]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agoruntime: offset the heap arena index by 2^47 on amd64
Austin Clements [Mon, 19 Feb 2018 21:10:58 +0000 (16:10 -0500)]
runtime: offset the heap arena index by 2^47 on amd64

On amd64, the virtual address space, when interpreted as signed
values, is [-2^47, 2^47). Currently, we only support heap addresses in
the "positive" half of this, [0, 2^47). This suffices for linux/amd64
and windows/amd64, but solaris/amd64 can map user addresses in the
negative part of this range. Specifically, addresses
0xFFFF8000'00000000 to 0xFFFFFD80'00000000 are part of user space.
This leads to "memory allocated by OS not in usable address space"
panic, since we don't map heap arena index space for these addresses.

Fix this by offsetting addresses when computing arena indexes so that
arena entry 0 corresponds to address -2^47 on amd64. We already map
enough arena space for 2^48 heap addresses on 64-bit (because arm64's
virtual address space is [0, 2^48)), so we don't need to grow any
structures to support this.

A different approach would be to simply mask out the top 16 bits.
However, there are two advantages to the offset approach: 1) invalid
heap addresses continue to naturally map to invalid arena indexes so
we don't need extra checks and 2) it perturbs the mapping of addresses
to arena indexes more, which helps check that we don't accidentally
compute incorrect arena indexes somewhere that happen to be right most
of the time.

Several comments and constant names are now somewhat misleading. We'll
fix that in the next CL. This CL is the core change the arena
indexing.

Fixes #23862.

Change-Id: Idb8e299fded04593a286b01a9582da6ddbac2f9a
Reviewed-on: https://go-review.googlesource.com/95497
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: abstract indexing of arena index
Austin Clements [Fri, 16 Feb 2018 22:53:16 +0000 (17:53 -0500)]
runtime: abstract indexing of arena index

Accessing the arena index is about to get slightly more complicated.
Abstract this away into a set of functions for going back and forth
between addresses and arena slice indexes.

For #23862.

Change-Id: I0b20e74ef47a07b78ed0cf0a6128afe6f6e40f4b
Reviewed-on: https://go-review.googlesource.com/95496
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: simplify bulkBarrierPreWrite
Austin Clements [Fri, 16 Feb 2018 22:45:21 +0000 (17:45 -0500)]
runtime: simplify bulkBarrierPreWrite

Currently, bulkBarrierPreWrite uses inheap to decide whether the
destination is in the heap or whether to check for stack or global
data. However, this isn't the best question to ask.

Instead, get the span directly and query its state. This lets us
directly determine whether this might be a global, or is stack memory,
or is heap memory.

At this point, inheap is no longer used in the hot path, so drop it
from the must-be-inlined list and substitute spanOf.

This will help in a circuitous way with #23862, since fixing that is
going to push inheap very slightly over the inline-able threshold on a
few platforms.

Change-Id: I5360fc1181183598502409f12979899e1e4d45f7
Reviewed-on: https://go-review.googlesource.com/95495
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agocmd/trace: include P info in goroutine slices
Hana Kim [Fri, 26 Jan 2018 15:28:10 +0000 (10:28 -0500)]
cmd/trace: include P info in goroutine slices

The task-oriented trace view presents the execution trace organized
based on goroutines. Often, which P a goroutine was running on is
useful, so this CL includes the P ids in the goroutine execution slices.

R=go1.11

Change-Id: I96539bf8215e5c1cd8cc997a90204f57347c48c8
Reviewed-on: https://go-review.googlesource.com/90221
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/trace: add user log event in the task-oriented trace view
Hana Kim [Fri, 26 Jan 2018 15:18:16 +0000 (10:18 -0500)]
cmd/trace: add user log event in the task-oriented trace view

Also append stack traces to task create/end slices.

R=go1.11

Change-Id: I2adb342e92b36d30bee2860393618eb4064450cf
Reviewed-on: https://go-review.googlesource.com/90220
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/trace: present the GC time in the usertask view
Hana Kim [Thu, 25 Jan 2018 20:38:09 +0000 (15:38 -0500)]
cmd/trace: present the GC time in the usertask view

The GC time for a task is defined by the sum of GC duration
overlapping with the task's duration.

Also, grey out non-overlapping slices in the task-oriented
trace view.

R=go1.11

Change-Id: I42def0eb520f5d9bd07edd265e558706f6fab552
Reviewed-on: https://go-review.googlesource.com/90219
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/compile/internal: reuse more memory
Heschi Kreinick [Mon, 5 Feb 2018 22:04:44 +0000 (17:04 -0500)]
cmd/compile/internal: reuse more memory

Reuse even more memory, and keep track of it in a long-lived debugState
object rather than piecemeal in the Cache.

Change-Id: Ib6936b4e8594dc6dda1f59ece753c00fd1c136ba
Reviewed-on: https://go-review.googlesource.com/92404
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile/internal/ssa: refactor buildLocationLists
Heschi Kreinick [Mon, 5 Feb 2018 21:55:54 +0000 (16:55 -0500)]
cmd/compile/internal/ssa: refactor buildLocationLists

Change the closures to methods on debugState, mostly just for aesthetic
reasons.

