]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agoos: use poller when NewFile is called with a blocking descriptor.
Nick Patavalis [Sun, 11 Mar 2018 17:11:33 +0000 (19:11 +0200)]
os: use poller when NewFile is called with a blocking descriptor.

If NewFile is called with a file descriptor that is already set to
non-blocking mode, it tries to return a pollable file (one for which
SetDeadline methods work) by adding the filedes to the poll/netpoll
mechanism. If called with a filedes in blocking mode, it returns a
non-pollable file, as it always did.

Fixes #22939
Updates #24331

Change-Id: Id54c8be1b83e6d35e14e76d7df0e57a9fd64e176
Reviewed-on: https://go-review.googlesource.com/100077
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: use Block.Likely to put likely branch first
David Chase [Thu, 5 Apr 2018 20:14:42 +0000 (16:14 -0400)]
cmd/compile: use Block.Likely to put likely branch first

When a neither of a conditional block's successors follows,
the block must end with a conditional branch followed by a
an unconditional branch.  If the (conditional) branch is
"unlikely", invert it and swap successors to make it
likely instead.

This doesn't matter to most benchmarks on amd64, but in one
instance on amd64 it caused a 30% improvement, and it is
otherwise harmless.  The problematic loop is

for i := 0; i < w; i++ {
if pw[i] != 0 {
return true
}
}

compiled under GOEXPERIMENT=preemptibleloops
This the very worst-case benchmark for that experiment.

Also in this CL is a commoning up of heavily-repeated
boilerplate, which made it much easier to see that the
changes were applied correctly.  In the future this should
allow un-exporting of SSAGenState.Branches once the
boilerplate-replacement is done everywhere.

Change-Id: I0e5ded6eeb3ab1e3e0138e12d54c7e056bd99335
Reviewed-on: https://go-review.googlesource.com/104977
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: use existing flag bits to record 'used' property of Names (cleanup)
Robert Griesemer [Wed, 11 Apr 2018 00:02:07 +0000 (17:02 -0700)]
cmd/compile: use existing flag bits to record 'used' property of Names (cleanup)

Change-Id: I804d5ab111e33bd2c2554e2bac75b5273b0b4160
Reviewed-on: https://go-review.googlesource.com/106121
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agonet/http: skip a test on netbsd-arm
Benny Siegert [Wed, 11 Apr 2018 10:38:04 +0000 (12:38 +0200)]
net/http: skip a test on netbsd-arm

TestServerDuplicateBackgroundRead has been causing crashes on the
netbsd-arm-bsiegert builder, with the system becoming completely
unresponsive (probably deadlocked). Skip this test while the crash
is under investigation.

Upstream bug report is http://gnats.netbsd.org/53173.

Change-Id: Ib48f19005cf2cbba8a27e75e689c2acb025d8870
Reviewed-on: https://go-review.googlesource.com/106295
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
7 years agotest/codegen: port last stack and memcombining tests
Alberto Donizetti [Wed, 11 Apr 2018 15:03:14 +0000 (17:03 +0200)]
test/codegen: port last stack and memcombining tests

And delete them from asm_test.

Also delete an arm64 cmov test has been already ported to the new test
harness.

Change-Id: I4458721e1f512bc9ecbbe1c22a2c9c7109ad68fe
Reviewed-on: https://go-review.googlesource.com/106335
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
7 years agocmd/internal/obj/arm64: support SWPD/SWPW/SWPH/SWPB
Ben Shi [Tue, 20 Mar 2018 08:25:08 +0000 (08:25 +0000)]
cmd/internal/obj/arm64: support SWPD/SWPW/SWPH/SWPB

SWPD/SWPW/SWPH/SWPB were introduced in ARMv8.1. They swap content
of register and memory atomically. And their difference is
SWPD: 64-bit double word data
SWPW: 32-bit word data (zero extended to 64-bit)
SWPH: 16-bit half word data (zero extended to 64-bit)
SWPB: 8-bit byte data (zero extended to 64-bit)

This CL implements them in the arm64 assembler.

Change-Id: I2d9fb2310674bd92693531210e187143e7eed602
Reviewed-on: https://go-review.googlesource.com/101516
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agogo/build, runtime/internal/sys: reserve RISC-V arch names
Tobias Klauser [Wed, 11 Apr 2018 09:12:28 +0000 (11:12 +0200)]
go/build, runtime/internal/sys: reserve RISC-V arch names

In #17528 it was discussed (off-topic to the actual issue) to reserve
GOARCH names for the RISC-V architecture. With the first RISC-V
Linux-capable development boards released (e.g. HiFive Unleashed),
Linux distributions being ported to RISC-V (e.g. Debian, Fedora) and
RISC-V support being added to gccgo (CL 96377), it becomes more likely
that Go software (and maybe Go itself) will be ported as well.

Add riscv and riscv64 (which is already used by gccgo), so Go 1.11 will
already recognize "*_riscv{,64}.go" as reserved files.

Change-Id: I042aab19c68751d82ea513e40f7b1d7e1ad924ea
Reviewed-on: https://go-review.googlesource.com/106256
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agointernal/bytealg: add optimized Compare for arm64
fanzha02 [Tue, 20 Mar 2018 03:53:21 +0000 (03:53 +0000)]
internal/bytealg: add optimized Compare for arm64

Use LDP instructions to load 16 bytes per loop when the source length is long. Specially
process the 8 bytes length, 4 bytes length and 2 bytes length to get a better performance.

Benchmark result:
name                           old time/op   new time/op    delta
BytesCompare/1-8                21.0ns ± 0%    10.5ns ± 0%      ~     (p=0.079 n=4+5)
BytesCompare/2-8                11.5ns ± 0%    10.5ns ± 0%    -8.70%  (p=0.008 n=5+5)
BytesCompare/4-8                13.5ns ± 0%    10.0ns ± 0%   -25.93%  (p=0.008 n=5+5)
BytesCompare/8-8                28.8ns ± 0%     9.5ns ± 0%      ~     (p=0.079 n=4+5)
BytesCompare/16-8               40.5ns ± 0%    10.5ns ± 0%   -74.07%  (p=0.008 n=5+5)
BytesCompare/32-8               64.6ns ± 0%    12.5ns ± 0%   -80.65%  (p=0.008 n=5+5)
BytesCompare/64-8                112ns ± 0%      16ns ± 0%   -85.27%  (p=0.008 n=5+5)
BytesCompare/128-8               208ns ± 0%      24ns ± 0%   -88.22%  (p=0.008 n=5+5)
BytesCompare/256-8               400ns ± 0%      50ns ± 0%   -87.62%  (p=0.008 n=5+5)
BytesCompare/512-8               785ns ± 0%      82ns ± 0%   -89.61%  (p=0.008 n=5+5)
BytesCompare/1024-8             1.55µs ± 0%    0.14µs ± 0%      ~     (p=0.079 n=4+5)
BytesCompare/2048-8             3.09µs ± 0%    0.27µs ± 0%      ~     (p=0.079 n=4+5)
CompareBytesEqual-8             39.0ns ± 0%    12.0ns ± 0%   -69.23%  (p=0.008 n=5+5)
CompareBytesToNil-8             8.57ns ± 5%    8.23ns ± 2%    -3.99%  (p=0.016 n=5+5)
CompareBytesEmpty-8             7.37ns ± 0%    7.36ns ± 4%      ~     (p=0.690 n=5+5)
CompareBytesIdentical-8         7.39ns ± 0%    7.46ns ± 2%      ~     (p=0.667 n=5+5)
CompareBytesSameLength-8        17.0ns ± 0%    10.5ns ± 0%   -38.24%  (p=0.008 n=5+5)
CompareBytesDifferentLength-8   17.0ns ± 0%    10.5ns ± 0%   -38.24%  (p=0.008 n=5+5)
CompareBytesBigUnaligned-8      1.58ms ± 0%    0.19ms ± 0%   -88.31%  (p=0.016 n=4+5)
CompareBytesBig-8               1.59ms ± 0%    0.19ms ± 0%   -88.27%  (p=0.016 n=5+4)
CompareBytesBigIdentical-8      7.01ns ± 0%    6.60ns ± 3%    -5.91%  (p=0.008 n=5+5)

name                           old speed     new speed      delta
CompareBytesBigUnaligned-8     662MB/s ± 0%  5660MB/s ± 0%  +755.15%  (p=0.016 n=4+5)
CompareBytesBig-8              661MB/s ± 0%  5636MB/s ± 0%  +752.57%  (p=0.016 n=5+4)
CompareBytesBigIdentical-8     150TB/s ± 0%   159TB/s ± 3%    +6.27%  (p=0.008 n=5+5)

This is resubmit of CL90175.

Change-Id: Ie841daedb3123a68dd2554f27ebef0b3f8a855c2
Reviewed-on: https://go-review.googlesource.com/101635
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/internal/obj/arm64: add support for a series of load/store with register offset...
fanzha02 [Fri, 26 Jan 2018 08:15:49 +0000 (08:15 +0000)]
cmd/internal/obj/arm64: add support for a series of load/store with register offset instrucitons

The patch adds support for arm64 instructions LDRB, LDRH, LDRSB,
LDRSH, LDRSW, STR, STRB and STRH with register offset.

Test cases are also added.

Change-Id: I8d17fddd2963c0bc366e12b00bac49b93f3f0957
Reviewed-on: https://go-review.googlesource.com/91575
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: use internal/cpu instead of support_avx
Tobias Klauser [Wed, 11 Apr 2018 07:57:18 +0000 (09:57 +0200)]
runtime: use internal/cpu instead of support_avx

After CL 104636 cpu.X86.HasAVX is set early enough that it can be used
to determine useAVXmemmove. Use it and remove support_avx.

Change-Id: Ib7a627bede2bf96c92362507e742bd833cb42a74
Reviewed-on: https://go-review.googlesource.com/106235
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: document GOTOOLDIR environment variable
Alberto Donizetti [Wed, 11 Apr 2018 08:50:09 +0000 (10:50 +0200)]
cmd/go: document GOTOOLDIR environment variable

Also, make the variables list sorted.

Fixes #24794

Change-Id: I55f77004b00391875d26df8e55e54d79cef168dc
Reviewed-on: https://go-review.googlesource.com/106255
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc/install.html: address comments from review of previous edit
Rob Pike [Tue, 10 Apr 2018 23:49:32 +0000 (09:49 +1000)]
doc/install.html: address comments from review of previous edit

Point out that one can just run the commands now; it's not necessary
to log out first.

