 0%     ~     (p=0.094 n=9+9)
Template-8                  162ms ± 1%     160ms ± 2%     ~     (p=0.089 n=10+10)
TimeParse-8                 756ns ± 2%     763ns ± 1%     ~     (p=0.158 n=10+10)
TimeFormat-8                758ns ± 1%     746ns ± 1%   -1.52%  (p=0.000 n=10+10)

name                     old speed      new speed      delta
GobDecode-8              49.4MB/s ± 1%  50.3MB/s ± 1%   +1.84%  (p=0.000 n=10+10)
GobEncode-8              65.6MB/s ± 5%  65.4MB/s ± 2%     ~     (p=0.549 n=10+9)
Gzip-8                   31.2MB/s ± 0%  31.1MB/s ± 0%   -0.24%  (p=0.000 n=9+9)
Gunzip-8                  264MB/s ± 0%   263MB/s ± 1%     ~     (p=0.073 n=9+9)
JSONEncode-8             62.3MB/s ± 2%  67.7MB/s ± 1%   +8.67%  (p=0.000 n=10+9)
JSONDecode-8             13.4MB/s ± 0%  13.4MB/s ± 1%     ~     (p=0.508 n=9+10)
GoParse-8                7.66MB/s ± 1%  7.64MB/s ± 0%   -0.23%  (p=0.049 n=10+9)
RegexpMatchEasy0_32-8     154MB/s ± 9%   143MB/s ± 3%     ~     (p=0.303 n=10+7)
RegexpMatchEasy0_1K-8    1.46GB/s ± 4%  1.47GB/s ± 4%     ~     (p=0.912 n=10+10)
RegexpMatchEasy1_32-8     172MB/s ± 9%   170MB/s ±12%     ~     (p=0.971 n=10+10)
RegexpMatchEasy1_1K-8     908MB/s ± 1%   972MB/s ± 2%   +7.12%  (p=0.000 n=10+10)
RegexpMatchMedium_32-8   3.17MB/s ± 7%  3.46MB/s ± 1%   +9.14%  (p=0.000 n=10+7)
RegexpMatchMedium_1K-8   11.3MB/s ± 0%  12.0MB/s ± 2%   +5.51%  (p=0.000 n=10+10)
RegexpMatchHard_32-8     5.78MB/s ± 0%  8.21MB/s ± 0%  +41.93%  (p=0.000 n=9+10)
RegexpMatchHard_1K-8     8.62MB/s ± 0%  8.27MB/s ± 0%   -4.11%  (p=0.000 n=9+10)
Revcomp-8                 237MB/s ± 0%   237MB/s ± 0%     ~     (p=0.081 n=9+9)
Template-8               12.0MB/s ± 1%  12.1MB/s ± 2%     ~     (p=0.072 n=10+10)

Change-Id: I080801f520366b42d5f9699954bd33106976a81b
Reviewed-on: https://go-review.googlesource.com/c/120661
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/compile: intrinsify math/big.mulWW on ppc64x
Carlos Eduardo Seo [Tue, 16 Oct 2018 01:29:05 +0000 (22:29 -0300)]
cmd/compile: intrinsify math/big.mulWW on ppc64x

This change implements mulWW as an intrinsic for ppc64x. Performance
numbers below:

name                            old time/op    new time/op    delta
QuoRem                            4.54µs ±45%    3.22µs ± 0%  -29.22%  (p=0.029 n=4+4)
ModSqrt225_Tonelli                 765µs ± 3%     757µs ± 0%   -1.02%  (p=0.029 n=4+4)
ModSqrt225_3Mod4                   231µs ± 0%     231µs ± 0%   -0.10%  (p=0.029 n=4+4)
ModSqrt231_Tonelli                 789µs ± 0%     788µs ± 0%   -0.14%  (p=0.029 n=4+4)
ModSqrt231_5Mod8                   267µs ± 0%     267µs ± 0%   -0.13%  (p=0.029 n=4+4)
Sqrt                              49.5µs ±17%    45.3µs ± 0%   -8.48%  (p=0.029 n=4+4)
IntSqr/1                          32.2ns ±22%    24.2ns ± 0%  -24.79%  (p=0.029 n=4+4)
IntSqr/2                          60.6ns ± 0%    60.9ns ± 0%   +0.50%  (p=0.029 n=4+4)
IntSqr/3                          82.8ns ± 0%    83.3ns ± 0%   +0.51%  (p=0.029 n=4+4)
IntSqr/5                           122ns ± 0%     121ns ± 0%   -1.22%  (p=0.029 n=4+4)
IntSqr/8                           227ns ± 0%     226ns ± 0%   -0.44%  (p=0.029 n=4+4)
IntSqr/10                          300ns ± 0%     298ns ± 0%   -0.67%  (p=0.029 n=4+4)
IntSqr/20                         1.02µs ± 0%    0.89µs ± 0%  -13.08%  (p=0.029 n=4+4)
IntSqr/30                         1.73µs ± 0%    1.51µs ± 0%  -12.73%  (p=0.029 n=4+4)
IntSqr/50                         3.69µs ± 1%    3.29µs ± 0%  -10.70%  (p=0.029 n=4+4)
IntSqr/80                         7.64µs ± 0%    7.04µs ± 0%   -7.91%  (p=0.029 n=4+4)
IntSqr/100                        11.1µs ± 0%    10.3µs ± 0%   -7.04%  (p=0.029 n=4+4)
IntSqr/200                        37.9µs ± 0%    36.4µs ± 0%   -4.13%  (p=0.029 n=4+4)
IntSqr/300                        69.4µs ± 0%    66.0µs ± 0%   -4.94%  (p=0.029 n=4+4)
IntSqr/500                         174µs ± 0%     168µs ± 0%   -3.10%  (p=0.029 n=4+4)
IntSqr/800                         347µs ± 0%     333µs ± 0%   -4.06%  (p=0.029 n=4+4)
IntSqr/1000                        524µs ± 0%     507µs ± 0%   -3.21%  (p=0.029 n=4+4)

Change-Id: If067452f5b6579ad3a2e9daa76a7ffe6fceae1bb
Reviewed-on: https://go-review.googlesource.com/c/143217
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
6 years agoencoding/gob: increase "tooBig" from 1GB to 8GB on 64-bit machines
Rob Pike [Mon, 22 Oct 2018 03:17:06 +0000 (14:17 +1100)]
encoding/gob: increase "tooBig" from 1GB to 8GB on 64-bit machines

A little shift magic makes it easy to adjust the maximum buffer
size on machines with larger integers.

Fixes #27635

Change-Id: I1f26b07a363fbb9730df2377052475fa88bbb781
Reviewed-on: https://go-review.googlesource.com/c/143678
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: tweak example in Effective Go
Rob Pike [Mon, 22 Oct 2018 03:04:18 +0000 (14:04 +1100)]
doc: tweak example in Effective Go

A prior attempt at addressing the issue got bogged down in an
endless conversation around the subtleties of Read semantics.
Let's not go there.

Instead, we put the issue to bed, perhaps not in perfect comfort
but well enough, by moving a line of the example so that even
if there is a "benign" error as the issue suggests, the loop
terminates with n and err correctly set.

Fixes #27818

Change-Id: I4a32d56c9e782f17578565d90b22ce531e3d8667
Reviewed-on: https://go-review.googlesource.com/c/143677
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/internal/xcoff: don't use io.SeekCurrent for go1.4 compatibility
Tobias Klauser [Fri, 19 Oct 2018 16:09:42 +0000 (18:09 +0200)]
cmd/internal/xcoff: don't use io.SeekCurrent for go1.4 compatibility

This fixes the build with go1.4 during bootstrap.

Change-Id: I35fa1dc3d877d31d5ffe515f5f19188f4a8b77f4
Reviewed-on: https://go-review.googlesource.com/c/143377
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: update docs.html with new tour import path
Andrew Bonventre [Fri, 12 Oct 2018 19:39:30 +0000 (15:39 -0400)]
doc: update docs.html with new tour import path

As of golang.org/cl/141857 the import path has changed from
golang.org/x/tour/gotour to golang.org/x/tour

Change-Id: Ib54ab2e50188ef66c8a5c45136babfa49ad6934a
Reviewed-on: https://go-review.googlesource.com/c/141917
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agomisc/wasm: fix panic on os.Stdout.Sync() in the browser
Nick Craig-Wood [Thu, 18 Oct 2018 15:26:56 +0000 (16:26 +0100)]
misc/wasm: fix panic on os.Stdout.Sync() in the browser

Before this change running os.Stdout.Sync() in the browser would panic
the application with:

panic: syscall/js: Value.Call: property fsync is not a function, got undefined

Afterwards Sync() becomes a noop for compatibility reasons.

Change-Id: I1fcef694beb35fdee3173f87371e1ff233b15d32
Reviewed-on: https://go-review.googlesource.com/c/143138
Reviewed-by: Richard Musiol <neelance@gmail.com>
Run-TryBot: Richard Musiol <neelance@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agogo/types: temporarily disable a verification in Stdlib test
Robert Griesemer [Fri, 19 Oct 2018 22:11:37 +0000 (15:11 -0700)]
go/types: temporarily disable a verification in Stdlib test

Keeps the longtest builder green for now.
Proper fix to come ASAP.

Also, reword an internal comment that could easily be misread.

Updates #28282.

Change-Id: I8f41c9faa5a3eb638e6204bae3ff374ed49e5177
Reviewed-on: https://go-review.googlesource.com/c/143478
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: move argument stack construction to SSA generation
Josh Bleecher Snyder [Sun, 6 May 2018 19:58:53 +0000 (12:58 -0700)]
cmd/compile: move argument stack construction to SSA generation

The goal of this change is to move work from walk to SSA,
and simplify things along the way.

This is hard to accomplish cleanly with small incremental changes,
so this large commit message aims to provide a roadmap to the diff.

High level description:

Prior to this change, walk was responsible for constructing (most of) the stack for function calls.
ascompatte gathered variadic arguments into a slice.
It also rewrote n.List from a list of arguments to a list of assignments to stack slots.
ascompatte was called multiple times to handle the receiver in a method call.
reorder1 then introduced temporaries into n.List as needed to avoid smashing the stack.
adjustargs then made extra stack space for go/defer args as needed.

Node to SSA construction evaluated all the statements in n.List,
and issued the function call, assuming that the stack was correctly constructed.
Intrinsic calls had to dig around inside n.List to extract the arguments,
since intrinsics don't use the stack to make function calls.