Change-Id: I5242807f7300efafc7efb4eb3bd305ac3ec8e826
Reviewed-on: https://go-review.googlesource.com/92403
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile/internal: use sparseSet, optimize isSynthetic
Heschi Kreinick [Fri, 2 Feb 2018 21:26:58 +0000 (16:26 -0500)]
cmd/compile/internal: use sparseSet, optimize isSynthetic

changedVars was functionally a set, but couldn't be iterated over
efficiently. In functions with many variables, the wasted iteration was
costly. Use a sparseSet instead.

(*gc.Node).String() is very expensive: it calls Sprintf, which does
reflection, etc, etc. Instead, just look at .Sym.Name, which is all we
care about.

Change-Id: Ib61cd7b5c796e1813b8859135e85da5bfe2ac686
Reviewed-on: https://go-review.googlesource.com/92402
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile/internal/ssa: shrink commonly-used structs
Heschi Kreinick [Wed, 31 Jan 2018 22:56:14 +0000 (17:56 -0500)]
cmd/compile/internal/ssa: shrink commonly-used structs

Replace the OnStack boolean in VarLoc with a flag bit in StackOffset.
This doesn't get much memory savings since it's still 64-bit aligned,
but does seem to help a bit anyway.

Change liveSlot to fit into 16 bytes. Because nested structs still get
padding, this required inlining it. Fortunately there's not much logic
to copy.

Change-Id: Ie19a409daa41aa310275c4517a021eecf8886441
Reviewed-on: https://go-review.googlesource.com/92401
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile: use | in the most repetitive ppc64 rules
Alberto Donizetti [Wed, 21 Feb 2018 17:19:52 +0000 (18:19 +0100)]
cmd/compile: use | in the most repetitive ppc64 rules

For now, limited to the most repetitive rules that are also short and
simple, so that we can have a substantial conciseness win without
compromising rules readability.

Ran rulegen, no changes in the rewrite files.

Change-Id: I8d8cc67d02faca4756cc02402b763f1645ee31de
Reviewed-on: https://go-review.googlesource.com/95935
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: intrinsify math.Sqrt on mips64
Alberto Donizetti [Wed, 21 Feb 2018 16:48:33 +0000 (17:48 +0100)]
cmd/compile: intrinsify math.Sqrt on mips64

Fixes #24006

Change-Id: Ic1438b121fe705f9a6e3ed8340882e9dfd26ecf7
Reviewed-on: https://go-review.googlesource.com/95916
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>

7 years agocmd/compile: use | in the most repetitive mips64 rules
Alberto Donizetti [Wed, 21 Feb 2018 11:41:34 +0000 (12:41 +0100)]
cmd/compile: use | in the most repetitive mips64 rules

For now, limited to the most repetitive rules that are also short and
simple, so that we can have a substantial conciseness win without
compromising rules readability.

Ran rulegen, no change in the actual compiler code (as expected).

Change-Id: Ia74acc389cd8310eb7fe8f927171fa3d292d2a86
Reviewed-on: https://go-review.googlesource.com/95797
Reviewed-by: Giovanni Bajo <rasky@develer.com>
7 years agocmd/compile: use | in the most repetitive mips rules
Alberto Donizetti [Wed, 21 Feb 2018 10:54:55 +0000 (11:54 +0100)]
cmd/compile: use | in the most repetitive mips rules

For now, limited to the most repetitive rules that are also short and
simple, so that we can have a substantial conciseness win without
compromising rules readability.

Ran rulegen, no change in the actual compiler code (as expected).

Change-Id: Ib0bfbbc181fcec095fb78ac752addd1eee0c3575
Reviewed-on: https://go-review.googlesource.com/95796
Reviewed-by: Giovanni Bajo <rasky@develer.com>
7 years agocmd/compile: aggregate some rules in AMD64.rules
Giovanni Bajo [Wed, 21 Feb 2018 10:36:43 +0000 (11:36 +0100)]
cmd/compile: aggregate some rules in AMD64.rules

No changes in the generated file, as expected.

Change-Id: I30e0404612cd150f1455378b8db1c18b1e12d34e
Reviewed-on: https://go-review.googlesource.com/95616
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/vet: warn on unkeyed struct pointer literals
Daniel Martí [Thu, 25 Jan 2018 03:32:23 +0000 (10:32 +0700)]
cmd/vet: warn on unkeyed struct pointer literals

We did warn on them in some cases, but not others. In particular, if one
used a slice composite literal with struct pointer elements, and omitted
the type of an element's composite literal, it would not get any warning
even if it should get one.

The issue is that typ.Underlying() can be of type *types.Pointer. Skip
those levels of indirection before checking for a *types.Struct
underlying type.

isLocalType also needed a bit of tweaking to ignore dereferences.
Perhaps that can be rewritten now that we have type info, but let's
leave it for another time.

Fixes #23539.