Change-Id: I48d0cc0273d97ba54ce59b3a3bbcae0b5af9aaef
Reviewed-on: https://go-review.googlesource.com/106195
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc: switch from spaces to tabs for indentation in example code
Jeremy Schlatter [Wed, 11 Apr 2018 03:22:59 +0000 (03:22 +0000)]
doc: switch from spaces to tabs for indentation in example code

This is the standard that gofmt uses.

Change-Id: I30e70611ccd0d3bbaa500fa84e9d1ba75970f0bc
GitHub-Last-Rev: b2b7af0df9759ad447b6fd3551ea356804d136b2
GitHub-Pull-Request: golang/go#24802
Reviewed-on: https://go-review.googlesource.com/106123
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: use fixed TLS offsets on darwin/amd64 and darwin/386
Keith Randall [Tue, 10 Apr 2018 22:56:14 +0000 (22:56 +0000)]
runtime: use fixed TLS offsets on darwin/amd64 and darwin/386

Fixes #23617

Note that this CL does not affect darwin/arm and darwin/arm64,
still TBD what, if anything, needs to be done for those.

This is a fix of CL 105975 which was reverted in CL 106155.
Needed to use movl instead of movq for 386.

Change-Id: I0db7f8087173869e60cc22c6c3124fa0a0739b46
Reviewed-on: https://go-review.googlesource.com/106156
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: better error message when referring to ambiguous method/field
Robert Griesemer [Tue, 10 Apr 2018 22:56:17 +0000 (15:56 -0700)]
cmd/compile: better error message when referring to ambiguous method/field

Fixes #14321.

Change-Id: I9c92c767b01cf7938c4808a8fef9f2936fc667bc
Reviewed-on: https://go-review.googlesource.com/106119
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile: fix check that ensures main.main is a function
Matthew Dempsky [Tue, 10 Apr 2018 23:01:23 +0000 (16:01 -0700)]
cmd/compile: fix check that ensures main.main is a function

The check was previously disallowing package main from even importing
a non-function symbol named "main".

Fixes #24801.

Change-Id: I849b9713890429f0a16860ef16b5dc7e970d04a4
Reviewed-on: https://go-review.googlesource.com/106120
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: omit unnecessary interface method expression wrappers
Matthew Dempsky [Tue, 10 Apr 2018 22:13:49 +0000 (15:13 -0700)]
cmd/compile: omit unnecessary interface method expression wrappers

We'll always generate method expression wrappers for declared
interface types in their own package, so no need to generate them in
downstream packages.

Noticed by gri@ while looking into #21282.

Change-Id: I4fb7051b4e15297933da05fdd2b111d6b8f4178e
Reviewed-on: https://go-review.googlesource.com/106175
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agogo/doc: tune association of a function with a type
Agniva De Sarker [Sun, 8 Apr 2018 10:37:25 +0000 (16:07 +0530)]
go/doc: tune association of a function with a type

Previously, we used to associate a function with its first returned type
assuming that it is a factory function for that type.

However, a function may return multiple types in which case it is usually
doing something else. Check for multiple return types, and treat it as
a normal function in that case. Maintain same behavior if the function
returns just one type.

Fixes #12839

Change-Id: Ic4ac11d322996f216f593b71f4e61ad4270d5213
Reviewed-on: https://go-review.googlesource.com/105575
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/trace: pprof computation for span types
Hana Kim [Wed, 4 Apr 2018 18:24:02 +0000 (14:24 -0400)]
cmd/trace: pprof computation for span types

/spanio, /spanblock, /spansched, /spansyscall provide
the pprof-style summary of span execution's
io, block, scheduling, syscall latency distributions
respectively.

The computation logic for /io, /block, /sched, /syscall
analysis was refactored and extended for reuse in these
new types of analysis. Upon the analysis query, we create
a map of goroutine id to time intervals based on the query
parameter, that represents the interesting time intervals
of matching goroutines. Only the events from the matching
goroutines that fall into the intervals are considered
in the pprof computation.

The new endpoints are not yet hooked into other span
analysis page (e.g. /userspan) yet.

Change-Id: I80c3396e45a2d6631758710de67d132e5832c7ce
Reviewed-on: https://go-review.googlesource.com/105822
Reviewed-by: Heschi Kreinick <heschi@google.com>
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoRevert "runtime: use fixed TLS offsets on darwin/amd64 and darwin/386"
Keith Randall [Tue, 10 Apr 2018 18:40:01 +0000 (18:40 +0000)]
Revert "runtime: use fixed TLS offsets on darwin/amd64 and darwin/386"

This reverts commit 76e92d1c9e2943de7093af14d58663fa2993e608.

Reason for revert: Seems to have broken the darwin/386 builder, the toolchain is barfing on the new inline assembly.

Change-Id: Ic83fa3c85148946529c5fd47d1e1669898031ace
Reviewed-on: https://go-review.googlesource.com/106155
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocrypto/elliptic: improve P256 implementation on amd64 a bit
Vlad Krasnov [Fri, 9 Mar 2018 00:07:42 +0000 (16:07 -0800)]
crypto/elliptic: improve P256 implementation on amd64 a bit

Minor modifications to the optimized amd64 implememntation.

* Reduce window size: reduces size of the lookup tables by 40%
* Revised scalar inversion formula, with less operations
* Field square function now uses intental loop, saving call overhead

This change will serve as a basis for an arm64 implementation.

Performance results on Skylake MacBook Pro:

pkg:crypto/elliptic goos:darwin goarch:amd64
BaseMultP256      17.8µs ± 1%    17.5µs ± 1%  -1.41%  (p=0.003 n=10+10)
ScalarMultP256    70.7µs ± 1%    68.9µs ± 2%  -2.57%  (p=0.000 n=9+9)
pkg:crypto/ecdsa goos:darwin goarch:amd64
SignP256          32.7µs ± 1%    31.4µs ± 1%  -3.96%  (p=0.000 n=10+8)
VerifyP256        95.1µs ± 1%    93.5µs ± 2%  -1.73%  (p=0.001 n=10+9)

name            old alloc/op   new alloc/op   delta
pkg:crypto/elliptic goos:darwin goarch:amd64
BaseMultP256        288B ± 0%      288B ± 0%    ~     (all equal)
ScalarMultP256      256B ± 0%      256B ± 0%    ~     (all equal)
pkg:crypto/ecdsa goos:darwin goarch:amd64
SignP256          2.90kB ± 0%    2.90kB ± 0%    ~     (all equal)
VerifyP256          976B ± 0%      976B ± 0%    ~     (all equal)

name            old allocs/op  new allocs/op  delta
pkg:crypto/elliptic goos:darwin goarch:amd64
BaseMultP256        6.00 ± 0%      6.00 ± 0%    ~     (all equal)
ScalarMultP256      5.00 ± 0%      5.00 ± 0%    ~     (all equal)
pkg:crypto/ecdsa goos:darwin goarch:amd64
SignP256            34.0 ± 0%      34.0 ± 0%    ~     (all equal)
VerifyP256          17.0 ± 0%      17.0 ± 0%    ~     (all equal)

Change-Id: I3f0e2e197a54e7bc7916dedc5dbf085e2c4aea24
Reviewed-on: https://go-review.googlesource.com/99622
Reviewed-by: Vlad Krasnov <vlad@cloudflare.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Vlad Krasnov <vlad@cloudflare.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/trace: skip GoroutineInSyscall in case time order err
Hana Kim [Mon, 9 Apr 2018 14:31:10 +0000 (10:31 -0400)]
cmd/trace: skip GoroutineInSyscall in case time order err

All tests involving trace collection and parsing
still need handling of failures caused by #16755
(Timestamp issue)

Fixes #24738

Change-Id: I6cd0f9c6f49854a22fad6fce1a00964c168aa614
Reviewed-on: https://go-review.googlesource.com/105821
Reviewed-by: Peter Weinberger <pjw@google.com>
7 years agonet/http: don't sniff Content-type in Server when X-Content-Type-Options:nosniff
Mike Samuel [Tue, 23 Jan 2018 19:27:47 +0000 (14:27 -0500)]
net/http: don't sniff Content-type in Server when X-Content-Type-Options:nosniff

The docs for ResponseWriter.Write say
// If the Header
// does not contain a Content-Type line, Write adds a Content-Type set
// to the result of passing the initial 512 bytes of written data to
// DetectContentType.

The header X-Content-Type-Options:nosniff is an explicit directive that
content-type should not be sniffed.

This changes the behavior of Response.WriteHeader so that, when
there is an X-Content-Type-Options:nosniff header, but there is
no Content-type header, the following happens:
1.  A Content-type:application/octet-stream is added
2.  A warning is logged via the server's logging mechanism.

Previously, a content-type would have been silently added based on
heuristic analysis of the first 512B which might allow a hosted
GIF like http://www.thinkfu.com/blog/gifjavascript-polyglots to be
categorized as JavaScript which might allow a CSP bypass, loading
as a script despite `Content-Security-Policy: script-src 'self' `.

----

https://fetch.spec.whatwg.org/#x-content-type-options-header
defines the X-Content-Type-Options header.

["Polyglots: Crossing Origins by Crossing Formats"](http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.905.2946&rep=rep1&type=pdf)
explains Polyglot attacks in more detail.

Change-Id: I2c8800d2e4b4d10d9e08a0e3e5b20334a75f03c0
Reviewed-on: https://go-review.googlesource.com/89275
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agomisc/ios: make detect.go more robust
Elias Naur [Sat, 7 Apr 2018 15:22:43 +0000 (17:22 +0200)]
misc/ios: make detect.go more robust

To enable the exec wrapper go_darwin_arm_exec.go to run binaries
on iOS devices, the GOIOS_DEV_ID variable needs to be set to a code
signing identity. The program detect.go attempts to detect suitable
values for GOIOS_DEV_ID (along with GOIOS_APP_ID and GOIOS_TEAM_ID).

Before this change, detect.go would use "security find-identity
-p codesigning -v" to list all available identities for code signing
and pick the first one with "iPhone Developer" in its name. However,
that pick might be invalid since if it was replaced by an identity
issued later.

For example, on the mobile builder:

$ security find-identity -p codesigning -v
  1) 0E251DE41FE4490574E475AC320B47F58D6D3635 "lldb_codesign"
  2) 0358588D07AA6A19478981BA405F40A97F95F187 "iPhone Developer: xxx@xxx (2754T98W8E)"
  3) FC6D96F24A3223C98BF7A2C2C5194D82E04CD23E "iPhone Developer: xxx@xxx (2754T98W8E)"
     3 valid identities found

In this case, the identity 0358588D07AA6A19478981BA405F40A97F95F187
is picked by detect.go even though it has been invalidated by
FC6D96F24A3223C98BF7A2C2C5194D82E04CD23E.