This change moves stack construction to the SSA construction phase.
ascompatte, now called walkParams, does all the work that ascompatte and reorder1 did.
It handles variadic arguments, inserts the method receiver if needed, and allocates temporaries.
It does not, however, make any assignments to stack slots.
Instead, it moves the function arguments to n.Rlist, leaving assignments to temporaries in n.List.
(It would be better to use Ninit instead of List; future work.)
During SSA construction, after doing all the temporary assignments in n.List,
the function arguments are assigned to stack slots by
constructing the appropriate SSA Value, using (*state).storeArg.
SSA construction also now handles adjustments for go/defer args.
This change also simplifies intrinsic calls, since we no longer need to undo walk's work.

Along the way, we simplify nodarg by pushing the fp==1 case to its callers, where it fits nicely.

Generated code differences:

There were a few optimizations applied along the way, the old way.
f(g()) was rewritten to do a block copy of function results to function arguments.
And reorder1 avoided introducing the final "save the stack" temporary in n.List.

The f(g()) block copy optimization never actually triggered; the order pass rewrote away g(), so that has been removed.

SSA optimizations mostly obviated the need for reorder1's optimization of avoiding the final temporary.
The exception was when the temporary's type was not SSA-able;
in that case, we got a Move into an autotmp and then an immediate Move onto the stack,
with the autotmp never read or used again.
This change introduces a new rewrite rule to detect such pointless double Moves
and collapse them into a single Move.
This is actually more powerful than the original optimization,
since the original optimization relied on the imprecise Node.HasCall calculation.

The other significant difference in the generated code is that the stack is now constructed
completely in SP-offset order. Prior to this change, the stack was constructed somewhat
haphazardly: first the final argument that Node.HasCall deemed to require a temporary,
then other arguments, then the method receiver, then the defer/go args.
SP-offset is probably a good default order. See future work.

There are a few minor object file size changes as a result of this change.
I investigated some regressions in early versions of this change.

One regression (in archive/tar) was the addition of a single CMPQ instruction,
which would be eliminated were this TODO from flagalloc to be done:
// TODO: Remove original instructions if they are never used.

One regression (in text/template) was an ADDQconstmodify that is now
a regular MOVQLoad+ADDQconst+MOVQStore, due to an unlucky change
in the order in which arguments are written. The argument change
order can also now be luckier, so this appears to be a wash.

All in all, though there will be minor winners and losers,
this change appears to be performance neutral.

Future work:

Move loading the result of function calls to SSA construction; eliminate OINDREGSP.

Consider pushing stack construction deeper into SSA world, perhaps in an arch-specific pass.
Among other benefits, this would make it easier to transition to a new calling convention.
This would require rethinking the handling of stack conflicts and is non-trivial.

Figure out some clean way to indicate that stack construction Stores/Moves
do not alias each other, so that subsequent passes may do things like
CSE+tighten shared stack setup, do DSE using non-first Stores, etc.
This would allow us to eliminate the minor text/template regression.

Possibly make assignments to stack slots not treated as statements by DWARF.

Compiler benchmarks:

name        old time/op       new time/op       delta
Template          182ms ± 2%        179ms ± 2%  -1.69%  (p=0.000 n=47+48)
Unicode          86.3ms ± 5%       85.1ms ± 4%  -1.36%  (p=0.001 n=50+50)
GoTypes           646ms ± 1%        642ms ± 1%  -0.63%  (p=0.000 n=49+48)
Compiler          2.89s ± 1%        2.86s ± 2%  -1.36%  (p=0.000 n=48+50)
SSA               8.47s ± 1%        8.37s ± 2%  -1.22%  (p=0.000 n=47+50)
Flate             122ms ± 2%        121ms ± 2%  -0.66%  (p=0.000 n=47+45)
GoParser          147ms ± 2%        146ms ± 2%  -0.53%  (p=0.006 n=46+49)
Reflect           406ms ± 2%        403ms ± 2%  -0.76%  (p=0.000 n=48+43)
Tar               162ms ± 3%        162ms ± 4%    ~     (p=0.191 n=46+50)
XML               223ms ± 2%        222ms ± 2%  -0.37%  (p=0.031 n=45+49)
[Geo mean]        382ms             378ms       -0.89%

name        old user-time/op  new user-time/op  delta
Template          219ms ± 3%        216ms ± 3%  -1.56%  (p=0.000 n=50+48)
Unicode           109ms ± 6%        109ms ± 5%    ~     (p=0.190 n=50+49)
GoTypes           836ms ± 2%        828ms ± 2%  -0.96%  (p=0.000 n=49+48)
Compiler          3.87s ± 2%        3.80s ± 1%  -1.81%  (p=0.000 n=49+46)
SSA               12.0s ± 1%        11.8s ± 1%  -2.01%  (p=0.000 n=48+50)
Flate             142ms ± 3%        141ms ± 3%  -0.85%  (p=0.003 n=50+48)
GoParser          178ms ± 4%        175ms ± 4%  -1.66%  (p=0.000 n=48+46)
Reflect           520ms ± 2%        512ms ± 2%  -1.44%  (p=0.000 n=45+48)
Tar               200ms ± 3%        198ms ± 4%  -0.61%  (p=0.037 n=47+50)
XML               277ms ± 3%        275ms ± 3%  -0.85%  (p=0.000 n=49+48)
[Geo mean]        482ms             476ms       -1.23%

name        old alloc/op      new alloc/op      delta
Template         36.1MB ± 0%       35.3MB ± 0%  -2.18%  (p=0.008 n=5+5)
Unicode          29.8MB ± 0%       29.3MB ± 0%  -1.58%  (p=0.008 n=5+5)
GoTypes           125MB ± 0%        123MB ± 0%  -2.13%  (p=0.008 n=5+5)
Compiler          531MB ± 0%        513MB ± 0%  -3.40%  (p=0.008 n=5+5)
SSA              2.00GB ± 0%       1.93GB ± 0%  -3.34%  (p=0.008 n=5+5)
Flate            24.5MB ± 0%       24.3MB ± 0%  -1.18%  (p=0.008 n=5+5)
GoParser         29.4MB ± 0%       28.7MB ± 0%  -2.34%  (p=0.008 n=5+5)
Reflect          87.1MB ± 0%       86.0MB ± 0%  -1.33%  (p=0.008 n=5+5)
Tar              35.3MB ± 0%       34.8MB ± 0%  -1.44%  (p=0.008 n=5+5)
XML              47.9MB ± 0%       47.1MB ± 0%  -1.86%  (p=0.008 n=5+5)
[Geo mean]       82.8MB            81.1MB       -2.08%

name        old allocs/op     new allocs/op     delta
Template           352k ± 0%         347k ± 0%  -1.32%  (p=0.008 n=5+5)
Unicode            342k ± 0%         339k ± 0%  -0.66%  (p=0.008 n=5+5)
GoTypes           1.29M ± 0%        1.27M ± 0%  -1.30%  (p=0.008 n=5+5)
Compiler          4.98M ± 0%        4.87M ± 0%  -2.14%  (p=0.008 n=5+5)
SSA               15.7M ± 0%        15.2M ± 0%  -2.86%  (p=0.008 n=5+5)
Flate              233k ± 0%         231k ± 0%  -0.83%  (p=0.008 n=5+5)
GoParser           296k ± 0%         291k ± 0%  -1.54%  (p=0.016 n=5+4)
Reflect           1.05M ± 0%        1.04M ± 0%  -0.65%  (p=0.008 n=5+5)
Tar                343k ± 0%         339k ± 0%  -0.97%  (p=0.008 n=5+5)
XML                432k ± 0%         426k ± 0%  -1.19%  (p=0.008 n=5+5)
[Geo mean]         815k              804k       -1.35%

name        old object-bytes  new object-bytes  delta
Template          505kB ± 0%        505kB ± 0%  -0.01%  (p=0.008 n=5+5)
Unicode           224kB ± 0%        224kB ± 0%    ~     (all equal)
GoTypes          1.82MB ± 0%       1.83MB ± 0%  +0.06%  (p=0.008 n=5+5)
Flate             324kB ± 0%        324kB ± 0%  +0.00%  (p=0.008 n=5+5)
GoParser          402kB ± 0%        402kB ± 0%  +0.04%  (p=0.008 n=5+5)
Reflect          1.39MB ± 0%       1.39MB ± 0%  -0.01%  (p=0.008 n=5+5)
Tar               449kB ± 0%        449kB ± 0%  -0.02%  (p=0.008 n=5+5)
XML               598kB ± 0%        597kB ± 0%  -0.05%  (p=0.008 n=5+5)

Change-Id: Ifc9d5c1bd01f90171414b8fb18ffe2290d271143
Reviewed-on: https://go-review.googlesource.com/c/114797
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agocmd/compile: in wasm, allocate approximately right number of locals for functions
Denys Smirnov [Fri, 19 Oct 2018 19:04:29 +0000 (19:04 +0000)]
cmd/compile: in wasm, allocate approximately right number of locals for functions

Currently, WASM binary writer requests 16 int registers (locals) and
16 float registers for every function regardless of how many locals the
function uses.

This change counts the number of used registers and requests a number
of locals matching the highest register index. The change has no effect
on performance and neglectable binary size improvement, but it makes
WASM code more readable and easy to analyze.

Change-Id: Ic1079623c0d632b215c68482db909fa440892700
GitHub-Last-Rev: 184634fa918aff74e280904dc2efafcc80735a8b
GitHub-Pull-Request: golang/go#28116
Reviewed-on: https://go-review.googlesource.com/c/140999
Reviewed-by: Richard Musiol <neelance@gmail.com>
Run-TryBot: Richard Musiol <neelance@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agogo/types: accept recv base type that is alias to a pointer type
Robert Griesemer [Thu, 18 Oct 2018 22:19:15 +0000 (15:19 -0700)]
go/types: accept recv base type that is alias to a pointer type

Per the spec clarification https://golang.org/cl/142757 (issue #27995).

Fixes #28251.
Updates #27995.

Change-Id: Idc142829955f9306a8698c5ed1c24baa8ee2b109
Reviewed-on: https://go-review.googlesource.com/c/143179
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agogo/types: collect type info for type ...T in variadic functions
Robert Griesemer [Fri, 19 Oct 2018 04:21:46 +0000 (21:21 -0700)]
go/types: collect type info for type ...T in variadic functions

Because the code type-checks T rather than ...T (and then corrects
the type to []T "manually"), it didn't automatically record the
type for the ast.Expr corresponding to ...T. Do it manually.

Fixes #28277.

Change-Id: I3d9aae310c90b01f52d189e70c48dd9007f72207
Reviewed-on: https://go-review.googlesource.com/c/143317
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/compile: remove compiling_wrappers
Matthew Dempsky [Thu, 18 Oct 2018 23:39:49 +0000 (16:39 -0700)]
cmd/compile: remove compiling_wrappers

It's no longer needed after removing safemode.