Change-Id: I727a497284df1325b70d47a756519f5db1add25d
Reviewed-on: https://go-review.googlesource.com/89715
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: regenerate rewrite rules for amd64
Giovanni Bajo [Wed, 21 Feb 2018 10:31:47 +0000 (11:31 +0100)]
cmd/compile: regenerate rewrite rules for amd64

Sometimes, multiple CLs being merged that create rules on the same
opcodes can cause the generated file to differ compared to a new
regeneration. This is caused by the fact that rulegen splits
generated functions in chunks of 10 rules per function (to avoid
creating functions that are too big). If two CLs add rules to
the same function, they might cause a generated function to
have more than 10 rules, even though each CL individually didn't
pass this limit.

Change-Id: Ib641396b7e9028f80ec8718746969d390a9fbba9
Reviewed-on: https://go-review.googlesource.com/95795
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet: fix UDPConn readers to return truncated payload size instead of 0
Mansour Rahimi [Wed, 7 Feb 2018 01:17:50 +0000 (02:17 +0100)]
net: fix UDPConn readers to return truncated payload size instead of 0

Calling UDPConn readers (Read, ReadFrom, ReadMsgUDP) to read part of
datagram returns error (in Windows), mentioning there is more data
available, and 0 as size of read data, even though part of data is
already read.

This fix makes UDPConn readers to return truncated payload size,
even there is error due more data available to read.

Fixes #14074
Updates #18056

Change-Id: Id7eec7f544dd759b2d970fa2561eef2937ec4662
Reviewed-on: https://go-review.googlesource.com/92475
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
7 years agocmd/compile: use | in the most repetitive arm rules
Alberto Donizetti [Tue, 20 Feb 2018 20:27:28 +0000 (21:27 +0100)]
cmd/compile: use | in the most repetitive arm rules

For now, limited to the most repetitive rules that are also short and
simple, so that we can have a substantial conciseness win without
compromising rules readability.

Ran rulegen, no change in the actual compiler code (as expected).

Change-Id: Ib1d2b9fbc787379105ec9baf10d2c1e2ff3c4c5c
Reviewed-on: https://go-review.googlesource.com/95615
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoA+C: update my name and spelling
Ahmed W [Fri, 16 Feb 2018 18:33:15 +0000 (18:33 +0000)]
A+C: update my name and spelling

GitHub-Last-Rev: ff68319f4c46271ddb927d176756962a3a2b4332
GitHub-Pull-Request: golang/go#23874
Change-Id: I7242c5cc35f04e23d807f5e91180c4ef510e7d1a
Reviewed-on: https://go-review.googlesource.com/94840
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: ensure sysStat for mheap_.arenas is aligned
Austin Clements [Tue, 20 Feb 2018 23:16:56 +0000 (18:16 -0500)]
runtime: ensure sysStat for mheap_.arenas is aligned

We don't want to account the memory for mheap_.arenas because most of
it is never touched, so currently we pass the address of a uint64 on
the heap. However, at least on mips, it's possible for this uint64 to
be unaligned, which causes the atomic add in mSysStatInc to crash.

Fix this by instead passing a nil stat pointer.

Fixes #23946.

Change-Id: I091587df1b3066c330b6bb4d834e4596c407910f
Reviewed-on: https://go-review.googlesource.com/95695
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agogithub: update Pull Request template
Andrew Bonventre [Wed, 21 Feb 2018 00:14:00 +0000 (19:14 -0500)]
github: update Pull Request template

+ Move from Markdown checklist to text. The first PR comment is
  presented as text when creating it.
+ Add the note about Signed-Off-By: not being required.

Change-Id: I0650891dcf11ed7dd367007148730ba2917784fe
Reviewed-on: https://go-review.googlesource.com/95696
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoarchive/zip: make benchmarks more representative
Ilya Tocar [Wed, 13 Dec 2017 21:25:50 +0000 (15:25 -0600)]
archive/zip: make benchmarks more representative

Currently zip benchmarks spend 60% in the rleBuffer code,
which is used only to test zip archive/zip itself:
    17.48s 37.02% 37.02%     18.12s 38.37%  archive/zip.(*rleBuffer).ReadAt
     9.51s 20.14% 57.16%     10.43s 22.09%  archive/zip.(*rleBuffer).Write
     9.15s 19.38% 76.54%     10.85s 22.98%  compress/flate.(*compressor).deflate

This means that benchmarks currently test performance of test helper.
Updating ReadAt/Write methods to be more performant makes benchmarks closer to real world.

name                       old time/op    new time/op    delta
CompressedZipGarbage-8       2.34ms ± 0%    2.34ms ± 1%     ~     (p=0.684 n=10+10)
Zip64Test-8                  58.1ms ± 2%    10.7ms ± 1%  -81.54%  (p=0.000 n=10+10)
Zip64TestSizes/4096-8        4.05µs ± 2%    3.65µs ± 5%   -9.96%  (p=0.000 n=9+10)
Zip64TestSizes/1048576-8      238µs ± 0%      43µs ± 0%  -82.06%  (p=0.000 n=10+10)
Zip64TestSizes/67108864-8    15.3ms ± 1%     2.6ms ± 0%  -83.12%  (p=0.000 n=10+9)

name                       old alloc/op   new alloc/op   delta
CompressedZipGarbage-8       17.9kB ±14%    16.0kB ±24%  -10.48%  (p=0.026 n=9+10)

name                       old allocs/op  new allocs/op  delta
CompressedZipGarbage-8         44.0 ± 0%      44.0 ± 0%     ~     (all equal)

Change-Id: Idfd920d0e4bed4aec2f5be84dc7e3919d9f1dd2d
Reviewed-on: https://go-review.googlesource.com/83857
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: shorten reflect.unsafe_New call chain
Martin Möhrmann [Sun, 28 Jan 2018 18:46:57 +0000 (19:46 +0100)]
runtime: shorten reflect.unsafe_New call chain

reflect.unsafe_New is an often called function according
to profiling in a large production environment.