Instead of attempting to find an identity from the "security
find-identity" list, use the identity from the CommonName in the
embedded certificate in the provisioning file. The CommonName only
lists the identity name (iPhone Developer: xxx@xxx (2754T98W8E)),
not the fingerprint (FC6D96F24A3223C98BF7A2C2C5194D82E04CD23E), but
fortunately the codesign tool accepts both.

Identity names may not be unique, as demonstrated by the example,
but that will result in an ambiguity error at codesigning instead of
a more obscure error about an invalid identity when
go_darwin_arm_exec.go runs a binary.
The fix is then to delete the invalid identity from the system
keychain.

While here, find all connected devices instead of the first connected
and only consider provision files that covers them all. This matters
for the mobile builder where two devices are connected.

Change-Id: I6beb59ace3fc5e071ba76222a20a607765943989
Reviewed-on: https://go-review.googlesource.com/105436
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
7 years agoruntime: use internal/cpu in alginit
Meng Zhuo [Tue, 10 Apr 2018 08:42:44 +0000 (16:42 +0800)]
runtime: use internal/cpu in alginit

After CL 104636 the feature flags in internal/cpu are initialized before
alginit and can now be used for aeshash feature detection. Also remove
now unused runtime variables:
x86:
support_ssse3
support_sse42
support_aes
arm64:
support_aes

Change-Id: I2f64198d91750eaf3c6cf2aac6e9e17615811ec8
Reviewed-on: https://go-review.googlesource.com/106015
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agomisc/ios,runtime/cgo: remove SIGINT handshake for the iOS exec wrapper
Elias Naur [Tue, 10 Apr 2018 11:26:05 +0000 (13:26 +0200)]
misc/ios,runtime/cgo: remove SIGINT handshake for the iOS exec wrapper

Once upon a time, the iOS exec wrapper needed to change the current
working directory for the binary being tested. To allow that, the
runtime raised a SIGINT signal that the wrapper caught, changed the
working directory and resumed the process.

These days, the current working directory is passed from the wrapper
to the runtime through a special entry in the app metadata and the
SIGINT handshake is not necessary anymore.

Remove the signaling from the runtime and the exec harness.

Change-Id: Ia53bcc9e4724d2ca00207e22b91ce80a05271b55
Reviewed-on: https://go-review.googlesource.com/106096
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
7 years agointernal/cpu: align capability definitions for x86 with other architectures
Martin Möhrmann [Sun, 4 Feb 2018 19:39:39 +0000 (20:39 +0100)]
internal/cpu: align capability definitions for x86 with other architectures

Use constant masks and align the definition of isSet with
arm64 and ppc64x.

Change-Id: I0c6eae30da5e3ce797cde0dab4a39855d4d245d9
Reviewed-on: https://go-review.googlesource.com/94759
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agointernal/cpu,runtime: call cpu.initialize before alginit
Meng Zhuo [Tue, 10 Apr 2018 14:33:03 +0000 (22:33 +0800)]
internal/cpu,runtime: call cpu.initialize before alginit

runtime.alginit needs runtime/support_{aes,ssse3,sse41} feature flag
to init aeshash function but internal/cpu.init not be called yet.
This CL will call internal/cpu.initialize before runtime.alginit, so
that we can move all cpu features related code to internal/cpu.

Change-Id: I00b8e403ace3553f8c707563d95f27dade0bc853
Reviewed-on: https://go-review.googlesource.com/104636
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: remove last manual node copies
Daniel Martí [Thu, 5 Apr 2018 13:45:38 +0000 (14:45 +0100)]
cmd/compile: remove last manual node copies

When I added the Node.copy method, I converted most of the occurrences
but missed a few.

One of them, used only for gdata, was an unnecessary copy given that
gdata does not modify the node it is passed.

No allocation changes in compilebench.

Passes toolstash -cmp on std cmd.

Change-Id: I7fba5212377b75c6d6b3785e594a30568ff0732e
Reviewed-on: https://go-review.googlesource.com/104937
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agodebug/elf: add missing EM_AARCH64 to machineStrings
Tobias Klauser [Tue, 10 Apr 2018 07:57:14 +0000 (09:57 +0200)]
debug/elf: add missing EM_AARCH64 to machineStrings

EM_AARCH64 is defined as a constant, but the corresponding entry in
machineStrings is missing. Add it.

Change-Id: I6506404386efe608877095e635a290bbc0686215
Reviewed-on: https://go-review.googlesource.com/106035
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: remove support_bmi{1,2}
Tobias Klauser [Tue, 10 Apr 2018 10:19:24 +0000 (12:19 +0200)]
runtime: remove support_bmi{1,2}

The code reading these variables was removed in CL 41476. They are only
set but never read now, so remove them.

Change-Id: I6b0b8d813e9a3ec2a13586ff92746e00ad1b5bf0
Reviewed-on: https://go-review.googlesource.com/106095
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: use fixed TLS offsets on darwin/amd64 and darwin/386
Keith Randall [Mon, 9 Apr 2018 23:00:54 +0000 (16:00 -0700)]
runtime: use fixed TLS offsets on darwin/amd64 and darwin/386

Fixes #23617

Note that this CL does not affect darwin/arm and darwin/arm64,
still TBD what, if anything, needs to be done for those.

Change-Id: Ie1ee02a9f4d4d1fb9cd5dc432d900f926cc157db
Reviewed-on: https://go-review.googlesource.com/105975
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotext/template: copy Decl field when copying PipeNode
Ian Lance Taylor [Tue, 10 Apr 2018 14:08:45 +0000 (07:08 -0700)]
text/template: copy Decl field when copying PipeNode

Fixes #24791

Change-Id: I62ac17313e6e09796586911d88191a36d67f9aa1
Reviewed-on: https://go-review.googlesource.com/106115
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: compile packages imported by runtime with -+
Ian Lance Taylor [Tue, 10 Apr 2018 03:01:36 +0000 (20:01 -0700)]
cmd/go: compile packages imported by runtime with -+

Packages imported by the runtime should follow the same memory
allocation rules as the runtime package.

Change-Id: Id3fadb05605acbd6ac61d7fe6068a2add8bb2308
Reviewed-on: https://go-review.googlesource.com/105995
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agotest/codegen: port arm64 BIC/EON/ORN and masking tests
Alberto Donizetti [Tue, 10 Apr 2018 09:20:20 +0000 (11:20 +0200)]
test/codegen: port arm64 BIC/EON/ORN and masking tests

And delete them from asm_test.

Change-Id: I24f421b87e8cb4770c887a6dfd58eacd0088947d
Reviewed-on: https://go-review.googlesource.com/106056
Reviewed-by: Keith Randall <khr@golang.org>
7 years agotest/codegen: port last remaining misc bit/arithmetic tests
Alberto Donizetti [Mon, 9 Apr 2018 12:24:35 +0000 (14:24 +0200)]
test/codegen: port last remaining misc bit/arithmetic tests

And delete them from asm_test.

Change-Id: I9a75efe9858ef9d7ac86065f860c2ae3f25b0941
Reviewed-on: https://go-review.googlesource.com/105597
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: sort method sets using package height
Matthew Dempsky [Thu, 5 Apr 2018 19:48:28 +0000 (12:48 -0700)]
cmd/compile: sort method sets using package height

Also, when statically building itabs, compare *types.Sym instead of
name alone so that method sets with duplicate non-exported methods are
handled correctly.

Fixes #24693.

Change-Id: I2db8a3d6e80991a71fef5586a15134b6de116269
Reviewed-on: https://go-review.googlesource.com/105039
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: add package height to export data
Matthew Dempsky [Thu, 5 Apr 2018 21:29:32 +0000 (14:29 -0700)]
cmd/compile: add package height to export data

A package's height is defined as the length of the longest import path
between itself and a leaf package (i.e., package with no imports).

We can't rely on knowing the path of the package being compiled, so
package height is useful for defining a package ordering.

Updates #24693.

Change-Id: I965162c440b6c5397db91b621ea0be7fa63881f1
Reviewed-on: https://go-review.googlesource.com/105038
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: fix constant pointer comparison failure
Matthew Dempsky [Sat, 7 Apr 2018 10:08:46 +0000 (03:08 -0700)]
cmd/compile: fix constant pointer comparison failure

Previously, constant pointer-typed expressions could use either Mpint
or NilVal as their Val depending on their construction, but const.go
expects each type to have a single corresponding Val kind.

This CL changes pointer-typed expressions to exclusively use Mpint.

Fixes #21221.

Change-Id: I6ba36c9b11eb19a68306f0b296acb11a8c254c41
Reviewed-on: https://go-review.googlesource.com/105315
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: refactor symbol sorting logic
Matthew Dempsky [Mon, 9 Apr 2018 20:57:56 +0000 (13:57 -0700)]
cmd/compile: refactor symbol sorting logic

This used to be duplicated in methcmp and siglt, because Sig used its
own representation for Syms. Instead, just use Syms, and add a
(*Sym).Less method that both methcmp and siglt can use.

Also, prune some impossible cases purportedly related to blank
methods: the Go spec disallows blank methods in interface method sets,
and addmethod drops blank methods without actually recording them in
the type's method set.

Passes toolstash-check.

Updates #24693.

Change-Id: I24e981659b68504d71518160486989a82505f513
Reviewed-on: https://go-review.googlesource.com/105936
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: rename gc.exportname to types.IsExported
Matthew Dempsky [Mon, 9 Apr 2018 22:22:01 +0000 (15:22 -0700)]
cmd/compile: rename gc.exportname to types.IsExported

gofmt -r 'exportname(s) -> types.IsExported(s)'

Passes toolstash-check.

Change-Id: I6b428bd039c135be66d8b81c325d4e08bae69f24
Reviewed-on: https://go-review.googlesource.com/105938
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoapi: remove unnecessary lines from except.txt
Daniel Martí [Sat, 7 Apr 2018 15:24:52 +0000 (16:24 +0100)]
api: remove unnecessary lines from except.txt

When I added the text/template/parse lines, I thought that both removed
and added APIs should be listed here (i.e. both -pkg and +pkg lines).
However that was wrong, as one can see by reading cmd/api/goapi.go, or
seeing how removing the +pkg lines does not break the API test.

Change-Id: I0a8dcd6db44762dadb58728acfb844bf118c9d45
Reviewed-on: https://go-review.googlesource.com/105376
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: optimize comparisons using load merging where available
Michael Munday [Mon, 26 Mar 2018 20:18:27 +0000 (21:18 +0100)]
cmd/compile: optimize comparisons using load merging where available

Multi-byte comparison operations were used on amd64, arm64, i386
and s390x for comparisons with constant arrays, but only amd64 and
i386 for comparisons with string constants. This CL combines the
check for platform capability, since they have the same requirements,
and also enables both on ppc64le which also supports load merging.