Change-Id: I7581d77a86342e3b6d7c632839f5eb7a5c20902e
Reviewed-on: https://go-review.googlesource.com/c/143397
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agogo/types: fix unsymmetric test when typechecking comparisons
Robert Griesemer [Fri, 19 Oct 2018 00:39:35 +0000 (17:39 -0700)]
go/types: fix unsymmetric test when typechecking comparisons

The existing code assumed that comparability and orderedness
was implied for the 2nd operand if the 1st operand satisfied
these predicates.

Fixes #28164.

Change-Id: I61d4e5eedb3297731a20a14acb3645d11b36fcc5
Reviewed-on: https://go-review.googlesource.com/c/143277
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agoruntime: port assembly files for aix/ppc64
Clément Chigot [Mon, 1 Oct 2018 07:50:36 +0000 (09:50 +0200)]
runtime: port assembly files for aix/ppc64

This commit adds the change on asm_ppc64.s and tls_ppc64.s files for AIX
operating system.

R2 does not need to be set for aix/ppc64 since it should remain valid
througout Go execution, except after a call to a C function.
Moreover, g must always be saved on the tls as syscalls are made with
C functions.

Some modifications on asm_ppc64.s are due to AIX stack layout.

It also removes a useless part in asmcgocall which was done twice.

Change-Id: Ie037ab73da00562bb978f2d0f17fcdabd4a40aa2
Reviewed-on: https://go-review.googlesource.com/c/138735
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/internal/xcoff: add new debug package for cmd
Clément Chigot [Fri, 28 Sep 2018 14:15:13 +0000 (16:15 +0200)]
cmd/internal/xcoff: add new debug package for cmd

This commit adds a new package in cmd/internal which aims
to debug and load XCOFF files.

Updates: #25893, #28037

Change-Id: I47db495bedfa43e9129a831b9b8bbc35b703567b
Reviewed-on: https://go-review.googlesource.com/c/138727
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocrypto/tls: replace net.Pipe in tests with real TCP connections
Filippo Valsorda [Wed, 17 Oct 2018 03:47:55 +0000 (23:47 -0400)]
crypto/tls: replace net.Pipe in tests with real TCP connections

crypto/tls is meant to work over network connections with buffering, not
synchronous connections, as explained in #24198. Tests based on net.Pipe
are unrealistic as reads and writes are matched one to one. Such tests
worked just thanks to the implementation details of the tls.Conn
internal buffering, and would break if for example the flush of the
first flight of the server was not entirely assimilated by the client
rawInput buffer before the client attempted to reply to the ServerHello.

Note that this might run into the Darwin network issues at #25696.

Fixed a few test races that were either hidden or synchronized by the
use of the in-memory net.Pipe.

Also, this gets us slightly more realistic benchmarks, reflecting some
syscall cost of Read and Write operations.

Change-Id: I5a597b3d7a81b8ccc776030cc837133412bf50f8
Reviewed-on: https://go-review.googlesource.com/c/142817
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotext/template: drop unused sortKeys function
Seebs [Thu, 18 Oct 2018 22:00:02 +0000 (17:00 -0500)]
text/template: drop unused sortKeys function

Recent change golang.org/cl/142737 drops the only call site for the
sortKeys function. If it's not in use, it should probably not be there in
the code, lurking and preparing to bite us when someone calls that instead
of the new key sorter in fmtsort, resulting in strange inconsistencies.

Since the function isn't called, this should have no impact.
Related to, but does not fix, #21095.

Change-Id: I4695503ef4d5ce90d989ec952f01ea00cc15c79d
Reviewed-on: https://go-review.googlesource.com/c/143178
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoflag: return a consistent parse error if the flag value is invalid
Rob Pike [Thu, 18 Oct 2018 23:48:53 +0000 (10:48 +1100)]
flag: return a consistent parse error if the flag value is invalid

Return a consistently formatted error string that reports either
a parse error or a range error.

Before:
invalid boolean value "3" for -debug: strconv.ParseBool: parsing "3": invalid syntax

After:
invalid boolean value "3" for -debug: parse error

Fixes #26822

Change-Id: I60992bf23da32a4c0cf32472a8af486a3c9674ad
Reviewed-on: https://go-review.googlesource.com/c/143257
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: optimize store combination on 386/amd64
Ben Shi [Thu, 18 Oct 2018 05:11:38 +0000 (05:11 +0000)]
cmd/compile: optimize store combination on 386/amd64

This CL add 3 rules to combine byte-store to word-store on386 and
amd64.

Change-Id: Iffd9cda42f1961680c81def4edc773ad58f211b3
Reviewed-on: https://go-review.googlesource.com/c/143057
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/compile: extract gc.eqtype as types.Identical
Matthew Dempsky [Thu, 18 Oct 2018 22:24:50 +0000 (15:24 -0700)]
cmd/compile: extract gc.eqtype as types.Identical

For symmetry with go/types.Identical.

Passes toolstash-check.

Change-Id: Id19c3956e44ed8e2d9f203d15824322cc5842d3d
Reviewed-on: https://go-review.googlesource.com/c/143180
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agofmt: print maps in key-sorted order
Rob Pike [Tue, 16 Oct 2018 02:54:39 +0000 (13:54 +1100)]
fmt: print maps in key-sorted order

For easier testing, change the way maps are printed so they
appear in a consistent order between runs. Do this by printing
them in key-sorted order.

To do this, we add a package at the root, internal/fmtsort,
that implements a general mechanism for sorting map keys
regardless of their type. This is a little messy and probably
slow, but formatted printing of maps has never been fast and
is already always reflection-driven.

The rules are:

The ordering rules are more general than with Go's < operator:

 - when applicable, nil compares low
 - ints, floats, and strings order by <
 - NaN compares less than non-NaN floats
 - bool compares false before true
 - complex compares real, then imag
 - pointers compare by machine address
 - channel values compare by machine address
 - structs compare each field in turn
 - arrays compare each element in turn.
 - interface values compare first by reflect.Type describing the concrete type
   and then by concrete value as described in the previous rules.

The new package is internal because we really do not want
everyone using this to sort things. It is slow, not general, and
only suitable for the subset of types that can be map keys.

Also use the package in text/template, which already had a
weaker version of this mechanism.

This change requires adding a dependency on sort to the fmt
package, but that isn't disruptive to the dependency tree.

Fixes #21095

Change-Id: Ia602115c7de5d95993dbd609611d8bd96e054157
Reviewed-on: https://go-review.googlesource.com/c/142737
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agocmd/compile: fix gdb stepping test, take 2
Keith Randall [Mon, 15 Oct 2018 20:40:00 +0000 (13:40 -0700)]
cmd/compile: fix gdb stepping test, take 2

The fix in CL 141649 is not right, the line in question got moved,
not added. Not sure why the -u option didn't do the right thing
when preparing the diff.

Fixes #28198

Change-Id: I6d45fdbbd5a9487cc70da07ab84e090b689a57f5
Reviewed-on: https://go-review.googlesource.com/c/142298
Reviewed-by: Alberto Donizetti <alb.donizetti@gmail.com>
6 years agocmd/doc: minor code simplification
Robert Griesemer [Thu, 18 Oct 2018 19:37:25 +0000 (12:37 -0700)]
cmd/doc: minor code simplification

Follow-up on https://golang.org/cl/143037.

Change-Id: Ia36760b499f7d46ce03e0fa81ec455a4a20208ac
Reviewed-on: https://go-review.googlesource.com/c/143061
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
6 years agocmd/compile: attach slots to incoming params for better debugging
David Chase [Thu, 11 Oct 2018 21:48:33 +0000 (17:48 -0400)]
cmd/compile: attach slots to incoming params for better debugging

This change attaches a slots to the OpArg values for
incoming params, and this in turn causes location lists
to be generated for params, and that yields better
debugging, in delve and sometimes in gdb.

The parameter lifetimes could start earlier; they are in
fact defined on entry, not at the point where the OpArg is
finally mentioned.  (that will be addressed in another CL)

Change-Id: Icca891e118291d260c35a14acd5bc92bb82d9e9f
Reviewed-on: https://go-review.googlesource.com/c/141697
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agotest: update issue5089.go for recent gccgo changes
Ian Lance Taylor [Tue, 24 Jan 2017 21:22:21 +0000 (13:22 -0800)]
test: update issue5089.go for recent gccgo changes

As of https://golang.org/cl/43456 gccgo now gives a better error
message for this test.

Before:
    fixedbugs/issue5089.go:13:1: error: redefinition of ‘bufio.Buffered’: receiver name changed
     func (b *bufio.Reader) Buffered() int { // ERROR "non-local|redefinition"
     ^
    fixedbugs/issue5089.go:11:13: note: previous definition of ‘bufio.Buffered’ was here
     import "bufio" // GCCGO_ERROR "previous"
                 ^

Now:
    fixedbugs/issue5089.go:13:7: error: may not define methods on non-local type
     func (b *bufio.Reader) Buffered() int { // ERROR "non-local|redefinition"
           ^

Change-Id: I4112ca8d91336f6369f780c1d45b8915b5e8e235
Reviewed-on: https://go-review.googlesource.com/c/130955
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agosyscall/js: make zero js.Value represent "undefined"
Richard Musiol [Thu, 18 Oct 2018 13:53:38 +0000 (15:53 +0200)]
syscall/js: make zero js.Value represent "undefined"

This commit changes the encoding of js.Value so that the zero js.Value
represents the JavaScript value "undefined". This is what users
intuitively expect.

Specifically, the encodings of "undefined" and the number zero have
been swapped.

Fixes #27592.

Change-Id: Icfc832c8cdf7a8a78bd69d20e00a04dbed0ccd10
Reviewed-on: https://go-review.googlesource.com/c/143137
Run-TryBot: Richard Musiol <neelance@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotest: add fixedbugs/bug506 for gccgo
Ian Lance Taylor [Mon, 11 Dec 2017 20:03:44 +0000 (12:03 -0800)]
test: add fixedbugs/bug506 for gccgo

Building with gccgo failed with an undefined symbol error from an
unnecessary hash function.

Updates #19773

Change-Id: Ic78bf1b086ff5ee26d464089c0e14987d3fe8b02
Reviewed-on: https://go-review.googlesource.com/c/130956
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/doc: make comments inside functions appear with -src
Rob Pike [Thu, 18 Oct 2018 02:46:54 +0000 (13:46 +1100)]
cmd/doc: make comments inside functions appear with -src

The old godoc didn't do this either, perhaps because it's a little
tricky, but it can be done using a special type from the go/printer
package. (Usually we just use go/format).

Fixes #28195.

Change-Id: Ic6d3df3953ba71128398ceaf9a133c798551b6b8
Reviewed-on: https://go-review.googlesource.com/c/143037
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agotest/codegen: add more combined load/store test cases
Ben Shi [Wed, 17 Oct 2018 10:24:28 +0000 (10:24 +0000)]
test/codegen: add more combined load/store test cases

This CL adds more combined load/store test cases for 386/amd64.