Since newobject is not inlined currently there
is call overhead that can be avoided by calling
mallocgc directly.

name  old time/op  new time/op  delta
New   32.4ns ± 2%  29.8ns ± 1%  -8.03%  (p=0.000 n=19+20)

Change-Id: I572e4be830ed8e5c0da555dc3a8864c8363112be
Reviewed-on: https://go-review.googlesource.com/95015
Reviewed-by: Austin Clements <austin@google.com>
7 years agocrypto/sha512: speed-up for very small blocks
Ilya Tocar [Fri, 8 Dec 2017 20:59:55 +0000 (14:59 -0600)]
crypto/sha512: speed-up for very small blocks

Similar to https://golang.org/cl/54391, but for sha512
name          old time/op    new time/op    delta
Hash8Bytes-8     289ns ± 1%     253ns ± 2%  -12.59%  (p=0.000 n=10+10)
Hash1K-8        1.85µs ± 1%    1.82µs ± 1%   -1.77%  (p=0.000 n=9+10)
Hash8K-8        12.7µs ± 2%    12.5µs ± 1%     ~     (p=0.075 n=10+10)

name          old speed      new speed      delta
Hash8Bytes-8  27.6MB/s ± 1%  31.6MB/s ± 2%  +14.43%  (p=0.000 n=10+10)
Hash1K-8       554MB/s ± 1%   564MB/s ± 1%   +1.81%  (p=0.000 n=9+10)
Hash8K-8       647MB/s ± 2%   653MB/s ± 1%     ~     (p=0.075 n=10+10)

Change-Id: I437668c96ad55f8dbb62c89c8fc3f433453b5330
Reviewed-on: https://go-review.googlesource.com/82996
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <hi@filippo.io>
7 years agocrypto/sha256: speed-up for very small blocks
Ilya Tocar [Fri, 8 Dec 2017 19:28:07 +0000 (13:28 -0600)]
crypto/sha256: speed-up for very small blocks

Similar to https://golang.org/cl/54391, but for sha256
name          old time/op    new time/op    delta
Hash8Bytes-8     209ns ± 1%     191ns ± 1%  -8.65%  (p=0.000 n=10+9)
Hash1K-8        2.49µs ± 1%    2.47µs ± 2%  -0.74%  (p=0.045 n=9+10)
Hash8K-8        18.4µs ± 1%    18.2µs ± 0%  -0.98%  (p=0.009 n=10+10)

name          old speed      new speed      delta
Hash8Bytes-8  38.1MB/s ± 1%  41.8MB/s ± 1%  +9.47%  (p=0.000 n=10+9)
Hash1K-8       412MB/s ± 1%   415MB/s ± 2%    ~     (p=0.051 n=9+10)
Hash8K-8       445MB/s ± 1%   450MB/s ± 0%  +0.98%  (p=0.009 n=10+10)

Change-Id: I50ca80fc28c279fbb758b7c849f67d8c66391eb6
Reviewed-on: https://go-review.googlesource.com/82995
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <hi@filippo.io>
7 years agocmd/compile/internal/ssa: only store relevant slots in pendingEntries
Heschi Kreinick [Wed, 31 Jan 2018 21:30:16 +0000 (16:30 -0500)]
cmd/compile/internal/ssa: only store relevant slots in pendingEntries

For functions with many local variables, keeping track of every
LocalSlot for every variable is very expensive. Only track the slots
that are actually used by a given variable.

Change-Id: Iaafbce030a782b8b8c4a0eb7cf025e59af899ea4
Reviewed-on: https://go-review.googlesource.com/92400
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile/internal/ssa: don't store block start states
Heschi Kreinick [Wed, 31 Jan 2018 20:08:05 +0000 (15:08 -0500)]
cmd/compile/internal/ssa: don't store block start states

Keeping the start state of each block around costs more than just
recomputing them as necessary, especially because many blocks only have
one predecessor and don't need any merging at all. Stop storing the
start state, and reuse predecessors' end states as much as conveniently
possible.

Change-Id: I549bad9e1a35af76a974e46fe69f74cd4dce873b
Reviewed-on: https://go-review.googlesource.com/92399
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile: fold LEAQ/ADDQconst into SETx ops
Giovanni Bajo [Sun, 18 Feb 2018 19:02:17 +0000 (20:02 +0100)]
cmd/compile: fold LEAQ/ADDQconst into SETx ops

This saves an instruction and a register. The new rules
match ~4900 times during all.bash.