Note that these optimizations currently use little endian byte order
which results in byte reversal instructions on s390x. This should
be fixed at some point.

Change-Id: Ie612d13359b50c77f4d7c6e73fea4a59fa11f322
Reviewed-on: https://go-review.googlesource.com/102558
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agodoc: add a note about loading profile files
Daniel Martí [Sun, 8 Apr 2018 16:52:16 +0000 (17:52 +0100)]
doc: add a note about loading profile files

If one is somewhat new to the command line or shell, it might be
surprising that changes applied to a file like $HOME/.profile will
seemingly not take effect, even if new shells are started.

Add a note about how shells usually only load these when the user logs
into a machine, to minimize the amount of people stuck and confused by
this.

Fixes #24756.

Change-Id: Ic68d8c97933f3f080b151a107633ecad76a163a4
Reviewed-on: https://go-review.googlesource.com/105557
Reviewed-by: Agniva De Sarker <agniva.quicksilver@gmail.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
7 years agoencoding/json: remove unnecessary else conditions
bontequero [Mon, 9 Apr 2018 08:06:18 +0000 (11:06 +0300)]
encoding/json: remove unnecessary else conditions

Fixes golint warning about "if block ends with a return statement, so drop this else and outdent its block".

Change-Id: Id17ad0bf37ba939386b177b709e9e3c067d8ba21
Reviewed-on: https://go-review.googlesource.com/105736
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/internal/obj/x86: remove redundant code (fix TODO)
isharipo [Fri, 6 Apr 2018 19:22:03 +0000 (22:22 +0300)]
cmd/internal/obj/x86: remove redundant code (fix TODO)

Special cases described in TODO comments are
fixed in https://golang.org/cl/6901.

One of those blocks was needed until old 6a assembler was removed.
This happened in https://golang.org/cl/12784.

Fixes #24734 (Also contains more details and reasoning)

Change-Id: If1f2f155d36ab236b16ae6f309a0716e00aa6371
Reviewed-on: https://go-review.googlesource.com/105156
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocrypto/rsa: improve the comment of PublicKey.Size
Filippo Valsorda [Mon, 9 Apr 2018 18:43:52 +0000 (14:43 -0400)]
crypto/rsa: improve the comment of PublicKey.Size

Change-Id: Ic507cb740395e76d1d011a5a2f395b96c3d172a2
Reviewed-on: https://go-review.googlesource.com/105915
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agotest: check that unaligned load-add opcodes work.
Keith Randall [Mon, 9 Apr 2018 17:25:15 +0000 (10:25 -0700)]
test: check that unaligned load-add opcodes work.

A test for CL 102036.

Change-Id: Ief6dcb4f478670813fbe22ea75a06815a4b201a3
Reviewed-on: https://go-review.googlesource.com/105875
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: insert instrumentation during SSA building
Matthew Dempsky [Tue, 27 Mar 2018 20:50:08 +0000 (13:50 -0700)]
cmd/compile: insert instrumentation during SSA building

Insert appropriate race/msan calls before each memory operation during
SSA construction.

This is conceptually simple, but subtle because we need to be careful
that inserted instrumentation calls don't clobber arguments that are
currently being prepared for a user function call.

reorder1 already handles introducing temporary variables for arguments
in some cases. This CL changes it to use them for all arguments when
instrumenting.

Also, we can't SSA struct types with more than one field while
instrumenting. Otherwise, concurrent uses of disjoint fields within an
SSA-able struct can introduce false races.

This is both somewhat better and somewhat worse than the old racewalk
instrumentation pass. We're now able to easily recognize cases like
constructing non-escaping closures on the stack or accessing closure
variables don't need instrumentation calls. On the other hand,
spilling escaping parameters to the heap now results in an
instrumentation call.

Overall, this CL results in a small net reduction in the number of
instrumentation calls, but a small net increase in binary size for
instrumented executables. cmd/go ends up with 5.6% fewer calls, but a
2.4% larger binary.

Fixes #19054.

Change-Id: I70d1dd32ad6340e6fdb691e6d5a01452f58e97f3
Reviewed-on: https://go-review.googlesource.com/102817
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/internal/obj/arm64: add assembler support for load with register offset
fanzha02 [Fri, 8 Dec 2017 08:19:32 +0000 (08:19 +0000)]
cmd/internal/obj/arm64: add assembler support for load with register offset

The patch adds support for LDR(register offset) instruction.
And add the test cases and negative tests.

Change-Id: I5b32c6a5065afc4571116d4896f7ebec3c0416d3
Reviewed-on: https://go-review.googlesource.com/87955
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd: remove a few more unused parameters
Daniel Martí [Tue, 3 Apr 2018 14:52:30 +0000 (15:52 +0100)]
cmd: remove a few more unused parameters

ssa's pos parameter on the Const* funcs is unused, so remove it.

ld's alloc parameter on elfnote is always true, so remove the arguments
and simplify the code.

Finally, arm's addpltreloc never has its return parameter used, so
remove it.

Change-Id: I63387ecf6ab7b5f7c20df36be823322bb98427b8
Reviewed-on: https://go-review.googlesource.com/104456
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agotest/codegen: port various mem-combining tests
Alberto Donizetti [Mon, 9 Apr 2018 07:52:40 +0000 (09:52 +0200)]
test/codegen: port various mem-combining tests

And delete them from asm_test.

Change-Id: I0e33d58274951ab5acb67b0117b60ef617ea887a
Reviewed-on: https://go-review.googlesource.com/105735
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agonet/http: fix comment on LocalAddrContextKey
Ryan Slade [Thu, 5 Apr 2018 13:27:44 +0000 (13:27 +0000)]
net/http: fix comment on LocalAddrContextKey

Change-Id: Ibb793158a190758cebc00b58c12e3b8b106c0548
GitHub-Last-Rev: 6a6d949eb252ec6d3e76920fc2f50ef5ac3cdb5a
GitHub-Pull-Request: golang/go#24702
Reviewed-on: https://go-review.googlesource.com/104896
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agointernal/cpu: skip arm64 feature test on unsupported GOOS
Meng Zhuo [Sun, 8 Apr 2018 02:30:15 +0000 (10:30 +0800)]
internal/cpu: skip arm64 feature test on unsupported GOOS

Fixes #24753

Change-Id: Ic6049da98b8eee41223df71ffafa71a894915bd7
Reviewed-on: https://go-review.googlesource.com/105455
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: handle blank struct fields in NumComponents
Josh Bleecher Snyder [Fri, 30 Mar 2018 19:25:02 +0000 (12:25 -0700)]
cmd/compile: handle blank struct fields in NumComponents

NumComponents is used by racewalk to decide whether reads and writes
might occur to subobjects of an address. For that purpose,
blank fields matter.

It is also used to decide whether to inline == and != for a type.
For that purpose, blank fields may be ignored.

Add a parameter to NumComponents to support this distinction.
While we're here, document NumComponents, as requested in CL 59334.

Change-Id: I8c2021b172edadd6184848a32a74774dde1805c8
Reviewed-on: https://go-review.googlesource.com/103755
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/internal/obj/x86: use raw string literals in regexp
Alekseev Artem [Sat, 7 Apr 2018 14:10:45 +0000 (17:10 +0300)]
cmd/internal/obj/x86: use raw string literals in regexp

Found with megacheck (S1007).

Change-Id: Icb15fd5bfefa8e0b39a1bfa9ec3e9af3eff6b390
Reviewed-on: https://go-review.googlesource.com/105415
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotest/codegen: port arm64 slice zeroing tests
Alberto Donizetti [Fri, 6 Apr 2018 16:44:00 +0000 (18:44 +0200)]
test/codegen: port arm64 slice zeroing tests

Finish porting arm64 slice zeroing codegen tests; delete them from
asm_test.

Change-Id: Id2532df8ba1c340fa662a6b5238daa3de30548be
Reviewed-on: https://go-review.googlesource.com/105136
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
7 years agoruntime: improve Windows gdb tests
Egon Elbre [Sat, 24 Mar 2018 16:07:58 +0000 (18:07 +0200)]
runtime: improve Windows gdb tests

This ensures that gdb tests run on Windows by ignoring any line ending.

Works with gdb 7.7, however with gdb 7.9 and 7.12 gets an error
error:

    internal-error: buildsym_init: Assertion `free_pendings == NULL'
    failed.

Updates #21380

Change-Id: I6a6e5b2a1b5efdca4dfce009fcb9c134c87497d6
Reviewed-on: https://go-review.googlesource.com/102419
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
7 years agocmd/vendor/.../pprof: update to 0f7d9ba1
Daniel Martí [Fri, 6 Apr 2018 21:38:48 +0000 (22:38 +0100)]
cmd/vendor/.../pprof: update to 0f7d9ba1

In particular, to pull a few fixes that were causing some tests to be
flaky on our build dashboard.

Fixes #24405.
Fixes #24508.
Fixes #24611.

Change-Id: I713156ad11c924e4a4b603144d10395523d526ed
Reviewed-on: https://go-review.googlesource.com/105275
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: replace SOCKS client implementation
Mikio Hara [Tue, 18 Apr 2017 20:35:58 +0000 (05:35 +0900)]
net/http: replace SOCKS client implementation

This change replaces the vendored socks client implementation with the
bundle of golang.org/x/net/internal/socks package which contains fixes
for 19354 and 11682.

golang.org/x/net/internal/socks becomes socks_bundle.go.

At git rev 61147c4. (golang.org/cl/38278)

Updates #11682.
Updates #18508.
Updates #19354.
Fixes #19688.
Updates #21333.

Change-Id: I8cf6c3f5eb87c24685a7592be015729f84fbed77
Reviewed-on: https://go-review.googlesource.com/41031
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoos/user: add a way to enforce pure Go implementation
Kir Kolyshkin [Tue, 6 Feb 2018 21:16:02 +0000 (13:16 -0800)]
os/user: add a way to enforce pure Go implementation

This provides a way to enforce pure Go implementation of os/user
lookup functions on UNIX platforms by means of "osusergo" build tag,
in a manner similar to netgo/netcgo tags in the net package.

If "osusergo" build tag is set, Go implementation is selected.

If "osusergo" build tag is NOT set, the old behavior is retained,
that is to use cgo (libc-backed) implementation if both cgo and such
and such implementation are available.