Change-Id: I0a483a6ed0212b65c5e84d67ed8c9f50c389ce2d
Reviewed-on: https://go-review.googlesource.com/c/142878
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agoos: ignore SIGTTIN during TestTTYClose
Ian Lance Taylor [Thu, 20 Sep 2018 03:09:02 +0000 (20:09 -0700)]
os: ignore SIGTTIN during TestTTYClose

Otherwise, if the test is run in the background, it will stop waiting
for access to the terminal.

Change-Id: Ib5224c6cb9060281e05c3b00cd2964445421e774
Reviewed-on: https://go-review.googlesource.com/c/136415
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agodatabase/sql: remove commented debug prints
Umang Parmar [Wed, 17 Oct 2018 08:10:28 +0000 (08:10 +0000)]
database/sql: remove commented debug prints

Fixes #28234

Change-Id: I89090ffb8285c4936b0c9b5c2475849c0643186a
GitHub-Last-Rev: 4dd0ec162d0ce1548045d4119fd3295570f65d85
GitHub-Pull-Request: golang/go#28246
Reviewed-on: https://go-review.googlesource.com/c/142877
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/go: make go vet query cmd/vet for its flags
Alan Donovan [Tue, 16 Oct 2018 18:35:01 +0000 (14:35 -0400)]
cmd/go: make go vet query cmd/vet for its flags

Add -flags flag to cmd/vet that causes it to describe its flags as JSON.

go vet's "-vettool" flag has been replaced with an environment
variable, GOVETTOOL, for two reasons:

  1) we need its value before flag processing,
     because we must run vet to discover its flags.

  2) users may change the env var to opt in/out of the new vet tool
     during the upcoming transition to vet based on the analysis API.

Change-Id: I5d8f90817623022f4170b88fab3c92c9b2fbdc37
Reviewed-on: https://go-review.googlesource.com/c/142617
Run-TryBot: Alan Donovan <adonovan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agogo/internal/gccgoimporter: backport from x/tools to ensure identical code
Robert Griesemer [Wed, 17 Oct 2018 21:15:54 +0000 (14:15 -0700)]
go/internal/gccgoimporter: backport from x/tools to ensure identical code

This change backports a minor modification of the x/tools version of this
code back into the std library. It simply ensures that both versions of
the code are the same and will simplify keeping them in sync down the
road.

While this is an API change, this is an internal package, so we're ok.

Updates #27891.

Change-Id: Ib153141382f727a2692ca80179ae09c4a383ba4f
Reviewed-on: https://go-review.googlesource.com/c/142894
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agospec: clarify rules for receiver base types
Robert Griesemer [Tue, 16 Oct 2018 23:50:25 +0000 (16:50 -0700)]
spec: clarify rules for receiver base types

The spec currently provides a syntactic rule for receiver base types,
and a strict reading of those rules prohibits the use of type aliases
referring to pointer types as receiver types.

This strict interpretation breaks an assumed rule for aliases, which
is that a type literal can always be replaced by an alias denoting
that literal.

Furthermore, cmd/compile always accepted this new formulation of the
receiver type rules and so this change will simply validate what has
been implemented all along.

Fixes #27995.

Change-Id: I032289c926a4f070d6f7795431d86635fe64d907
Reviewed-on: https://go-review.googlesource.com/c/142757
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agoos/signal: wait for goroutine in TestTerminalSignal
Ian Lance Taylor [Wed, 17 Oct 2018 19:46:20 +0000 (12:46 -0700)]
os/signal: wait for goroutine in TestTerminalSignal

Fixes #28169

Change-Id: I187d9effea56357bbb04d4971d284a52ffae61f8
Reviewed-on: https://go-review.googlesource.com/c/142889
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/cgo: write a string rather than building an AST
Ian Lance Taylor [Wed, 17 Oct 2018 04:36:34 +0000 (21:36 -0700)]
cmd/cgo: write a string rather than building an AST

This generates the same code as before, but does so directly rather
than building an AST and printing that. This is in preparation for
later changes.

Change-Id: Ifec141120bcc74847f0bff8d3d47306bfe69b454
Reviewed-on: https://go-review.googlesource.com/c/142883
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/cgo: split name rewriting out of rewriteRef
Ian Lance Taylor [Wed, 17 Oct 2018 01:32:35 +0000 (18:32 -0700)]
cmd/cgo: split name rewriting out of rewriteRef

This is in preparation for later changes.

Change-Id: I2b9b77a782cf65a2fcec5e700ec6bb8b1476f6b5
Reviewed-on: https://go-review.googlesource.com/c/142882
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/cgo: mangle names before rewriting calls
Ian Lance Taylor [Wed, 17 Oct 2018 01:18:40 +0000 (18:18 -0700)]
cmd/cgo: mangle names before rewriting calls

Move name mangling before rewriting calls rather than after.
This is in preparation for later changes.

Change-Id: I74bc351f4290dad7ebf6d0d361bb684087786053
Reviewed-on: https://go-review.googlesource.com/c/142881
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoos: change UserHomeDir to use USERPROFILE on windows
Jordan Rhee [Wed, 17 Oct 2018 16:41:09 +0000 (09:41 -0700)]
os: change UserHomeDir to use USERPROFILE on windows

Fixes #28182

Change-Id: I49c2117fba6325c234512f937ff2edfa9477f52f
Reviewed-on: https://go-review.googlesource.com/c/142886
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: remove obsolete "safe" mode
Matthew Dempsky [Tue, 16 Oct 2018 21:02:13 +0000 (14:02 -0700)]
cmd/compile: remove obsolete "safe" mode

Nowadays there are better ways to safely run untrusted Go programs, like
NaCl and gVisor.

Change-Id: I20c45f13a50dbcf35c343438b720eb93e7b4e13a
Reviewed-on: https://go-review.googlesource.com/c/142717
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agotest: limit runoutput concurrency with -v
Josh Bleecher Snyder [Wed, 17 Oct 2018 04:37:44 +0000 (21:37 -0700)]
test: limit runoutput concurrency with -v

This appears to have simply been an oversight.

Change-Id: Ia5d1309b3ebc99c9abbf0282397693272d8178aa
Reviewed-on: https://go-review.googlesource.com/c/142885
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: use only treaps for tracking spans
Michael Anthony Knyszek [Thu, 4 Oct 2018 21:03:26 +0000 (21:03 +0000)]
runtime: use only treaps for tracking spans

Currently, mheap tracks spans in both mSpanLists and mTreaps, but
mSpanLists, while they tend to be smaller, complicate the
implementation. Here we simplify the implementation by removing
free and busy from mheap and renaming freelarge -> free and busylarge
-> busy.

This change also slightly changes the reclamation policy. Previously,
for allocations under 1MB we would attempt to find a small span of the
right size. Now, we just try to find any number of spans totaling the
right size. This may increase heap fragmentation, but that will be dealt
with using virtual memory tricks in follow-up CLs.

For #14045.

Garbage-heavy benchmarks show very little change, except what appears
to be a decrease in STW times and peak RSS.

name                      old STW-ns/GC       new STW-ns/GC       delta
Garbage/benchmem-MB=64-8           263k ±64%           217k ±24%  -17.66%  (p=0.028 n=25+23)

name                      old STW-ns/op       new STW-ns/op       delta
Garbage/benchmem-MB=64-8          9.39k ±65%          7.80k ±24%  -16.88%  (p=0.037 n=25+23)

name                      old peak-RSS-bytes  new peak-RSS-bytes  delta
Garbage/benchmem-MB=64-8           281M ± 0%           249M ± 4%  -11.40%  (p=0.000 n=19+18)

https://perf.golang.org/search?q=upload:20181005.1

Go1 benchmarks perform roughly the same, the most notable regression
being the JSON encode/decode benchmark with worsens by ~2%.

name                     old time/op    new time/op    delta
BinaryTree17-8              3.02s ± 2%     2.99s ± 2%  -1.18%  (p=0.000 n=25+24)
Fannkuch11-8                3.05s ± 1%     3.02s ± 2%  -1.20%  (p=0.000 n=25+25)
FmtFprintfEmpty-8          43.6ns ± 5%    43.4ns ± 3%    ~     (p=0.528 n=25+25)
FmtFprintfString-8         74.9ns ± 3%    73.4ns ± 1%  -2.03%  (p=0.001 n=25+24)
FmtFprintfInt-8            79.3ns ± 3%    77.9ns ± 1%  -1.73%  (p=0.003 n=25+25)
FmtFprintfIntInt-8          119ns ± 6%     116ns ± 0%  -2.68%  (p=0.000 n=25+18)
FmtFprintfPrefixedInt-8     134ns ± 4%     132ns ± 1%  -1.52%  (p=0.004 n=25+25)
FmtFprintfFloat-8           240ns ± 1%     241ns ± 1%    ~     (p=0.403 n=24+23)
FmtManyArgs-8               543ns ± 1%     537ns ± 1%  -1.00%  (p=0.000 n=25+25)
GobDecode-8                6.88ms ± 1%    6.92ms ± 4%    ~     (p=0.088 n=24+22)
GobEncode-8                5.92ms ± 1%    5.93ms ± 1%    ~     (p=0.898 n=25+24)
Gzip-8                      267ms ± 2%     266ms ± 2%    ~     (p=0.213 n=25+24)
Gunzip-8                   35.4ms ± 1%    35.6ms ± 1%  +0.70%  (p=0.000 n=25+25)
HTTPClientServer-8          104µs ± 2%     104µs ± 2%    ~     (p=0.686 n=25+25)
JSONEncode-8               9.67ms ± 1%    9.80ms ± 4%  +1.32%  (p=0.000 n=25+25)
JSONDecode-8               47.7ms ± 1%    48.8ms ± 5%  +2.33%  (p=0.000 n=25+25)
Mandelbrot200-8            4.87ms ± 1%    4.91ms ± 1%  +0.79%  (p=0.000 n=25+25)
GoParse-8                  3.59ms ± 4%    3.55ms ± 1%    ~     (p=0.199 n=25+24)
RegexpMatchEasy0_32-8      90.3ns ± 1%    89.9ns ± 1%  -0.47%  (p=0.000 n=25+21)
RegexpMatchEasy0_1K-8       204ns ± 1%     204ns ± 1%    ~     (p=0.914 n=25+24)
RegexpMatchEasy1_32-8      84.9ns ± 0%    84.6ns ± 1%  -0.36%  (p=0.000 n=24+25)
RegexpMatchEasy1_1K-8       350ns ± 1%     348ns ± 3%  -0.59%  (p=0.007 n=25+25)
RegexpMatchMedium_32-8      122ns ± 1%     121ns ± 0%  -1.08%  (p=0.000 n=25+18)
RegexpMatchMedium_1K-8     36.1µs ± 1%    34.6µs ± 1%  -4.02%  (p=0.000 n=25+25)
RegexpMatchHard_32-8       1.69µs ± 2%    1.65µs ± 1%  -2.38%  (p=0.000 n=25+25)
RegexpMatchHard_1K-8       50.8µs ± 1%    49.4µs ± 1%  -2.69%  (p=0.000 n=25+24)
Revcomp-8                   453ms ± 2%     449ms ± 3%  -0.74%  (p=0.022 n=25+24)
Template-8                 63.2ms ± 2%    63.4ms ± 1%    ~     (p=0.127 n=25+24)
TimeParse-8                 313ns ± 1%     315ns ± 3%    ~     (p=0.924 n=24+25)
TimeFormat-8                294ns ± 1%     292ns ± 2%  -0.65%  (p=0.004 n=23+24)
[Geo mean]                 49.9µs         49.6µs       -0.65%