Change-Id: I2f867c5e70262004e31f545f3bb89e939c45b718
Reviewed-on: https://go-review.googlesource.com/94767
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoall: fix misspellings
Shawn Smith [Tue, 20 Feb 2018 20:50:20 +0000 (20:50 +0000)]
all: fix misspellings

GitHub-Last-Rev: 468df242d07419c228656985702325aa78952d99
GitHub-Pull-Request: golang/go#23935
Change-Id: If751ce3ffa3a4d5e00a3138211383d12cb6b23fc
Reviewed-on: https://go-review.googlesource.com/95577
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
7 years agocmd/compile: use | in the most repetitive 386 rules
Alberto Donizetti [Tue, 20 Feb 2018 20:04:56 +0000 (21:04 +0100)]
cmd/compile: use | in the most repetitive 386 rules

For now, limited to the most repetitive rules that are also short and
simple, so that we can have a substantial conciseness win without
compromising rules readability.

Ran rulegen, no change in the actual compiler code (as expected).

Change-Id: Ibf157382fb4544c063fbc80406fb9302430728fe
Reviewed-on: https://go-review.googlesource.com/95595
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: use | in the most repetitive boolean rules
Alberto Donizetti [Tue, 20 Feb 2018 19:18:16 +0000 (20:18 +0100)]
cmd/compile: use | in the most repetitive boolean rules

For now, limited to a few repetitive boolean rules where the win is
substantial (4+ variants).

Change-Id: I67bce0d356ca7d71a0f15ff98551fe2caff8abf9
Reviewed-on: https://go-review.googlesource.com/95535
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: aggregate rules that fold LEA/ADD into MOVx ops
Giovanni Bajo [Sun, 18 Feb 2018 18:32:26 +0000 (19:32 +0100)]
cmd/compile: aggregate rules that fold LEA/ADD into MOVx ops

No functional changes.

Change-Id: I4a3642d6dedf602a62f5a69cb630d35965ad6b98
Reviewed-on: https://go-review.googlesource.com/94763
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: aggregate bit-test rules
Giovanni Bajo [Fri, 16 Feb 2018 23:10:50 +0000 (00:10 +0100)]
cmd/compile: aggregate bit-test rules

No functional changes.

Change-Id: I4ea186b09a0309dfa1a80ff71208af2223997ffe
Reviewed-on: https://go-review.googlesource.com/94762
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/trace: task-oriented view includes child tasks
Hana Kim [Thu, 25 Jan 2018 20:13:12 +0000 (15:13 -0500)]
cmd/trace: task-oriented view includes child tasks

R=go1.11

Change-Id: Ibb09e309c745eba811a0b53000c063bc10a055e1
Reviewed-on: https://go-review.googlesource.com/90218
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Peter Weinberger <pjw@google.com>
7 years agocmd/trace: extend trace view (/trace) for task-oriented view
Hana Kim [Wed, 24 Jan 2018 22:48:28 +0000 (17:48 -0500)]
cmd/trace: extend trace view (/trace) for task-oriented view

R=go1.11

Change-Id: I2d2db148fed96d0fcb228bee414b050fe4e46e2c
Reviewed-on: https://go-review.googlesource.com/90217
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/trace: add analyzeAnnotation and /usertasks view.
Hana Kim [Mon, 22 Jan 2018 21:20:30 +0000 (16:20 -0500)]
cmd/trace: add analyzeAnnotation and /usertasks view.

R=go1.11

Change-Id: I5078ab714c8ac2c652e6ec496e01b063235a014a
Reviewed-on: https://go-review.googlesource.com/90216
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/trace: encode selection in trace URL
Austin Clements [Fri, 28 Jul 2017 14:23:04 +0000 (10:23 -0400)]
cmd/trace: encode selection in trace URL

This adds the ability to add a #x:y anchor to the trace view URL that
causes the viewer to initially select from x ms to y ms.

Change-Id: I4a980d8128ecc85dbe41f224e8ae336707a4eaab
Reviewed-on: https://go-review.googlesource.com/60794
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
7 years agocmd/compile: normalize spaces in rewrite rule comments.
Giovanni Bajo [Tue, 20 Feb 2018 01:18:54 +0000 (02:18 +0100)]
cmd/compile: normalize spaces in rewrite rule comments.

In addition to look nicer to the eye, this allows to reformat
and indent rules without causing spurious changes to the generated
file, making it easier to spot functional changes.

After this CL, all CLs that will aggregate rules through
the new "|" functionality should cause no changes to the
generated files.

Change-Id: Icec283585ba8d7b91c79d76513c1d83dca4b30aa
Reviewed-on: https://go-review.googlesource.com/95216
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/internal/obj/x86: small refactoring
Ilya Tocar [Wed, 13 Dec 2017 22:38:02 +0000 (16:38 -0600)]
cmd/internal/obj/x86: small refactoring

Replace some ints with bool and use arrays istead of slices where
possible.