The reason behind this change is to make it possible to build proper
static binaries on Linux. The problem is, glibc implementation of
getpw*, getgrp* and getgrouplist functions relies on presense of
libnss*.so libraries during runtime, making it impossible to build
a self-contained static binary which uses both cgo and os/user.
In such case, linker warnings like this are shown:

> warning: Using 'getgrouplist' in statically linked applications
> requires at runtime the shared libraries from the glibc version
> used for linking

While this can be solved by recompiling glibc with --enable-static-nss
flag or using a different libc implementation (like musl on Alpine Linux),
it is not always practical or even possible.

Fixes #23265

Change-Id: I383a448a2ecf15493ec93dbd5d076b6330cb14cb
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Reviewed-on: https://go-review.googlesource.com/92456
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agotime: simplify comparison in test
Veselkov Konstantin [Fri, 6 Apr 2018 18:54:29 +0000 (22:54 +0400)]
time: simplify comparison in test

Found using go-simple.

Change-Id: I349e80a8c083688539bb4267564f02e3d7913da5
Reviewed-on: https://go-review.googlesource.com/105195
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: change cmplx{mpy,div} into Mpcplx methods
Matthew Dempsky [Fri, 6 Apr 2018 16:44:29 +0000 (09:44 -0700)]
cmd/compile: change cmplx{mpy,div} into Mpcplx methods

Passes toolstash-check.

Change-Id: Icae55fe4fa1bb8e4f2f83b7c69e08d30a5559d9e
Reviewed-on: https://go-review.googlesource.com/105047
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: factor waiting on mark phase
Austin Clements [Mon, 26 Mar 2018 21:44:05 +0000 (17:44 -0400)]
runtime: factor waiting on mark phase

There are three places where we wait for the GC mark phase to
complete. Factor these all into a single helper function.

Fixes #24362.

Change-Id: I47f6a7147974f5b9a2869c527a024519070ba6f3
Reviewed-on: https://go-review.googlesource.com/102605
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agohtml/template: grow srcset buffer in proportion to need
Rhys Hiltner [Fri, 6 Apr 2018 18:06:03 +0000 (11:06 -0700)]
html/template: grow srcset buffer in proportion to need

In particular, avoid exponential memory usage from growing it in
proportion to its current size.

Fixes #24731

Change-Id: I277d2fbac2ef7b00ae4b83d6d1dcd7f2e630a5cd
Reviewed-on: https://go-review.googlesource.com/105155
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: distinguish semaphore wait from sync.Cond.Wait
Austin Clements [Mon, 26 Mar 2018 21:45:01 +0000 (17:45 -0400)]
runtime: distinguish semaphore wait from sync.Cond.Wait

Updates #24362.

Change-Id: Ided1ab31792f05d9d7a86f17c1bcbd9e9b80052c
Reviewed-on: https://go-review.googlesource.com/102606
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile/internal/gc: use bool in racewalk
Ilya Tocar [Thu, 5 Apr 2018 19:15:56 +0000 (14:15 -0500)]
cmd/compile/internal/gc: use bool in racewalk

Replace int variables with 0/1 as only possible values with bools,
where possible.

Change-Id: I958c082e703bbc1540309da3e17612fc8e247932
Reviewed-on: https://go-review.googlesource.com/105036
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: ensure first instruction of function is stmt
David Chase [Thu, 5 Apr 2018 15:11:39 +0000 (11:11 -0400)]
cmd/compile: ensure first instruction of function is stmt

In gdb, "b f" gets confused if the first instruction of "f"
is not marked as a statement in the DWARF line table.

To ensure gdb is not confused, move the first statement
marker in "f" to its first instruction.

The screwy-looking conditional for "what's the first
instruction with a statement marker" will become simpler in
the future.

Fixes #24695.

Change-Id: I2eef81676b64d1bd9bff5da03b89b9dc0c18f44f
Reviewed-on: https://go-review.googlesource.com/104955
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: cleanup method expression type checking
Matthew Dempsky [Fri, 6 Apr 2018 07:54:20 +0000 (00:54 -0700)]
cmd/compile: cleanup method expression type checking

Passes toolstash-check.

Change-Id: I804e73447b6fdbb75af6235c193c4ee7cbcf8d3a
Reviewed-on: https://go-review.googlesource.com/105045
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: fix method expressions with anonymous receivers
Matthew Dempsky [Fri, 6 Apr 2018 05:42:16 +0000 (22:42 -0700)]
cmd/compile: fix method expressions with anonymous receivers

Method expressions with anonymous receiver types like "struct { T }.m"
require wrapper functions, which we weren't always creating. This in
turn resulted in linker errors.

This CL ensures that we generate wrapper functions for any anonymous
receiver types used in a method expression.

Fixes #22444.

Change-Id: Ia8ac27f238c2898965e57b82a91d959792d2ddd4
Reviewed-on: https://go-review.googlesource.com/105044
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet: use same sendFile for freebsd/dragonfly and solaris
Tobias Klauser [Thu, 5 Apr 2018 10:20:12 +0000 (12:20 +0200)]
net: use same sendFile for freebsd/dragonfly and solaris

FreeBSD/Dragonfly and Solaris have identical implementations of
sendFile. Keep one and adjust the comments accordingly.

Change-Id: I77b0f88a4816dd6e40f5cb33919c44606401ac6b
Reviewed-on: https://go-review.googlesource.com/104915
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/rpc: unlock client.mutex as early as possible
chanxuehong [Fri, 6 Apr 2018 14:05:01 +0000 (14:05 +0000)]
net/rpc: unlock client.mutex as early as possible

Although these changes have no essential influence, I think this is a better point.

Change-Id: I571d3a14c948d2fd7bc9561f47f33e9e4c90683f
GitHub-Last-Rev: d8c5d180069bb91b6eb1247769bd14f728922fc0
GitHub-Pull-Request: golang/go#24697
Reviewed-on: https://go-review.googlesource.com/104895
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/asm/internal/arch: unexport ParseARM64Suffix
Tobias Klauser [Fri, 6 Apr 2018 12:40:18 +0000 (14:40 +0200)]
cmd/asm/internal/arch: unexport ParseARM64Suffix

ParseARM64Suffix is not used outside cmd/asm/internal/arch.

Change-Id: I8e7782dce11cf8cd2fd08dd17e555ced8d87ba24
Reviewed-on: https://go-review.googlesource.com/105115
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd: some semi-automated cleanups
Daniel Martí [Thu, 5 Apr 2018 20:28:06 +0000 (21:28 +0100)]
cmd: some semi-automated cleanups

* Remove some redundant returns
* Replace HasPrefix with TrimPrefix
* Remove some obviously dead code

Passes toolstash -cmp on std cmd.

Change-Id: Ifb0d70a45cbb8a8553758a8c4878598b7fe932bc
Reviewed-on: https://go-review.googlesource.com/105017
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: eliminate unused Sig.offset field
Matthew Dempsky [Fri, 6 Apr 2018 05:46:36 +0000 (22:46 -0700)]
cmd/compile: eliminate unused Sig.offset field

Change-Id: If498d1fc6e8c0c4e8cf7ed38c4997adf05e003a6
Reviewed-on: https://go-review.googlesource.com/105043
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
7 years agocmd/compile: rewrite a & 1 != 1 into a & 1 == 0 on amd64
Josh Bleecher Snyder [Thu, 5 Apr 2018 17:20:05 +0000 (10:20 -0700)]
cmd/compile: rewrite a & 1 != 1 into a & 1 == 0 on amd64

These rules trigger 190 times during make.bash.

Change-Id: I20d1688db5d8c904a7237c08635c6c9d8bd58b1c
Reviewed-on: https://go-review.googlesource.com/105037
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
7 years agomath/big: clean up z.div(z, x, y) calls
Brian Kessler [Thu, 30 Nov 2017 16:32:07 +0000 (09:32 -0700)]
math/big: clean up z.div(z, x, y) calls

Updates #22830

Due to not checking if the output slices alias in divLarge,
calls of the form z.div(z, x, y) caused the slice z
to attempt to be used to store both the quotient and the
remainder of the division.  CL 78995 applies an alias
check to correct that error.  This CL cleans up the
additional div calls that attempt to supply the same slice
to hold both the quotient and remainder.

Note that the call in expNN was responsible for the reported
error in r.Exp(x, 1, m) when r was initialized to a non-zero value.

The second instance in expNNMontgomery did not result in an error
due to the size of the arguments.

// RR = 2**(2*_W*len(m)) mod m
RR := nat(nil).setWord(1)
zz := nat(nil).shl(RR, uint(2*numWords*_W))
_, RR = RR.div(RR, zz, m)

Specifically,

cap(RR) == 5 after setWord(1) due to const e = 4 in z.make(1)
len(zz) == 2*len(m) + 1 after shifting left, numWords = len(m)

Reusing the backing array for z and z2 in div was only triggered if
cap(RR) >= len(zz) + 1 and len(m) > 1 so that divLarge was called.

But, 5 < 2*len(m) + 2 if len(m) > 1, so new arrays were allocated
and the error was never triggered in this case.

Change-Id: Iedac80dbbde13216c94659e84d28f6f4be3aaf24
Reviewed-on: https://go-review.googlesource.com/81055
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile: cleanup method symbol creation
Matthew Dempsky [Thu, 5 Apr 2018 01:42:39 +0000 (18:42 -0700)]
cmd/compile: cleanup method symbol creation

There were multiple ad hoc ways to create method symbols, with subtle
and confusing differences between them. This CL unifies them into a
single well-documented encoding and implementation.

This introduces some inconsequential changes to symbol format for the
sake of simplicity and consistency. Two notable changes:

1) Symbol construction is now insensitive to the package currently
being compiled. Previously, non-exported methods on anonymous types
received different method symbols depending on whether the method was
local or imported.

2) Symbols for method values parenthesized non-pointer receiver types
and non-exported method names, and also always package-qualified
non-exported method names. Now they use the same rules as normal
method symbols.

The methodSym function is also now stricter about rejecting
non-sensical method/receiver combinations. Notably, this means that
typecheckfunc needs to call addmethod to validate the method before
calling declare, which also means we no longer emit errors about
redeclaring bogus methods.