name                     old speed      new speed      delta
GobDecode-8               112MB/s ± 1%   110MB/s ± 4%  -1.00%  (p=0.036 n=24+24)
GobEncode-8               130MB/s ± 1%   129MB/s ± 1%    ~     (p=0.894 n=25+24)
Gzip-8                   72.7MB/s ± 2%  73.0MB/s ± 2%    ~     (p=0.208 n=25+24)
Gunzip-8                  549MB/s ± 1%   545MB/s ± 1%  -0.70%  (p=0.000 n=25+25)
JSONEncode-8              201MB/s ± 1%   198MB/s ± 3%  -1.29%  (p=0.000 n=25+25)
JSONDecode-8             40.7MB/s ± 1%  39.8MB/s ± 5%  -2.23%  (p=0.000 n=25+25)
GoParse-8                16.2MB/s ± 4%  16.3MB/s ± 1%    ~     (p=0.211 n=25+24)
RegexpMatchEasy0_32-8     354MB/s ± 1%   356MB/s ± 1%  +0.47%  (p=0.000 n=25+21)
RegexpMatchEasy0_1K-8    5.00GB/s ± 0%  4.99GB/s ± 1%    ~     (p=0.588 n=24+24)
RegexpMatchEasy1_32-8     377MB/s ± 1%   378MB/s ± 1%  +0.39%  (p=0.000 n=25+25)
RegexpMatchEasy1_1K-8    2.92GB/s ± 1%  2.94GB/s ± 3%  +0.65%  (p=0.008 n=25+25)
RegexpMatchMedium_32-8   8.14MB/s ± 1%  8.22MB/s ± 1%  +0.98%  (p=0.000 n=25+24)
RegexpMatchMedium_1K-8   28.4MB/s ± 1%  29.6MB/s ± 1%  +4.19%  (p=0.000 n=25+25)
RegexpMatchHard_32-8     18.9MB/s ± 2%  19.4MB/s ± 1%  +2.43%  (p=0.000 n=25+25)
RegexpMatchHard_1K-8     20.2MB/s ± 1%  20.7MB/s ± 1%  +2.76%  (p=0.000 n=25+24)
Revcomp-8                 561MB/s ± 2%   566MB/s ± 3%  +0.75%  (p=0.021 n=25+24)
Template-8               30.7MB/s ± 2%  30.6MB/s ± 1%    ~     (p=0.131 n=25+24)
[Geo mean]                120MB/s        121MB/s       +0.48%

https://perf.golang.org/search?q=upload:20181004.6

Change-Id: I97f9fee34577961a116a8ddd445c6272253f0f95
Reviewed-on: https://go-review.googlesource.com/c/139837
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agoruntime: de-duplicate span scavenging
Michael Anthony Knyszek [Tue, 2 Oct 2018 20:23:41 +0000 (20:23 +0000)]
runtime: de-duplicate span scavenging

Currently, span scavenging was done nearly identically in two different
locations. This change deduplicates that into one shared routine.

For #14045.

Change-Id: I15006b2c9af0e70b7a9eae9abb4168d3adca3860
Reviewed-on: https://go-review.googlesource.com/c/139297
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agocmd/compile: make tabs narrow in src column of ssa.html
Yury Smolsky [Wed, 17 Oct 2018 07:40:44 +0000 (10:40 +0300)]
cmd/compile: make tabs narrow in src column of ssa.html

Too deeply nested code is hard to fit in ssa.html.
This CL reduces the tab size to 4 characters.

Change-Id: I08643b0868bce3439567084c7d701654655f23d7
Reviewed-on: https://go-review.googlesource.com/c/142857
Reviewed-by: David Chase <drchase@google.com>
6 years agocmd/doc: fix repeated header bug added in previous CL
Rob Pike [Wed, 17 Oct 2018 11:32:27 +0000 (22:32 +1100)]
cmd/doc: fix repeated header bug added in previous CL

One too many lines was deleted, and it would print a header multiple times.
Add a test.

Change-Id: I4906b454dbb66193d515ffacf43849ffdc2dede6
Reviewed-on: https://go-review.googlesource.com/c/142937
Reviewed-by: Ralph Corderoy <ralph@inputplus.co.uk>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/doc: add -all flag to print all documentation for package
Rob Pike [Sat, 13 Oct 2018 10:06:43 +0000 (21:06 +1100)]
cmd/doc: add -all flag to print all documentation for package

Unlike the one for the old godoc, you need the -u flag to see
unexported symbols. This seems like the right behavior: it's
consistent.

For now at least, the argument must be a package, not a symbol.
This is also different from old godoc.

Required a little refactoring but also cleaned up a few things.

Update #25595

Leaving the bug open for now until we tackle
go doc -all symbol

Change-Id: Ibc1975bfa592cb1e92513eb2e5e9e11e01a60095
Reviewed-on: https://go-review.googlesource.com/c/141977
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agocrypto/tls,crypto/x509: normalize RFC references
Filippo Valsorda [Fri, 12 Oct 2018 21:07:04 +0000 (17:07 -0400)]
crypto/tls,crypto/x509: normalize RFC references

Use the format "RFC XXXX, Section X.X" (or "Appendix Y.X") as it fits
more properly in prose than a link, is more future-proof, and as there
are multiple ways to render an RFC. Capital "S" to follow the quoting
standard of RFCs themselves.

Applied the new goimports grouping to all files in those packages, too.

Change-Id: I01267bb3a3b02664f8f822e97b129075bb14d404
Reviewed-on: https://go-review.googlesource.com/c/141918
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
6 years agocrypto/aes: remove redundant nil check around loop
Rijnard van Tonder [Wed, 17 Oct 2018 01:10:14 +0000 (01:10 +0000)]
crypto/aes: remove redundant nil check around loop

Change-Id: I8cc4b5efe798e74b6daabd64fc2dd5486dcb7c5e
GitHub-Last-Rev: 694509e33df7c5729ec0bf7b053ff18ebd87c36b
GitHub-Pull-Request: golang/go#28115
Reviewed-on: https://go-review.googlesource.com/c/140998
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: simplified test case (cleanup)
Robert Griesemer [Tue, 16 Oct 2018 22:53:38 +0000 (15:53 -0700)]
cmd/compile: simplified test case (cleanup)

Follow-up on https://golang.org/cl/124595; no semantic changes.

Updates #26411.

Change-Id: Ic1c4622dbf79529ff61530f9c25ec742c2abe5ca
Reviewed-on: https://go-review.googlesource.com/c/142720
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: fix label redefinition error column numbers
Emmanuel T Odeke [Wed, 18 Jul 2018 02:02:53 +0000 (19:02 -0700)]
cmd/compile: fix label redefinition error column numbers

Ensure that label redefinition error column numbers
print the actual start of the label instead of the
position of the label's delimiting token ":".

For example, given this program:

package main

func main() {

            foo:
   foo:
foo:
foo            :
}

* Before:
main.go:5:13: label foo defined and not used
main.go:6:7: label foo already defined at main.go:5:13
main.go:7:4: label foo already defined at main.go:5:13
main.go:8:16: label foo already defined at main.go:5:13

* After:
main.go:5:13: label foo defined and not used
main.go:6:4: label foo already defined at main.go:5:13
main.go:7:1: label foo already defined at main.go:5:13
main.go:8:1: label foo already defined at main.go:5:13

Fixes #26411

Change-Id: I8eb874b97fdc8862547176d57ac2fa0f075f2367
Reviewed-on: https://go-review.googlesource.com/c/124595
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agocmd/compile/internal/gc: inline autogenerated (*T).M wrappers
Ilya Tocar [Mon, 17 Sep 2018 19:08:03 +0000 (14:08 -0500)]
cmd/compile/internal/gc: inline autogenerated (*T).M wrappers

Currently all inlining of autogenerated wrappers is disabled,
because it causes build failures, when indexed export format is enabled.
Turns out we can reenable it for common case of (*T).M wrappers.
This fixes most performance degradation of 1.11 vs 1.10.

encoding/binary:
name                    old time/op   new time/op   delta
ReadSlice1000Int32s-6    14.8µs ± 2%   11.5µs ± 2%  -22.01%  (p=0.000 n=10+10)
WriteSlice1000Int32s-6   14.8µs ± 2%   11.7µs ± 2%  -20.95%  (p=0.000 n=10+10)

bufio:
name           old time/op    new time/op    delta
WriterFlush-6    32.4ns ± 1%    28.8ns ± 0%  -11.17%  (p=0.000 n=9+10)

sort:
SearchWrappers-6       231ns ± 1%   231ns ± 0%     ~     (p=0.129 n=9+10)
SortString1K-6         365µs ± 1%   298µs ± 1%  -18.43%  (p=0.000 n=9+10)
SortString1K_Slice-6   274µs ± 2%   276µs ± 1%     ~     (p=0.105 n=10+10)
StableString1K-6       490µs ± 1%   373µs ± 1%  -23.73%  (p=0.000 n=10+10)
SortInt1K-6            210µs ± 1%   142µs ± 1%  -32.69%  (p=0.000 n=10+10)
StableInt1K-6          243µs ± 0%   151µs ± 1%  -37.75%  (p=0.000 n=10+10)
StableInt1K_Slice-6    130µs ± 1%   130µs ± 0%     ~     (p=0.237 n=10+8)
SortInt64K-6          19.9ms ± 1%  13.5ms ± 1%  -32.32%  (p=0.000 n=10+10)
SortInt64K_Slice-6    11.5ms ± 1%  11.5ms ± 1%     ~     (p=0.912 n=10+10)
StableInt64K-6        21.5ms ± 0%  13.5ms ± 1%  -37.30%  (p=0.000 n=9+10)
Sort1e2-6              108µs ± 2%    83µs ± 3%  -23.26%  (p=0.000 n=10+10)
Stable1e2-6            218µs ± 0%   161µs ± 1%  -25.99%  (p=0.000 n=8+9)
Sort1e4-6             22.6ms ± 1%  16.8ms ± 0%  -25.45%  (p=0.000 n=10+7)
Stable1e4-6           67.6ms ± 1%  49.7ms ± 0%  -26.48%  (p=0.000 n=10+10)
Sort1e6-6              3.44s ± 0%   2.55s ± 1%  -26.05%  (p=0.000 n=8+9)
Stable1e6-6            13.7s ± 0%    9.9s ± 1%  -27.68%  (p=0.000 n=8+10)

Fixes #27621
Updates #25338

Change-Id: I6fe633202f63fa829a6ab849c44d7e45f8835dff
Reviewed-on: https://go-review.googlesource.com/c/135697
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agoRevert "fmt: fix incorrect format of whole-number floats when using %#v"
Filippo Valsorda [Tue, 16 Oct 2018 17:01:07 +0000 (13:01 -0400)]
Revert "fmt: fix incorrect format of whole-number floats when using %#v"

Numbers without decimals are valid Go representations of whole-number
floats. That is, "var x float64 = 5" is valid Go. Avoid breakage in
tests that expect a certain output from %#v by reverting to it.