Change-Id: I510bdaec48f9c437685e72c4a3291cffeb7ef5fc
Reviewed-on: https://go-review.googlesource.com/83859
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoall: remove duplicate word "the"
Ryuma Yoshida [Tue, 20 Feb 2018 15:10:49 +0000 (15:10 +0000)]
all: remove duplicate word "the"

Change-Id: Ia5908e94a6bd362099ca3c63f6ffb7e94457131d
GitHub-Last-Rev: 545a40571a912f433546d8c94a9d63459313515d
GitHub-Pull-Request: golang/go#23942
Reviewed-on: https://go-review.googlesource.com/95435
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agohtml/template: make more use of stringer
Daniel Martí [Tue, 20 Feb 2018 10:02:10 +0000 (10:02 +0000)]
html/template: make more use of stringer

The code was maintaining manual versions of it in multiple places -
replace all of them.

Change-Id: I04c3063877b05ba914de9f5dddb33ffe09f308fe
Reviewed-on: https://go-review.googlesource.com/95356
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotext/template: avoid assiging unnecessary variable
Kunpei Sakai [Mon, 19 Feb 2018 23:42:11 +0000 (08:42 +0900)]
text/template: avoid assiging unnecessary variable

This follows up CL95235

Change-Id: I62652654e5bb46d2f8f37af468e1fbcc1835bb8a
Reviewed-on: https://go-review.googlesource.com/95295
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoregexp/syntax: make Op an fmt.Stringer
Daniel Martí [Tue, 20 Feb 2018 09:55:28 +0000 (09:55 +0000)]
regexp/syntax: make Op an fmt.Stringer

Using stringer.

Fixes #22684.

Change-Id: I62fbde5dcb337cf269686615616bd39a27491ac1
Reviewed-on: https://go-review.googlesource.com/95355
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: optimize ARM64 code with MNEG
Ben Shi [Mon, 19 Feb 2018 13:13:13 +0000 (13:13 +0000)]
cmd/compile: optimize ARM64 code with MNEG

A pair of MUL/NEG instructions can be combined to a single MNEG on ARM64.
This CL implements this optimization.

1. A special test case gets big improvement.
(https://github.com/benshi001/ugo1/blob/master/mneg_test.go)
name                     old time/op    new time/op    delta
MNEG-4                      315µs ± 0%     260µs ± 0%  -17.39%  (p=0.000 n=24+25)

2. There is little change in the go1 benchmark, excluding noise.
name                     old time/op    new time/op    delta
BinaryTree17-4              42.2s ± 2%     41.9s ± 2%  -0.82%  (p=0.001 n=30+26)
Fannkuch11-4                32.9s ± 0%     32.9s ± 0%  -0.01%  (p=0.006 n=20+26)
FmtFprintfEmpty-4           541ns ± 3%     534ns ± 0%  -1.24%  (p=0.003 n=30+26)
FmtFprintfString-4         1.09µs ± 0%    1.10µs ± 3%    ~     (p=0.142 n=23+30)
FmtFprintfInt-4            1.14µs ± 0%    1.14µs ± 0%    ~     (p=0.435 n=24+24)
FmtFprintfIntInt-4         1.76µs ± 0%    1.76µs ± 0%    ~     (p=0.508 n=24+26)
FmtFprintfPrefixedInt-4    2.20µs ± 3%    2.17µs ± 0%  -1.10%  (p=0.017 n=30+24)
FmtFprintfFloat-4          3.28µs ± 0%    3.28µs ± 0%    ~     (p=0.579 n=24+24)
FmtManyArgs-4              7.30µs ± 0%    7.30µs ± 0%    ~     (p=0.662 n=26+27)
GobDecode-4                94.8ms ± 0%    94.8ms ± 0%  +0.07%  (p=0.010 n=25+23)
GobEncode-4                80.9ms ± 4%    80.6ms ± 4%    ~     (p=0.901 n=30+30)
Gzip-4                      4.45s ± 0%     4.49s ± 0%  +0.98%  (p=0.000 n=25+24)
Gunzip-4                    450ms ± 3%     443ms ± 0%    ~     (p=0.942 n=30+26)
HTTPClientServer-4          548µs ± 1%     551µs ± 1%  +0.60%  (p=0.000 n=29+30)
JSONEncode-4                210ms ± 0%     211ms ± 0%  +0.03%  (p=0.000 n=23+25)
JSONDecode-4                866ms ± 5%     877ms ± 5%    ~     (p=0.187 n=30+30)
Mandelbrot200-4            51.4ms ± 0%    52.0ms ± 3%  +1.15%  (p=0.001 n=24+30)
GoParse-4                  42.9ms ± 5%    41.9ms ± 0%  -2.24%  (p=0.000 n=30+26)
RegexpMatchEasy0_32-4      1.02µs ± 3%    1.01µs ± 0%    ~     (p=0.247 n=30+26)
RegexpMatchEasy0_1K-4      3.90µs ± 0%    3.90µs ± 0%    ~     (p=0.062 n=24+24)
RegexpMatchEasy1_32-4       955ns ± 0%     956ns ± 0%  +0.16%  (p=0.000 n=25+23)
RegexpMatchEasy1_1K-4      6.42µs ± 3%    6.37µs ± 0%  -0.81%  (p=0.012 n=30+24)
RegexpMatchMedium_32-4     1.77µs ± 3%    1.79µs ± 0%  +1.28%  (p=0.003 n=30+24)
RegexpMatchMedium_1K-4      561µs ± 0%     569µs ± 3%  +1.50%  (p=0.000 n=25+30)
RegexpMatchHard_32-4       31.0µs ± 4%    30.8µs ± 0%    ~     (p=1.000 n=26+26)
RegexpMatchHard_1K-4        945µs ± 3%     945µs ± 3%    ~     (p=0.513 n=30+30)
Revcomp-4                   7.76s ± 4%     7.68s ± 0%    ~     (p=0.464 n=29+23)
Template-4                  903ms ± 5%     904ms ± 5%    ~     (p=0.248 n=30+30)
TimeParse-4                4.80µs ± 0%    4.80µs ± 0%    ~     (p=0.081 n=25+26)
TimeFormat-4               4.70µs ± 1%    4.70µs ± 1%    ~     (p=0.763 n=24+26)
[Geo mean]                  709µs          708µs       -0.09%