Change-Id: I9501c7a53dd70ef60e5c74603974e5ecc06e2003
Reviewed-on: https://go-review.googlesource.com/104876
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoruntime: avoid calling adjustpointers unnecessarily
Josh Bleecher Snyder [Wed, 4 Apr 2018 16:24:13 +0000 (09:24 -0700)]
runtime: avoid calling adjustpointers unnecessarily

adjustpointers loops over a bitmap.
If the length of that bitmap is zero,
we can skip making the call entirely.
This speeds up stack copying when there are
no pointers present in either args or locals.

name                old time/op  new time/op  delta
StackCopyPtr-8       101ms ± 4%    90ms ± 4%  -10.95%  (p=0.000 n=87+93)
StackCopy-8         80.1ms ± 4%  72.6ms ± 4%   -9.41%  (p=0.000 n=98+100)
StackCopyNoCache-8   121ms ± 3%   113ms ± 3%   -6.57%  (p=0.000 n=98+97)

Change-Id: I7a272e19bc9a14fa3e3318771ebd082dc6247d25
Reviewed-on: https://go-review.googlesource.com/104737
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/compile/internal/gc: factor out beginning of SSAGenState.Call
Richard Musiol [Wed, 28 Mar 2018 22:53:26 +0000 (00:53 +0200)]
cmd/compile/internal/gc: factor out beginning of SSAGenState.Call

This commit does not change the semantics of the Call method. Its
purpose is to avoid duplication of code by making PrepareCall available
for separate use by the wasm backend.

Updates #18892

Change-Id: I04a3098f56ebf0d995791c5375dd4c03b6a202a3
Reviewed-on: https://go-review.googlesource.com/103275
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agocmd/trace: include taskless spans in /usertasks.
Hana Kim [Thu, 5 Apr 2018 20:15:52 +0000 (16:15 -0400)]
cmd/trace: include taskless spans in /usertasks.

Change-Id: Id4e3407ba497a018d5ace92813ba8e9653d0ac7d
Reviewed-on: https://go-review.googlesource.com/104976
Reviewed-by: Heschi Kreinick <heschi@google.com>
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomath/big: improve performance on ppc64x by unrolling loops
Carlos Eduardo Seo [Wed, 28 Mar 2018 03:38:32 +0000 (23:38 -0400)]
math/big: improve performance on ppc64x by unrolling loops

This change improves performance of addVV, subVV and mulAddVWW
by unrolling the loops, with improvements up to 1.45x.

benchmark                    old ns/op     new ns/op     delta
BenchmarkAddVV/1-16          5.79          5.85          +1.04%
BenchmarkAddVV/2-16          6.41          6.62          +3.28%
BenchmarkAddVV/3-16          6.89          7.35          +6.68%
BenchmarkAddVV/4-16          7.47          8.26          +10.58%
BenchmarkAddVV/5-16          8.04          8.18          +1.74%
BenchmarkAddVV/10-16         10.9          11.2          +2.75%
BenchmarkAddVV/100-16        81.7          57.0          -30.23%
BenchmarkAddVV/1000-16       714           500           -29.97%
BenchmarkAddVV/10000-16      7088          4946          -30.22%
BenchmarkAddVV/100000-16     71514         49364         -30.97%
BenchmarkSubVV/1-16          5.94          5.89          -0.84%
BenchmarkSubVV/2-16          12.9          6.82          -47.13%
BenchmarkSubVV/3-16          7.03          7.34          +4.41%
BenchmarkSubVV/4-16          7.58          8.23          +8.58%
BenchmarkSubVV/5-16          8.15          8.19          +0.49%
BenchmarkSubVV/10-16         11.2          11.4          +1.79%
BenchmarkSubVV/100-16        82.4          57.0          -30.83%
BenchmarkSubVV/1000-16       715           499           -30.21%
BenchmarkSubVV/10000-16      7089          4947          -30.22%
BenchmarkSubVV/100000-16     71568         49378         -31.01%

benchmark                    old MB/s     new MB/s      speedup
BenchmarkAddVV/1-16          11048.49     10939.92      0.99x
BenchmarkAddVV/2-16          19973.41     19323.60      0.97x
BenchmarkAddVV/3-16          27847.09     26123.06      0.94x
BenchmarkAddVV/4-16          34276.46     30976.54      0.90x
BenchmarkAddVV/5-16          39781.92     39140.68      0.98x
BenchmarkAddVV/10-16         58559.29     56894.68      0.97x
BenchmarkAddVV/100-16        78354.88     112243.69     1.43x
BenchmarkAddVV/1000-16       89592.74     127889.04     1.43x
BenchmarkAddVV/10000-16      90292.39     129387.06     1.43x
BenchmarkAddVV/100000-16     89492.92     129647.78     1.45x
BenchmarkSubVV/1-16          10781.03     10861.22      1.01x
BenchmarkSubVV/2-16          9949.27      18760.21      1.89x
BenchmarkSubVV/3-16          27319.40     26166.01      0.96x
BenchmarkSubVV/4-16          33764.35     31123.02      0.92x
BenchmarkSubVV/5-16          39272.40     39050.31      0.99x
BenchmarkSubVV/10-16         57262.87     56206.33      0.98x
BenchmarkSubVV/100-16        77641.78     112280.86     1.45x
BenchmarkSubVV/1000-16       89486.27     128064.08     1.43x
BenchmarkSubVV/10000-16      90274.37     129356.59     1.43x
BenchmarkSubVV/100000-16     89424.42     129610.50     1.45x

Change-Id: I2795a82134d1e3b75e2634c76b8ca165a723ec7b
Reviewed-on: https://go-review.googlesource.com/103495
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
7 years agoruntime: fix/improve exitThread on openbsd
Joel Sing [Thu, 5 Apr 2018 19:00:25 +0000 (05:00 +1000)]
runtime: fix/improve exitThread on openbsd

OpenBSD's __threxit syscall takes a pointer to a 32-bit value that will be
zeroed immediately before the thread exits. Make use of this instead of
zeroing freeWait from the exitThread assembly and using hacks like switching
to a static stack, so this works on 386.

Change-Id: I3ec5ead82b6496404834d148f713794d5d9da723
Reviewed-on: https://go-review.googlesource.com/105055
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile/internal/ssa: fix GO386=387 build
Ilya Tocar [Thu, 5 Apr 2018 18:40:51 +0000 (13:40 -0500)]
cmd/compile/internal/ssa: fix GO386=387 build

Don't generate FP ops with 1 operand in memory for 387.

Change-Id: I23b49dfa2a1e60c8778c920230e64785a3ddfbd1
Reviewed-on: https://go-review.googlesource.com/105035
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: drop legacy code for generating iface wrappers
Matthew Dempsky [Thu, 5 Apr 2018 04:49:49 +0000 (21:49 -0700)]
cmd/compile: drop legacy code for generating iface wrappers

Originally, scalar values were directly stored within interface values
as long as they fit into a pointer-sized slot of memory. And since
interface method calls always pass the full pointer-sized value as the
receiver argument, value-narrowing wrappers were necessary to adapt to
the calling convention for methods with smaller receiver types.

However, for precise garbage collection, we now only store actual
pointers within interface values, so these wrappers are no longer
necessary.

Passes toolstash-check.

Change-Id: I5303bfeb8d0f11db619b5a5d06b37ac898588670
Reviewed-on: https://go-review.googlesource.com/104875
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/internal/obj: various code cleanups
Daniel Martí [Thu, 5 Apr 2018 11:11:32 +0000 (12:11 +0100)]
cmd/internal/obj: various code cleanups

Mostly replacing C-Style loops with range expressions, but also other
simplifications like the introduction of writeBool and unindenting some
code.

Passes toolstash -cmp on std cmd.

Change-Id: I799bccd4e5d411428dcf122b8588a564a9217e7c
Reviewed-on: https://go-review.googlesource.com/104936
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agocmd/internal/obj/x86: cleanup comments and consts
isharipo [Wed, 4 Apr 2018 21:26:38 +0000 (00:26 +0300)]
cmd/internal/obj/x86: cleanup comments and consts

- Unexport MaxLoopPad and LoopAlign; associated comments updated
- Remove commented-out C code
- Replace C-style /**/ code comments with single-line comments

Change-Id: I51bd92a05b4d3823757b12efd798951c9f252bd4
Reviewed-on: https://go-review.googlesource.com/104795
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/internal/obj/x86: remove unused VEX constants
isharipo [Wed, 4 Apr 2018 21:21:13 +0000 (00:21 +0300)]
cmd/internal/obj/x86: remove unused VEX constants

VEX constants were used when instructions were added by hand.
Now all VEX-encoded instructions are auto-generated by x86avxgen,
so there is no need for those anymore.

Change-Id: Ida63e5e23a8b819b15f61ac98980dec45a21617c
Reviewed-on: https://go-review.googlesource.com/104775
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: optimize 386 binary operations with a memory operand
Ben Shi [Thu, 22 Mar 2018 02:18:50 +0000 (02:18 +0000)]
cmd/compile: optimize 386 binary operations with a memory operand

Some integer/float binary operations of 386 can take a direct memory
operand, which is more efficient than loading to a register.

These CL does this optimization by copying the similar solution
of amd64. And the go1 benchmark shows some inprovements, especially
the test case Template. (excluding noise)

name                     old time/op    new time/op    delta
BinaryTree17-4              3.42s ± 2%     3.40s ± 2%    ~     (p=0.069 n=38+39)
Fannkuch11-4                3.48s ± 1%     3.53s ± 1%  +1.59%  (p=0.000 n=40+40)
FmtFprintfEmpty-4          46.7ns ± 4%    46.3ns ± 3%  -1.03%  (p=0.001 n=40+40)
FmtFprintfString-4         80.1ns ± 3%    80.6ns ± 3%  +0.56%  (p=0.029 n=40+40)
FmtFprintfInt-4            92.4ns ± 2%    92.3ns ± 3%    ~     (p=0.847 n=40+40)
FmtFprintfIntInt-4          147ns ± 3%     144ns ± 3%  -1.87%  (p=0.000 n=40+40)
FmtFprintfPrefixedInt-4     182ns ± 2%     184ns ± 3%  +0.99%  (p=0.002 n=40+40)
FmtFprintfFloat-4           387ns ± 3%     384ns ± 3%    ~     (p=0.069 n=40+40)
FmtManyArgs-4               619ns ± 3%     616ns ± 3%    ~     (p=0.320 n=40+40)
GobDecode-4                7.28ms ± 6%    7.27ms ± 5%    ~     (p=0.897 n=40+40)
GobEncode-4                7.33ms ± 6%    7.21ms ± 6%  -1.56%  (p=0.022 n=38+40)
Gzip-4                      357ms ± 4%     357ms ± 4%    ~     (p=0.071 n=40+40)
Gunzip-4                   45.3ms ± 3%    45.4ms ± 3%    ~     (p=0.452 n=40+40)
HTTPClientServer-4         63.0µs ± 2%    62.9µs ± 3%    ~     (p=0.760 n=38+39)
JSONEncode-4               22.0ms ± 4%    21.7ms ± 4%  -1.49%  (p=0.000 n=40+40)
JSONDecode-4               67.7ms ± 4%    68.3ms ± 3%  +0.86%  (p=0.039 n=40+40)
Mandelbrot200-4            5.16ms ± 3%    5.17ms ± 3%    ~     (p=0.418 n=40+40)
GoParse-4                  3.30ms ± 2%    3.32ms ± 3%  +0.55%  (p=0.017 n=40+40)
RegexpMatchEasy0_32-4       104ns ± 3%     104ns ± 4%    ~     (p=0.992 n=40+40)
RegexpMatchEasy0_1K-4       852ns ± 3%     851ns ± 2%    ~     (p=0.344 n=40+40)
RegexpMatchEasy1_32-4       113ns ± 4%     113ns ± 5%    ~     (p=0.937 n=40+40)
RegexpMatchEasy1_1K-4      1.03µs ± 5%    1.04µs ± 4%    ~     (p=0.430 n=40+40)
RegexpMatchMedium_32-4      132ns ± 4%     131ns ± 3%  -1.06%  (p=0.027 n=40+40)
RegexpMatchMedium_1K-4     43.0µs ± 3%    43.2µs ± 3%    ~     (p=0.122 n=40+40)
RegexpMatchHard_32-4       2.21µs ± 4%    2.20µs ± 4%    ~     (p=0.146 n=40+40)
RegexpMatchHard_1K-4       67.1µs ± 4%    67.2µs ± 3%    ~     (p=0.859 n=40+40)
Revcomp-4                   1.85s ± 2%     1.85s ± 3%    ~     (p=0.184 n=40+40)
Template-4                 70.1ms ± 4%    67.5ms ± 3%  -3.65%  (p=0.000 n=40+40)
TimeParse-4                 457ns ±16%     439ns ± 4%    ~     (p=0.683 n=40+34)
TimeFormat-4                413ns ± 3%     414ns ± 3%    ~     (p=0.850 n=40+40)
[Geo mean]                 67.5µs         67.3µs       -0.38%