To guarantee the right type is generated by a print use %T(%#v) instead.

Added a test to lock in this behavior.

This reverts commit 7c7cecc1846aaaa0ce73931644fe1df2b4559e09.

Fixes #27634
Updates #26363

Change-Id: I544c400a0903777dd216452a7e86dfe60b0b0283
Reviewed-on: https://go-review.googlesource.com/c/142597
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/compile: remove -dolinkobj flag
Matthew Dempsky [Tue, 16 Oct 2018 19:49:17 +0000 (12:49 -0700)]
cmd/compile: remove -dolinkobj flag

This used to be used by cmd/vet and some assembly generation tests, but
those were removed in CL 37691 and CL 107336. No point in keeping an
unneeded flag around.

Fixes #28220.

Change-Id: I59f8546954ab36ea61ceba81c10d6e16d74b966a
Reviewed-on: https://go-review.googlesource.com/c/142677
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile/internal/gc: simplify typechecking definitions
Matthew Dempsky [Tue, 16 Oct 2018 18:58:00 +0000 (11:58 -0700)]
cmd/compile/internal/gc: simplify typechecking definitions

There are only a handful of nodes that we need to pass to
typecheckdef (OLITERAL, ONAME, OTYPE, and ONONAME), but typecheck1
takes the awkward approach of calling typecheckdef on every node with
Sym != nil, and then excluding a long list of uninteresting Ops that
have a non-nil Sym.

Passes toolstash-check.

Change-Id: I0271d2faff0208ad57ddc1f1a540a5fbed870234
Reviewed-on: https://go-review.googlesource.com/c/142657
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotest/codegen: enable more tests for ppc64/ppc64le
Lynn Boger [Mon, 15 Oct 2018 16:53:07 +0000 (12:53 -0400)]
test/codegen: enable more tests for ppc64/ppc64le

Adding cases for ppc64,ppc64le to the codegen tests
where appropriate.

Change-Id: Idf8cbe88a4ab4406a4ef1ea777bd15a58b68f3ed
Reviewed-on: https://go-review.googlesource.com/c/142557
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoencoding/json: always verify we can get a field's value
Daniel Martí [Thu, 11 Oct 2018 13:43:47 +0000 (14:43 +0100)]
encoding/json: always verify we can get a field's value

Calling .Interface on a struct field's reflect.Value isn't always safe.
For example, if that field is an unexported anonymous struct.

We only descended into this branch if the struct type had any methods,
so this bug had gone unnoticed for a few release cycles.

Add the check, and add a simple test case.

Fixes #28145.

Change-Id: I02f7e0ab9a4a0c18a5e2164211922fe9c3d30f64
Reviewed-on: https://go-review.googlesource.com/c/141537
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoencoding/json: fix "data changed underfoot?" panic
Daniel Martí [Tue, 16 Oct 2018 11:04:55 +0000 (12:04 +0100)]
encoding/json: fix "data changed underfoot?" panic

Given a program as follows:

data := []byte(`{"F": {
"a": 2,
"3": 4
}}`)
json.Unmarshal(data, &map[string]map[int]int{})

The JSON package should error, as "a" is not a valid integer. However,
we'd encounter a panic:

panic: JSON decoder out of sync - data changing underfoot?

The reason was that decodeState.object would return a nil error on
encountering the invalid map key string, while saving the key type error
for later. This broke if we were inside another object, as we would
abruptly end parsing the nested object, leaving the decoder in an
unexpected state.

To fix this, simply avoid storing the map element and continue decoding
the object, to leave the decoder state exactly as if we hadn't seen an
invalid key type.

This affected both signed and unsigned integer keys, so fix both and add
two test cases.

Updates #28189.

Change-Id: I8a6204cc3ff9fb04ed769df7a20a824c8b94faff
Reviewed-on: https://go-review.googlesource.com/c/142518
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: optimize 386's load/store combination
Ben Shi [Mon, 15 Oct 2018 01:19:48 +0000 (01:19 +0000)]
cmd/compile: optimize 386's load/store combination

This CL adds more combinations of two consequtive MOVBload/MOVBstore
to a unique MOVWload/MOVWstore.

1. The size of the go executable decreases about 4KB, and the total
size of pkg/linux_386 (excluding cmd/compile) decreases about 1.5KB.

2. There is no regression in the go1 benchmark result, excluding noise.
name                     old time/op    new time/op    delta
BinaryTree17-4              3.28s ± 2%     3.29s ± 2%    ~     (p=0.151 n=40+40)
Fannkuch11-4                3.52s ± 1%     3.51s ± 1%  -0.28%  (p=0.002 n=40+40)
FmtFprintfEmpty-4          45.4ns ± 4%    45.0ns ± 4%  -0.89%  (p=0.019 n=40+40)
FmtFprintfString-4         81.9ns ± 7%    81.3ns ± 1%    ~     (p=0.660 n=40+25)
FmtFprintfInt-4            91.9ns ± 9%    91.4ns ± 9%    ~     (p=0.249 n=40+40)
FmtFprintfIntInt-4          143ns ± 4%     143ns ± 4%    ~     (p=0.760 n=40+40)
FmtFprintfPrefixedInt-4     184ns ± 3%     183ns ± 4%    ~     (p=0.485 n=40+40)
FmtFprintfFloat-4           408ns ± 3%     409ns ± 3%    ~     (p=0.961 n=40+40)
FmtManyArgs-4               597ns ± 4%     602ns ± 3%    ~     (p=0.413 n=40+40)
GobDecode-4                7.13ms ± 6%    7.14ms ± 6%    ~     (p=0.859 n=40+40)
GobEncode-4                6.86ms ± 9%    6.94ms ± 7%    ~     (p=0.162 n=40+40)
Gzip-4                      395ms ± 4%     396ms ± 3%    ~     (p=0.099 n=40+40)
Gunzip-4                   40.9ms ± 4%    41.1ms ± 3%    ~     (p=0.064 n=40+40)
HTTPClientServer-4         63.6µs ± 2%    63.6µs ± 3%    ~     (p=0.832 n=36+39)
JSONEncode-4               16.1ms ± 3%    15.8ms ± 3%  -1.60%  (p=0.001 n=40+40)
JSONDecode-4               61.0ms ± 3%    61.5ms ± 4%    ~     (p=0.065 n=40+40)
Mandelbrot200-4            5.16ms ± 3%    5.18ms ± 3%    ~     (p=0.056 n=40+40)
GoParse-4                  3.25ms ± 2%    3.23ms ± 3%    ~     (p=0.727 n=40+40)
RegexpMatchEasy0_32-4      90.2ns ± 3%    89.3ns ± 6%  -0.98%  (p=0.002 n=40+40)
RegexpMatchEasy0_1K-4       812ns ± 3%     815ns ± 3%    ~     (p=0.309 n=40+40)
RegexpMatchEasy1_32-4       103ns ± 6%     103ns ± 5%    ~     (p=0.680 n=40+40)
RegexpMatchEasy1_1K-4      1.01µs ± 4%    1.02µs ± 3%    ~     (p=0.326 n=40+33)
RegexpMatchMedium_32-4      120ns ± 4%     120ns ± 5%    ~     (p=0.834 n=40+40)
RegexpMatchMedium_1K-4     40.1µs ± 3%    39.5µs ± 4%  -1.35%  (p=0.000 n=40+40)
RegexpMatchHard_32-4       2.27µs ± 6%    2.23µs ± 4%  -1.67%  (p=0.011 n=40+40)
RegexpMatchHard_1K-4       67.2µs ± 3%    67.2µs ± 3%    ~     (p=0.149 n=40+40)
Revcomp-4                   1.84s ± 2%     1.86s ± 3%  +0.70%  (p=0.020 n=40+40)
Template-4                 69.0ms ± 4%    69.8ms ± 3%  +1.20%  (p=0.003 n=40+40)
TimeParse-4                 438ns ± 3%     439ns ± 4%    ~     (p=0.650 n=40+40)
TimeFormat-4                412ns ± 3%     412ns ± 3%    ~     (p=0.888 n=40+40)
[Geo mean]                 65.2µs         65.2µs       -0.04%

name                     old speed      new speed      delta
GobDecode-4               108MB/s ± 6%   108MB/s ± 6%    ~     (p=0.855 n=40+40)
GobEncode-4               112MB/s ± 9%   111MB/s ± 8%    ~     (p=0.159 n=40+40)
Gzip-4                   49.2MB/s ± 4%  49.1MB/s ± 3%    ~     (p=0.102 n=40+40)
Gunzip-4                  474MB/s ± 3%   472MB/s ± 3%    ~     (p=0.063 n=40+40)
JSONEncode-4              121MB/s ± 3%   123MB/s ± 3%  +1.62%  (p=0.001 n=40+40)
JSONDecode-4             31.9MB/s ± 3%  31.6MB/s ± 4%    ~     (p=0.070 n=40+40)
GoParse-4                17.9MB/s ± 2%  17.9MB/s ± 3%    ~     (p=0.696 n=40+40)
RegexpMatchEasy0_32-4     355MB/s ± 3%   358MB/s ± 5%  +0.99%  (p=0.002 n=40+40)
RegexpMatchEasy0_1K-4    1.26GB/s ± 3%  1.26GB/s ± 3%    ~     (p=0.381 n=40+40)
RegexpMatchEasy1_32-4     310MB/s ± 5%   310MB/s ± 4%    ~     (p=0.655 n=40+40)
RegexpMatchEasy1_1K-4    1.01GB/s ± 4%  1.01GB/s ± 3%    ~     (p=0.351 n=40+33)
RegexpMatchMedium_32-4   8.32MB/s ± 4%  8.34MB/s ± 5%    ~     (p=0.696 n=40+40)
RegexpMatchMedium_1K-4   25.6MB/s ± 3%  25.9MB/s ± 4%  +1.36%  (p=0.000 n=40+40)
RegexpMatchHard_32-4     14.1MB/s ± 6%  14.3MB/s ± 4%  +1.64%  (p=0.011 n=40+40)
RegexpMatchHard_1K-4     15.2MB/s ± 3%  15.2MB/s ± 3%    ~     (p=0.147 n=40+40)
Revcomp-4                 138MB/s ± 2%   137MB/s ± 3%  -0.70%  (p=0.021 n=40+40)
Template-4               28.1MB/s ± 4%  27.8MB/s ± 3%  -1.19%  (p=0.003 n=40+40)
[Geo mean]               83.7MB/s       83.7MB/s       +0.03%

Change-Id: I2a2b3a942b5c45467491515d201179fd192e65c9
Reviewed-on: https://go-review.googlesource.com/c/141650
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agotest/codegen: fix confusing test cases
Ben Shi [Tue, 16 Oct 2018 04:02:03 +0000 (04:02 +0000)]
test/codegen: fix confusing test cases

ARMv7's MULAF/MULSF/MULAD/MULSD are not fused,
this CL fixes the confusing test cases.