name                     old speed      new speed      delta
GobDecode-4              8.10MB/s ± 0%  8.09MB/s ± 0%    ~     (p=0.160 n=25+23)
GobEncode-4              9.49MB/s ± 4%  9.53MB/s ± 4%    ~     (p=0.360 n=30+30)
Gzip-4                   4.36MB/s ± 0%  4.32MB/s ± 0%  -0.92%  (p=0.000 n=25+24)
Gunzip-4                 43.2MB/s ± 3%  43.8MB/s ± 0%    ~     (p=0.980 n=30+26)
JSONEncode-4             9.22MB/s ± 0%  9.22MB/s ± 0%  -0.04%  (p=0.005 n=23+25)
JSONDecode-4             2.24MB/s ± 5%  2.21MB/s ± 4%    ~     (p=0.252 n=30+30)
GoParse-4                1.35MB/s ± 5%  1.38MB/s ± 0%  +2.00%  (p=0.003 n=30+26)
RegexpMatchEasy0_32-4    31.5MB/s ± 3%  31.8MB/s ± 0%    ~     (p=0.110 n=30+26)
RegexpMatchEasy0_1K-4     263MB/s ± 0%   263MB/s ± 0%    ~     (p=0.111 n=24+24)
RegexpMatchEasy1_32-4    33.5MB/s ± 0%  33.4MB/s ± 0%  -0.16%  (p=0.003 n=25+23)
RegexpMatchEasy1_1K-4     160MB/s ± 3%   161MB/s ± 0%  +0.78%  (p=0.012 n=30+24)
RegexpMatchMedium_32-4    565kB/s ± 3%   560kB/s ± 0%  -0.83%  (p=0.001 n=30+24)
RegexpMatchMedium_1K-4   1.83MB/s ± 0%  1.80MB/s ± 3%  -1.56%  (p=0.000 n=25+30)
RegexpMatchHard_32-4     1.03MB/s ± 3%  1.04MB/s ± 0%  +1.46%  (p=0.000 n=30+26)
RegexpMatchHard_1K-4     1.08MB/s ± 3%  1.09MB/s ± 3%    ~     (p=0.444 n=30+30)
Revcomp-4                32.8MB/s ± 4%  33.1MB/s ± 0%    ~     (p=0.858 n=29+23)
Template-4               2.15MB/s ± 5%  2.15MB/s ± 5%    ~     (p=0.646 n=30+30)
[Geo mean]               7.79MB/s       7.81MB/s       +0.21%

3. There is no regression in the compilecmp benchmark.
name        old time/op       new time/op       delta
Template          2.35s ± 4%        2.33s ± 3%    ~     (p=0.796 n=10+10)
Unicode           1.35s ± 6%        1.35s ± 5%    ~     (p=1.000 n=9+10)
GoTypes           8.10s ± 3%        8.14s ± 3%    ~     (p=0.604 n=9+10)
Compiler          40.5s ± 2%        40.2s ± 2%    ~     (p=0.065 n=10+9)
SSA                115s ± 2%         115s ± 2%    ~     (p=0.447 n=9+10)
Flate             1.45s ± 3%        1.45s ± 4%    ~     (p=0.739 n=10+10)
GoParser          1.85s ± 3%        1.86s ± 2%    ~     (p=0.853 n=10+10)
Reflect           5.11s ± 2%        5.10s ± 2%    ~     (p=0.971 n=10+10)
Tar               2.23s ± 5%        2.23s ± 3%    ~     (p=0.796 n=10+10)
XML               2.67s ± 2%        2.69s ± 2%    ~     (p=0.549 n=9+10)
[Geo mean]        5.00s             5.00s       +0.02%