name                     old speed      new speed      delta
GobDecode-4               105MB/s ± 6%   106MB/s ± 5%    ~     (p=0.893 n=40+40)
GobEncode-4               105MB/s ± 6%   107MB/s ± 7%  +1.60%  (p=0.023 n=38+40)
Gzip-4                   54.4MB/s ± 4%  54.5MB/s ± 4%    ~     (p=0.073 n=40+40)
Gunzip-4                  429MB/s ± 3%   428MB/s ± 3%    ~     (p=0.453 n=40+40)
JSONEncode-4             88.3MB/s ± 5%  89.6MB/s ± 4%  +1.51%  (p=0.000 n=40+40)
JSONDecode-4             28.7MB/s ± 4%  28.4MB/s ± 3%  -0.87%  (p=0.039 n=40+40)
GoParse-4                17.6MB/s ± 3%  17.5MB/s ± 3%  -0.55%  (p=0.020 n=40+40)
RegexpMatchEasy0_32-4     308MB/s ± 4%   308MB/s ± 5%    ~     (p=0.988 n=40+40)
RegexpMatchEasy0_1K-4    1.20GB/s ± 3%  1.20GB/s ± 2%    ~     (p=0.329 n=40+40)
RegexpMatchEasy1_32-4     283MB/s ± 4%   283MB/s ± 4%    ~     (p=0.507 n=40+40)
RegexpMatchEasy1_1K-4     991MB/s ± 5%   987MB/s ± 4%    ~     (p=0.446 n=40+40)
RegexpMatchMedium_32-4   7.54MB/s ± 4%  7.63MB/s ± 3%  +1.26%  (p=0.004 n=40+40)
RegexpMatchMedium_1K-4   23.8MB/s ± 3%  23.7MB/s ± 4%    ~     (p=0.121 n=40+40)
RegexpMatchHard_32-4     14.5MB/s ± 4%  14.6MB/s ± 4%    ~     (p=0.145 n=40+40)
RegexpMatchHard_1K-4     15.3MB/s ± 4%  15.2MB/s ± 3%    ~     (p=0.874 n=40+40)
Revcomp-4                 137MB/s ± 2%   137MB/s ± 3%    ~     (p=0.179 n=40+40)
Template-4               27.7MB/s ± 4%  28.7MB/s ± 3%  +3.78%  (p=0.000 n=40+40)
[Geo mean]               78.9MB/s       79.2MB/s       +0.38%

Change-Id: I3ba688c253b665485c1ebdf5a75f4ce82cc3def3
Reviewed-on: https://go-review.googlesource.com/102036
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/internal/obj/x86: make AsmBuf receiver name consistent
isharipo [Wed, 4 Apr 2018 21:31:34 +0000 (00:31 +0300)]
cmd/internal/obj/x86: make AsmBuf receiver name consistent

Fixes golint receiver name complaints.

We can't go with "a" name as it sometimes is used for obj.Addr args.

Change-Id: I66556f4e3dc42cfaaa4db3ed7772fa6756ea9a9b
Reviewed-on: https://go-review.googlesource.com/104796
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: early return/continue to unindent some code
Daniel Martí [Tue, 3 Apr 2018 12:17:28 +0000 (13:17 +0100)]
cmd/compile: early return/continue to unindent some code

While at it, also simplify a couple of switches.

Doesn't pass toolstash -cmp on std cmd, because orderBlock(&n2.Nbody) is
moved further down to the n3 loop.

Change-Id: I20a2a6c21eb9a183a59572e0fca401a5041fc40a
Reviewed-on: https://go-review.googlesource.com/104416
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agotest: skip locklinear's lockmany test for now
Daniel Martí [Tue, 3 Apr 2018 18:11:53 +0000 (19:11 +0100)]
test: skip locklinear's lockmany test for now

Since it's been reliably failing on one of the linux-arm builders
(arm5spacemonkey) for a long time.

Updates #24221.

Change-Id: I8fccc7e16631de497ccc2c285e510a110a93ad95
Reviewed-on: https://go-review.googlesource.com/104535
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: stack-allocate worklist in ReachableBlocks
Alberto Donizetti [Wed, 4 Apr 2018 12:35:05 +0000 (14:35 +0200)]
cmd/compile: stack-allocate worklist in ReachableBlocks

Stack-allocate a local worklist in the deadcode pass. A size of 64 for
the pre-allocation is enough for >99% of the ReachableBlocks call in
a typical package.

name      old time/op       new time/op       delta
Template        281ms ± 3%        278ms ± 2%  -1.03%  (p=0.049 n=20+20)
Unicode         135ms ± 6%        134ms ± 6%    ~     (p=0.273 n=18+17)
GoTypes         882ms ± 3%        880ms ± 2%    ~     (p=0.925 n=20+20)
Compiler        4.01s ± 1%        4.02s ± 2%    ~     (p=0.640 n=20+20)
SSA             9.61s ± 1%        9.75s ± 1%  +1.39%  (p=0.000 n=20+19)
Flate           186ms ± 5%        185ms ± 7%    ~     (p=0.758 n=20+20)
GoParser        219ms ± 5%        218ms ± 4%    ~     (p=0.149 n=20+20)
Reflect         568ms ± 4%        562ms ± 1%    ~     (p=0.154 n=19+19)
Tar             258ms ± 2%        257ms ± 3%    ~     (p=0.428 n=19+20)
XML             316ms ± 2%        317ms ± 3%    ~     (p=0.901 n=20+19)

name      old user-time/op  new user-time/op  delta
Template        398ms ± 6%        388ms ± 6%  -2.55%  (p=0.007 n=20+20)
Unicode         217ms ± 5%        213ms ± 6%  -1.90%  (p=0.036 n=17+20)
GoTypes         1.21s ± 3%        1.20s ± 3%  -0.89%  (p=0.022 n=19+20)
Compiler        5.56s ± 3%        5.53s ± 5%    ~     (p=0.779 n=20+20)
SSA             13.9s ± 5%        14.0s ± 4%    ~     (p=0.529 n=20+20)
Flate           248ms ±10%        252ms ± 4%    ~     (p=0.409 n=20+18)
GoParser        305ms ± 4%        299ms ± 5%  -1.87%  (p=0.007 n=19+20)
Reflect         754ms ± 2%        747ms ± 3%    ~     (p=0.107 n=20+19)
Tar             360ms ± 5%        362ms ± 3%    ~     (p=0.534 n=20+18)
XML             425ms ± 6%        429ms ± 4%    ~     (p=0.496 n=20+19)

name      old alloc/op      new alloc/op      delta
Template       38.8MB ± 0%       38.7MB ± 0%  -0.15%  (p=0.000 n=20+20)
Unicode        29.1MB ± 0%       29.1MB ± 0%  -0.03%  (p=0.000 n=20+20)
GoTypes         115MB ± 0%        115MB ± 0%  -0.13%  (p=0.000 n=20+20)
Compiler        491MB ± 0%        490MB ± 0%  -0.15%  (p=0.000 n=18+19)
SSA            1.40GB ± 0%       1.40GB ± 0%  -0.16%  (p=0.000 n=20+20)
Flate          24.9MB ± 0%       24.8MB ± 0%  -0.17%  (p=0.000 n=20+20)
GoParser       30.7MB ± 0%       30.6MB ± 0%  -0.16%  (p=0.000 n=20+20)
Reflect        77.1MB ± 0%       77.0MB ± 0%  -0.11%  (p=0.000 n=19+20)
Tar            39.0MB ± 0%       39.0MB ± 0%  -0.14%  (p=0.000 n=20+20)
XML            44.6MB ± 0%       44.5MB ± 0%  -0.13%  (p=0.000 n=17+19)

name      old allocs/op     new allocs/op     delta
Template         379k ± 0%         378k ± 0%  -0.45%  (p=0.000 n=20+17)
Unicode          336k ± 0%         336k ± 0%  -0.08%  (p=0.000 n=20+20)
GoTypes         1.18M ± 0%        1.17M ± 0%  -0.37%  (p=0.000 n=20+20)
Compiler        4.58M ± 0%        4.56M ± 0%  -0.38%  (p=0.000 n=20+20)
SSA             11.4M ± 0%        11.4M ± 0%  -0.39%  (p=0.000 n=20+20)
Flate            233k ± 0%         232k ± 0%  -0.51%  (p=0.000 n=20+20)
GoParser         313k ± 0%         312k ± 0%  -0.48%  (p=0.000 n=19+20)
Reflect          946k ± 0%         943k ± 0%  -0.31%  (p=0.000 n=20+20)
Tar              388k ± 0%         387k ± 0%  -0.40%  (p=0.000 n=20+20)
XML              411k ± 0%         409k ± 0%  -0.35%  (p=0.000 n=17+20)

Change-Id: Iaec0b9471ded61be5eb3c9d1074e804672307644
Reviewed-on: https://go-review.googlesource.com/104675
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: extract inline related fields into separate Inline type
Matthew Dempsky [Wed, 4 Apr 2018 22:53:27 +0000 (15:53 -0700)]
cmd/compile: extract inline related fields into separate Inline type

Inl, Inldcl, and InlCost are only applicable to functions with bodies
that can be inlined, so pull them out into a separate Inline type to
make understanding them easier.