Change-Id: I35022e207e2f0d24a23a7f6f188e41ba8eee9886
Reviewed-on: https://go-review.googlesource.com/c/142439
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Akhil Indurti <aindurti@gmail.com>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
6 years agocmd/compile: don't panic on invalid map key declarations
Daniel Martí [Sun, 7 Oct 2018 11:45:56 +0000 (12:45 +0100)]
cmd/compile: don't panic on invalid map key declarations

In golang.org/cl/75310, the compiler's typechecker was changed so that
map key types were validated at a later stage, to make sure that all the
necessary type information was present.

This still worked for map type declarations, but caused a regression for
top-level map variable declarations. These now caused a fatal panic
instead of a typechecking error.

The cause was that checkMapKeys was run too early, before all
typechecking was done. In particular, top-level map variable
declarations are typechecked as external declarations, much later than
where checkMapKeys was run.

Add a test case for both exported and unexported top-level map
declarations, and add a second call to checkMapKeys at the actual end of
typechecking. Simply moving the one call isn't a good solution either;
the comments expand on that.

Fixes #28058.

Change-Id: Ia5febb01a1d877447cf66ba44fb49a7e0f4f18a5
Reviewed-on: https://go-review.googlesource.com/c/140417
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agointernal/cpu: add invalid option warnings and support to enable cpu features
Martin Möhrmann [Fri, 12 Oct 2018 17:17:21 +0000 (19:17 +0200)]
internal/cpu: add invalid option warnings and support to enable cpu features

This CL adds the ability to enable the cpu feature FEATURE by specifying
FEATURE=on in GODEBUGCPU. Syntax support to enable cpu features is useful
in combination with a preceeding all=off to disable all but some specific
cpu features. Example:

GODEBUGCPU=all=off,sse3=on

This CL implements printing of warnings for invalid GODEBUGCPU settings:
- requests enabling features that are not supported with the current CPU
- specifying values different than 'on' or 'off' for a feature
- settings for unkown cpu feature names

Updates #27218

Change-Id: Ic13e5c4c35426a390c50eaa4bd2a408ef2ee21be
Reviewed-on: https://go-review.googlesource.com/c/141800
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agostrconv: add comment explaining bounded shift in formatBits
Martin Möhrmann [Sun, 6 May 2018 10:19:25 +0000 (12:19 +0200)]
strconv: add comment explaining bounded shift in formatBits

The compiler can generate better code for shifts bounded to be less than 32
and thereby known to be less than any register width.
See https://golang.org/cl/109776.

Change-Id: I0c4c9f0faafa065fce3c10fd328830deb92f9e38
Reviewed-on: https://go-review.googlesource.com/c/111735
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile: avoid string allocations when map key is struct or array literal
Martin Möhrmann [Mon, 4 Jun 2018 06:24:11 +0000 (08:24 +0200)]
cmd/compile: avoid string allocations when map key is struct or array literal

x = map[string(byteslice)] is already optimized by the compiler to avoid a
string allocation. This CL generalizes this optimization to:

x = map[T1{ ... Tn{..., string(byteslice), ...} ... }]
where T1 to Tn is a nesting of struct and array literals.

Found in a hot code path that used a struct of strings made from []byte
slices to make a map lookup.

There are no uses of the more generalized optimization in the standard library.
Passes toolstash -cmp.

MapStringConversion/32/simple    21.9ns ± 2%    21.9ns ± 3%      ~     (p=0.995 n=17+20)
MapStringConversion/32/struct    28.8ns ± 3%    22.0ns ± 2%   -23.80%  (p=0.000 n=20+20)
MapStringConversion/32/array     28.5ns ± 2%    21.9ns ± 2%   -23.14%  (p=0.000 n=19+16)
MapStringConversion/64/simple    21.0ns ± 2%    21.1ns ± 3%      ~     (p=0.072 n=19+18)
MapStringConversion/64/struct    72.4ns ± 3%    21.3ns ± 2%   -70.53%  (p=0.000 n=20+20)
MapStringConversion/64/array     72.8ns ± 1%    21.0ns ± 2%   -71.13%  (p=0.000 n=17+19)

name                           old allocs/op  new allocs/op  delta
MapStringConversion/32/simple      0.00           0.00           ~     (all equal)
MapStringConversion/32/struct      0.00           0.00           ~     (all equal)
MapStringConversion/32/array       0.00           0.00           ~     (all equal)
MapStringConversion/64/simple      0.00           0.00           ~     (all equal)
MapStringConversion/64/struct      1.00 ± 0%      0.00       -100.00%  (p=0.000 n=20+20)
MapStringConversion/64/array       1.00 ± 0%      0.00       -100.00%  (p=0.000 n=20+20)

Change-Id: I483b4d84d8d74b1025b62c954da9a365e79b7a3a
Reviewed-on: https://go-review.googlesource.com/c/116275
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: add intrinsics for runtime/internal/math on 386 and amd64
Martin Möhrmann [Sat, 27 Jan 2018 10:55:34 +0000 (11:55 +0100)]
cmd/compile: add intrinsics for runtime/internal/math on 386 and amd64

Add generic, 386 and amd64 specific ops and SSA rules for multiplication
with overflow and branching based on overflow flags. Use these to intrinsify
runtime/internal/math.MulUinptr.

On amd64
  mul, overflow := math.MulUintptr(a, b)
  if overflow {
is lowered to two instructions:
  MULQ SI
  JO 0x10ee35c

No codegen tests as codegen can not currently test unexported internal runtime
functions.

amd64:
name              old time/op  new time/op  delta
MulUintptr/small  1.16ns ± 5%  0.88ns ± 6%  -24.36%  (p=0.000 n=19+20)
MulUintptr/large  10.7ns ± 1%   1.1ns ± 1%  -89.28%  (p=0.000 n=17+19)

Change-Id: If60739a86f820e5044d677276c21df90d3c7a86a
Reviewed-on: https://go-review.googlesource.com/c/141820
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/compile: avoid implicit bounds checks after explicit checks for append
Martin Möhrmann [Fri, 15 Jun 2018 23:22:07 +0000 (01:22 +0200)]
cmd/compile: avoid implicit bounds checks after explicit checks for append

The generated code for the append builtin already checks if the appended
to slice is large enough and calls growslice if that is not the case.
Trust that this ensures the slice is large enough and avoid the
implicit bounds check when slicing the slice to its new size.

Removes 365 panicslice calls (-14%) from the go binary which
reduces the binary size by ~12kbyte.

Change-Id: I1b88418675ff409bc0b956853c9e95241274d5a6
Reviewed-on: https://go-review.googlesource.com/c/119315
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agoruntime/internal/math: add multiplication with overflow check
Martin Möhrmann [Sat, 27 Jan 2018 10:55:34 +0000 (11:55 +0100)]
runtime/internal/math: add multiplication with overflow check

This CL adds a new internal math package for use by the runtime.
The new package exports a MulUintptr function with uintptr arguments
a and b and returns uintptr(a*b) and whether the full-width product
x*y does overflow the uintptr value range (uintptr(x*y) != x*y).

Uses of MulUinptr in the runtime and intrinsics for performance
will be added in followup CLs.

Updates #21588

Change-Id: Ia5a02eeabc955249118e4edf68c67d9fc0858058
Reviewed-on: https://go-review.googlesource.com/c/91755
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/compile: check order temp has correct type
Keith Randall [Mon, 15 Oct 2018 16:44:22 +0000 (09:44 -0700)]
cmd/compile: check order temp has correct type

Followon from CL 140306

Change-Id: Ic71033d2301105b15b60645d895a076107f44a2e
Reviewed-on: https://go-review.googlesource.com/c/142178
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agotest/codegen: test ppc64 TrailingZeros, OnesCount codegen
Alberto Donizetti [Fri, 12 Oct 2018 18:56:12 +0000 (20:56 +0200)]
test/codegen: test ppc64 TrailingZeros, OnesCount codegen

This change adds codegen tests for the intrinsification on ppc64 of
the OnesCount{64,32,16,8}, and TrailingZeros{64,32,16,8} math/bits
functions.

Change-Id: Id3364921fbd18316850e15c8c71330c906187fdb
Reviewed-on: https://go-review.googlesource.com/c/141897
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
6 years agocmd/compile: fuse before branchelim
Josh Bleecher Snyder [Sun, 27 May 2018 16:03:45 +0000 (09:03 -0700)]
cmd/compile: fuse before branchelim

The branchelim pass works better after fuse.
Running fuse before branchelim also increases
the stability of generated code amidst other compiler changes,
which was the original motivation behind this change.

The fuse pass is not cheap enough to run in its entirety
before branchelim, but the most important half of it is.
This change makes it possible to run "plain fuse" independently
and does so before branchelim.

During make.bash, elimIf occurrences increase from 4244 to 4288 (1%),
and elimIfElse occurrences increase from 989 to 1079 (9%).

Toolspeed impact is marginal; plain fuse pays for itself.

name        old time/op       new time/op       delta
Template          189ms ± 2%        189ms ± 2%    ~     (p=0.890 n=45+46)
Unicode          93.2ms ± 5%       93.4ms ± 7%    ~     (p=0.790 n=48+48)
GoTypes           662ms ± 4%        660ms ± 4%    ~     (p=0.186 n=48+49)
Compiler          2.89s ± 4%        2.91s ± 3%  +0.89%  (p=0.050 n=49+44)
SSA               8.23s ± 2%        8.21s ± 1%    ~     (p=0.165 n=46+44)
Flate             123ms ± 4%        123ms ± 3%  +0.58%  (p=0.031 n=47+49)
GoParser          154ms ± 4%        154ms ± 4%    ~     (p=0.492 n=49+48)
Reflect           430ms ± 4%        429ms ± 4%    ~     (p=1.000 n=48+48)
Tar               171ms ± 3%        170ms ± 4%    ~     (p=0.122 n=48+48)
XML               232ms ± 3%        232ms ± 2%    ~     (p=0.850 n=46+49)
[Geo mean]        394ms             394ms       +0.02%

name        old user-time/op  new user-time/op  delta
Template          236ms ± 5%        236ms ± 4%    ~     (p=0.934 n=50+50)
Unicode           132ms ± 7%        130ms ± 9%    ~     (p=0.087 n=50+50)
GoTypes           861ms ± 3%        867ms ± 4%    ~     (p=0.124 n=48+50)
Compiler          3.93s ± 4%        3.94s ± 3%    ~     (p=0.584 n=49+44)
SSA               12.2s ± 2%        12.3s ± 1%    ~     (p=0.610 n=46+45)
Flate             149ms ± 4%        150ms ± 4%    ~     (p=0.194 n=48+49)
GoParser          193ms ± 5%        191ms ± 6%    ~     (p=0.239 n=49+50)
Reflect           553ms ± 5%        556ms ± 5%    ~     (p=0.091 n=49+49)
Tar               218ms ± 5%        218ms ± 5%    ~     (p=0.359 n=49+50)
XML               299ms ± 5%        298ms ± 4%    ~     (p=0.482 n=50+49)
[Geo mean]        516ms             516ms       -0.01%

name        old alloc/op      new alloc/op      delta
Template         36.3MB ± 0%       36.3MB ± 0%  -0.02%  (p=0.000 n=49+49)
Unicode          29.7MB ± 0%       29.7MB ± 0%    ~     (p=0.270 n=50+50)
GoTypes           126MB ± 0%        126MB ± 0%  -0.34%  (p=0.000 n=50+49)
Compiler          534MB ± 0%        531MB ± 0%  -0.50%  (p=0.000 n=50+50)
SSA              1.98GB ± 0%       1.98GB ± 0%  -0.06%  (p=0.000 n=49+49)
Flate            24.6MB ± 0%       24.6MB ± 0%  -0.29%  (p=0.000 n=50+50)
GoParser         29.5MB ± 0%       29.4MB ± 0%  -0.15%  (p=0.000 n=49+50)
Reflect          87.3MB ± 0%       87.2MB ± 0%  -0.13%  (p=0.000 n=49+50)
Tar              35.6MB ± 0%       35.5MB ± 0%  -0.17%  (p=0.000 n=50+50)
XML              48.2MB ± 0%       48.0MB ± 0%  -0.30%  (p=0.000 n=48+50)
[Geo mean]       83.1MB            82.9MB       -0.20%

name        old allocs/op     new allocs/op     delta
Template           352k ± 0%         352k ± 0%  -0.01%  (p=0.004 n=49+49)
Unicode            341k ± 0%         341k ± 0%    ~     (p=0.341 n=48+50)
GoTypes           1.28M ± 0%        1.28M ± 0%  -0.03%  (p=0.000 n=50+49)
Compiler          4.96M ± 0%        4.96M ± 0%  -0.05%  (p=0.000 n=50+49)
SSA               15.5M ± 0%        15.5M ± 0%  -0.01%  (p=0.000 n=50+49)
Flate              233k ± 0%         233k ± 0%  +0.01%  (p=0.032 n=49+49)
GoParser           294k ± 0%         294k ± 0%    ~     (p=0.052 n=46+48)
Reflect           1.04M ± 0%        1.04M ± 0%    ~     (p=0.171 n=50+47)
Tar                343k ± 0%         343k ± 0%  -0.03%  (p=0.000 n=50+50)
XML                429k ± 0%         429k ± 0%  -0.04%  (p=0.000 n=50+50)
[Geo mean]         812k              812k       -0.02%

Object files grow slightly; branchelim often increases binary size, at least on amd64.

name        old object-bytes  new object-bytes  delta
Template          509kB ± 0%        509kB ± 0%  -0.01%  (p=0.008 n=5+5)
Unicode           224kB ± 0%        224kB ± 0%    ~     (all equal)
GoTypes          1.84MB ± 0%       1.84MB ± 0%  +0.00%  (p=0.008 n=5+5)
Compiler         6.71MB ± 0%       6.71MB ± 0%  +0.01%  (p=0.008 n=5+5)
SSA              21.2MB ± 0%       21.2MB ± 0%  +0.01%  (p=0.008 n=5+5)
Flate             324kB ± 0%        324kB ± 0%  -0.00%  (p=0.008 n=5+5)
GoParser          404kB ± 0%        404kB ± 0%  -0.02%  (p=0.008 n=5+5)
Reflect          1.40MB ± 0%       1.40MB ± 0%  +0.09%  (p=0.008 n=5+5)
Tar               452kB ± 0%        452kB ± 0%  +0.06%  (p=0.008 n=5+5)
XML               596kB ± 0%        596kB ± 0%  +0.00%  (p=0.008 n=5+5)
[Geo mean]       1.04MB            1.04MB       +0.01%

Change-Id: I535c711b85380ff657fc0f022bebd9cb14ddd07f
Reviewed-on: https://go-review.googlesource.com/c/129378
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/compile: provide types for all order-allocated temporaries
Keith Randall [Sat, 6 Oct 2018 21:31:08 +0000 (14:31 -0700)]
cmd/compile: provide types for all order-allocated temporaries

Ensure that we correctly type the stack temps for regular closures,
method function closures, and slice literals.

Then we don't need to override the dummy types later.
Furthermore, this allows order to reuse temporaries of these types.

OARRAYLIT doesn't need a temporary as far as I can tell, so I
removed that case from order.

Change-Id: Ic58520fa50c90639393ff78f33d3c831d5c4acb9
Reviewed-on: https://go-review.googlesource.com/c/140306
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/compile: fix gdb stepping test
Keith Randall [Sun, 14 Oct 2018 19:07:45 +0000 (12:07 -0700)]
cmd/compile: fix gdb stepping test

Not sure why this changed behavior, but seems mostly harmless.

Fixes #28198

Change-Id: Ie25c6e1fcb64912a582c7ae7bf92c4c1642e83cb
Reviewed-on: https://go-review.googlesource.com/c/141649
Reviewed-by: David Chase <drchase@google.com>
6 years agotest/codegen: add tests of FMA for arm/arm64
Ben Shi [Mon, 15 Oct 2018 05:39:23 +0000 (05:39 +0000)]
test/codegen: add tests of FMA for arm/arm64

This CL adds tests of fused multiplication-accumulation
on arm/arm64.

Change-Id: Ic85d5277c0d6acb7e1e723653372dfaf96824a39
Reviewed-on: https://go-review.googlesource.com/c/141652
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agointernal/cpu: expose ARM feature flags for FMA
Akhil Indurti [Fri, 27 Jul 2018 03:46:38 +0000 (23:46 -0400)]
internal/cpu: expose ARM feature flags for FMA

This change exposes feature flags needed to implement an FMA intrinsic
on ARM CPUs via auxv's HWCAP bits. Specifically, it exposes HasVFPv4 to
detect if an ARM processor has the fourth version of the vector floating
point unit. The relevant instruction for this CL is VFMA, emitted in Go
as FMULAD.

Updates #26630.

Change-Id: Ibbc04fb24c2b4d994f93762360f1a37bc6d83ff7
Reviewed-on: https://go-review.googlesource.com/c/126315
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
6 years agocmd/compile: simplify as2 method of *Order
Martin Möhrmann [Sun, 21 Jan 2018 12:02:27 +0000 (13:02 +0100)]
cmd/compile: simplify as2 method of *Order

Merge the two for loops that set up the node lists for
temporaries into one for loop.

Passes toolstash -cmp

Change-Id: Ibc739115f38c8869b0dcfbf9819fdc2fc96962e0
Reviewed-on: https://go-review.googlesource.com/c/141819
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/cgo: simplify switch statement to if statement
avsharapov [Sat, 13 Oct 2018 10:51:16 +0000 (13:51 +0300)]
cmd/cgo: simplify switch statement to if statement

Change-Id: Ie7dce45d554fde69d682680f55abba6a7fc55036
Reviewed-on: https://go-review.googlesource.com/c/142017
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agopprof: replace bits = bits + "..." to bits += "..." where bits is a string.
Ivan Sharavuev [Sat, 13 Oct 2018 10:25:20 +0000 (13:25 +0300)]
pprof: replace bits = bits + "..." to bits += "..." where bits is a string.

Change-Id: Ic77ebbdf2670b7fdf2c381cd1ba768624b07e57c
Reviewed-on: https://go-review.googlesource.com/c/141998
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agosrc/cmd/compile/internal/ssa: replace `s = s + x' => 's += x'.
OlgaVlPetrova [Sat, 13 Oct 2018 11:53:45 +0000 (14:53 +0300)]
src/cmd/compile/internal/ssa: replace `s = s + x' => 's += x'.

Change-Id: I1f399a8a0aa200bfda01f97f920b1345e59956ba
Reviewed-on: https://go-review.googlesource.com/c/142057
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agotest/codegen: add tests for multiplication-subtraction
Ben Shi [Mon, 15 Oct 2018 02:25:10 +0000 (02:25 +0000)]
test/codegen: add tests for multiplication-subtraction

This CL adds tests for armv7's MULS and arm64's MSUBW.

Change-Id: Id0fd5d26fd477e4ed14389b0d33cad930423eb5b
Reviewed-on: https://go-review.googlesource.com/c/141651
Run-TryBot: Ben Shi <powerman1st@163.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: reuse temporaries in order pass
Keith Randall [Fri, 5 Oct 2018 15:54:50 +0000 (08:54 -0700)]
cmd/compile: reuse temporaries in order pass

Instead of allocating a new temporary each time one
is needed, keep a list of temporaries which are free
(have already been VARKILLed on every path) and use
one of them.

Should save a lot of stack space. In a function like this:

func main() {
     fmt.Printf("%d %d\n", 2, 3)
     fmt.Printf("%d %d\n", 4, 5)
     fmt.Printf("%d %d\n", 6, 7)
}

The three [2]interface{} arrays used to hold the ... args
all use the same autotmp, instead of 3 different autotmps
as happened previous to this CL.

Change-Id: I2d728e226f81e05ae68ca8247af62014a1b032d3
Reviewed-on: https://go-review.googlesource.com/c/140301
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>