A side benefit is that we can check if a function can be inlined by
just checking if n.Func.Inl is non-nil, which simplifies handling of
empty function bodies.

While here, remove some unnecessary Curfn twiddling, and make imported
functions use Inl.Dcl instead of Func.Dcl for consistency for local
functions.

Passes toolstash-check.

Change-Id: Ifd4a80349d85d9e8e4484952b38ec4a63182e81f
Reviewed-on: https://go-review.googlesource.com/104756
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agotext/tabwriter: remove internal use of bytes.Buffer (cleanup)
Robert Griesemer [Wed, 4 Apr 2018 23:04:22 +0000 (16:04 -0700)]
text/tabwriter: remove internal use of bytes.Buffer (cleanup)

Noticed that we can simply use a []byte slice while investigating
a separate issue. Did the obvious simplification.

Change-Id: I921ebbb42135b5f1a10109236ceb9ae6e94ae7e2
Reviewed-on: https://go-review.googlesource.com/104757
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: adjust is-statement on Pos's to improve debugging
David Chase [Sat, 3 Mar 2018 01:33:15 +0000 (20:33 -0500)]
cmd/compile: adjust is-statement on Pos's to improve debugging

Stores to auto tmp variables can be hoisted to places
where the line numbers make debugging look "jumpy".
Turning those instructions into ones with is_stmt = 0 in
the DWARF (accomplished by marking ssa nodes with NotStmt)
makes debugging look better while still attributing the
instructions with the correct line number.

The same is true for certain register allocator spills and
reloads.

Change-Id: I97a394eb522d4911cc40b4bf5bf76d3d7221f6c0
Reviewed-on: https://go-review.googlesource.com/98415
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/link: process is_stmt data into dwarf line tables
David Chase [Tue, 13 Feb 2018 22:39:38 +0000 (17:39 -0500)]
cmd/link: process is_stmt data into dwarf line tables

To improve debugging, instructions should be annotated with
DWARF is_stmt.  The DWARF default before was is_stmt=1, and
to remove "jumpy" stepping the optimizer was tagging
instructions with a no-position position, which interferes
with the accuracy of profiling information.  This allows
that to be corrected, and also allows more "jumpy" positions
to be annotated with is_stmt=0 (these changes were not made
for 1.10 because of worries about further messing up
profiling).

The is_stmt values are placed in a pc-encoded table and
passed through a symbol derived from the name of the
function and processed in the linker alongside its
processing of each function's pc/line tables.

The only change in binary size is in the .debug_line tables
measured with "objdump -h --section=.debug_line go1.test"
For go1.test, these are 2614 bytes larger,
or 0.72% of the size of .debug_line,
or 0.025% of the file size.

This will increase in proportion to how much the is_stmt
flag is used (toggled).

Change-Id: Ic1f1aeccff44591ad0494d29e1a0202a3c506a7a
Reviewed-on: https://go-review.googlesource.com/93664
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/compile: add IsStmt breakpoint info to src.lico
David Chase [Wed, 3 Jan 2018 22:14:55 +0000 (17:14 -0500)]
cmd/compile: add IsStmt breakpoint info to src.lico

Add IsStmt information to src.lico so that suitable lines
for breakpoints (or not) can be noted, eventually for
communication to the debugger via the linker and DWARF.

The expectation is that the front end will apply statement
boundary marks because it has best information about the
input, and the optimizer will attempt to preserve these.
The exact method for placing these marks is still TBD;
ideally stopping "at" line N in unoptimized code will occur
at a point where none of the side effects of N have occurred
and all of the inputs for line N can still be observed.
The optimizer will work with the same markings supplied
for unoptimized code.

It is a goal that non-optimizing compilation should conserve
statement marks.

The optimizer will also use the not-a-statement annotation
to indicate instructions that have a line number (for
profiling purposes) but should not be the target of
debugger step, next, or breakpoints.  Because instructions
marked as statements are sometimes removed, a third value
indicating that a position (instruction) can serve as a
statement if the optimizer removes the current instruction
marked as a statement for the same line.  The optimizer
should attempt to conserve statement marks, but it is not
a bug if some are lost.

Includes changes to html output for GOSSAFUNC to indicate
not-default is-a-statement with bold and not-a-statement
with strikethrough.

Change-Id: Ia22c9a682f276e2ca2a4ef7a85d4b6ebf9c62b7f
Reviewed-on: https://go-review.googlesource.com/93663
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agogo/printer, gofmt: tuned table alignment for better results
Robert Griesemer [Wed, 4 Apr 2018 00:05:47 +0000 (17:05 -0700)]
go/printer, gofmt: tuned table alignment for better results

The go/printer (and thus gofmt) uses a heuristic to determine
whether to break alignment between elements of an expression
list which is spread across multiple lines. The heuristic only
kicked in if the entry sizes (character length) was above a
certain threshold (20) and the ratio between the previous and
current entry size was above a certain value (4).

This heuristic worked reasonably most of the time, but also
led to unfortunate breaks in many cases where a single entry
was suddenly much smaller (or larger) then the previous one.

The behavior of gofmt was sufficiently mysterious in some of
these situations that many issues were filed against it.

The simplest solution to address this problem is to remove
the heuristic altogether and have a programmer introduce
empty lines to force different alignments if it improves
readability. The problem with that approach is that the
places where it really matters, very long tables with many
(hundreds, or more) entries, may be machine-generated and
not "post-processed" by a human (e.g., unicode/utf8/tables.go).

If a single one of those entries is overlong, the result
would be that the alignment would force all comments or
values in key:value pairs to be adjusted to that overlong
value, making the table hard to read (e.g., that entry may
not even be visible on screen and all other entries seem
spaced out too wide).

Instead, we opted for a slightly improved heuristic that
behaves much better for "normal", human-written code.

1) The threshold is increased from 20 to 40. This disables
the heuristic for many common cases yet even if the alignment
is not "ideal", 40 is not that many characters per line with
todays screens, making it very likely that the entire line
remains "visible" in an editor.

2) Changed the heuristic to not simply look at the size ratio
between current and previous line, but instead considering the
geometric mean of the sizes of the previous (aligned) lines.
This emphasizes the "overall picture" of the previous lines,
rather than a single one (which might be an outlier).

3) Changed the ratio from 4 to 2.5. Now that we ignore sizes
below 40, a ratio of 4 would mean that a new entry would have
to be 4 times bigger (160) or smaller (10) before alignment
would be broken. A ratio of 2.5 seems more sensible.

Applied updated gofmt to all of src and misc. Also tested
against several former issues that complained about this
and verified that the output for the given examples is
satisfactory (added respective test cases).

Some of the files changed because they were not gofmt-ed
in the first place.

For #644.
For #7335.
For #10392.
(and probably more related issues)

Fixes #22852.

Change-Id: I5e48b3d3b157a5cf2d649833b7297b33f43a6f6e

7 years agocmd/compile: intrinsify math/big.mulWW on ARM64
Balaram Makam [Mon, 2 Apr 2018 20:22:08 +0000 (16:22 -0400)]
cmd/compile: intrinsify math/big.mulWW on ARM64

Performance numbers on amberwing:

pkg: math/big
name                            old time/op    new time/op    delta
QuoRem                            3.08µs ± 0%    2.93µs ± 1%   -4.89%  (p=0.008 n=5+5)
ModSqrt225_Tonelli                 721µs ± 0%     718µs ± 0%   -0.46%  (p=0.008 n=5+5)
ModSqrt224_3Mod4                   218µs ± 0%     217µs ± 0%   -0.27%  (p=0.008 n=5+5)
ModSqrt5430_Tonelli                2.91s ± 0%     2.91s ± 0%     ~     (p=0.222 n=5+5)
ModSqrt5430_3Mod4                  970ms ± 0%     970ms ± 0%     ~     (p=0.151 n=5+5)
Sqrt                              45.9µs ± 0%    43.8µs ± 0%   -4.63%  (p=0.008 n=5+5)
IntSqr/1                          19.9ns ± 0%    17.3ns ± 0%  -13.07%  (p=0.008 n=5+5)
IntSqr/2                          52.6ns ± 0%    50.8ns ± 0%   -3.35%  (p=0.008 n=5+5)
IntSqr/3                          70.4ns ± 0%    69.4ns ± 0%     ~     (p=0.079 n=4+5)
IntSqr/5                           103ns ± 0%      99ns ± 0%   -3.98%  (p=0.008 n=5+5)
IntSqr/8                           179ns ± 0%     178ns ± 0%   -0.56%  (p=0.008 n=5+5)
IntSqr/10                          272ns ± 0%     272ns ± 0%     ~     (all equal)
IntSqr/20                          763ns ± 0%     787ns ± 0%   +3.15%  (p=0.016 n=5+4)
IntSqr/30                         1.25µs ± 1%    1.29µs ± 1%   +3.27%  (p=0.008 n=5+5)
IntSqr/50                         2.64µs ± 0%    2.71µs ± 0%   +2.61%  (p=0.008 n=5+5)
IntSqr/80                         5.67µs ± 0%    5.72µs ± 0%   +0.88%  (p=0.008 n=5+5)
IntSqr/100                        8.05µs ± 0%    8.09µs ± 0%   +0.45%  (p=0.008 n=5+5)
IntSqr/200                        28.0µs ± 0%    28.1µs ± 0%     ~     (p=0.151 n=5+5)
IntSqr/300                        59.4µs ± 0%    59.6µs ± 0%   +0.36%  (p=0.008 n=5+5)
IntSqr/500                         141µs ± 0%     141µs ± 0%   +0.08%  (p=0.008 n=5+5)
IntSqr/800                         280µs ± 0%     280µs ± 0%   -0.12%  (p=0.008 n=5+5)
IntSqr/1000                        429µs ± 0%     428µs ± 0%   -0.27%  (p=0.008 n=5+5)

pkg: crypto-ecdsa
name      old time/op    new time/op    delta
SignP384    7.85ms ± 1%    7.61ms ± 1%  -3.12%  (p=0.008 n=5+5)

Change-Id: I1ab30856cc0e570f6312f0bd8914779b55adbc16
Reviewed-on: https://go-review.googlesource.com/104135
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>