name        old user-time/op  new user-time/op  delta
Template          2.88s ± 2%        2.86s ± 2%    ~     (p=0.529 n=10+10)
Unicode           1.70s ± 7%        1.69s ± 5%    ~     (p=0.853 n=10+10)
GoTypes           9.72s ± 1%        9.73s ± 1%    ~     (p=0.684 n=10+10)
Compiler          49.0s ± 1%        48.9s ± 1%    ~     (p=0.631 n=10+10)
SSA                144s ± 1%         144s ± 2%    ~     (p=0.684 n=10+10)
Flate             1.71s ± 4%        1.72s ± 4%    ~     (p=0.853 n=10+10)
GoParser          2.23s ± 2%        2.23s ± 2%    ~     (p=0.971 n=10+10)
Reflect           5.98s ± 2%        5.96s ± 2%    ~     (p=0.481 n=10+10)
Tar               2.68s ± 3%        2.67s ± 2%    ~     (p=0.393 n=10+10)
XML               3.21s ± 3%        3.22s ± 1%    ~     (p=0.604 n=10+9)
[Geo mean]        6.05s             6.05s       -0.04%

name        old text-bytes    new text-bytes    delta
HelloSize         641kB ± 0%        641kB ± 0%    ~     (all equal)

name        old data-bytes    new data-bytes    delta
HelloSize        9.46kB ± 0%       9.46kB ± 0%    ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize         125kB ± 0%        125kB ± 0%    ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.24MB ± 0%       1.24MB ± 0%    ~     (all equal)

Change-Id: I9ed9128f0114e0f1ebb08ca2d042c90fcb2b1dcd
Reviewed-on: https://go-review.googlesource.com/95075
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agosyscall: ensure Mkdir(path) on Plan 9 fails if path exists
Richard Miller [Mon, 19 Feb 2018 12:34:53 +0000 (12:34 +0000)]
syscall: ensure Mkdir(path) on Plan 9 fails if path exists

On Plan 9, the underlying create() syscall with DMDIR flag, which is
used to implement Mkdir, will fail silently if the path exists and
is not a directory.  Work around this by checking for existence
first and rejecting Mkdir with error EEXIST if the path is found.

Fixes #23918

Change-Id: I439115662307923c9f498d3e7b1f32c6d205e1ad
Reviewed-on: https://go-review.googlesource.com/94777
Reviewed-by: David du Colombier <0intro@gmail.com>
7 years agonet: adjust the test for IPv4 loopback address block
Mikio Hara [Tue, 20 Feb 2018 03:57:51 +0000 (12:57 +0900)]
net: adjust the test for IPv4 loopback address block

We live in the era of virtualization and isolation.
There is no reason to hesitate to use IPv4 loopback address block for
umbrella-type customer accommodating services.

Fixes #23903

Change-Id: I990dd98e2651a993dac1b105c0bc771f8631cb93
Reviewed-on: https://go-review.googlesource.com/95336
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile/internal/ssa: emit csel on arm64
philhofer [Sun, 13 Aug 2017 22:36:47 +0000 (22:36 +0000)]
cmd/compile/internal/ssa: emit csel on arm64

Introduce a new SSA pass to generate CondSelect intstrutions,
and add CondSelect lowering rules for arm64.

In order to make the CSEL instruction easier to optimize,
and to simplify the introduction of CSNEG, CSINC, and CSINV
in the future, modify the CSEL instruction to accept a condition
code in the aux field.

Notably, this change makes the go1 Gzip benchmark
more than 10% faster.

Benchmarks on a Cavium ThunderX:

name                      old time/op    new time/op    delta
BinaryTree17-96              15.9s ± 6%     16.0s ± 4%     ~     (p=0.968 n=10+9)
Fannkuch11-96                7.17s ± 0%     7.00s ± 0%   -2.43%  (p=0.000 n=8+9)
FmtFprintfEmpty-96           208ns ± 1%     207ns ± 0%     ~     (p=0.152 n=10+8)
FmtFprintfString-96          379ns ± 0%     375ns ± 0%   -0.95%  (p=0.000 n=10+9)
FmtFprintfInt-96             385ns ± 0%     383ns ± 0%   -0.52%  (p=0.000 n=9+10)
FmtFprintfIntInt-96          591ns ± 0%     586ns ± 0%   -0.85%  (p=0.006 n=7+9)
FmtFprintfPrefixedInt-96     656ns ± 0%     667ns ± 0%   +1.71%  (p=0.000 n=10+10)
FmtFprintfFloat-96           967ns ± 0%     984ns ± 0%   +1.78%  (p=0.000 n=10+10)
FmtManyArgs-96              2.35µs ± 0%    2.25µs ± 0%   -4.63%  (p=0.000 n=9+8)
GobDecode-96                31.0ms ± 0%    30.8ms ± 0%   -0.36%  (p=0.006 n=9+9)
GobEncode-96                24.4ms ± 0%    24.5ms ± 0%   +0.30%  (p=0.000 n=9+9)
Gzip-96                      1.60s ± 0%     1.43s ± 0%  -10.58%  (p=0.000 n=9+10)
Gunzip-96                    167ms ± 0%     169ms ± 0%   +0.83%  (p=0.000 n=8+9)
HTTPClientServer-96          311µs ± 1%     308µs ± 0%   -0.75%  (p=0.000 n=10+10)
JSONEncode-96               65.0ms ± 0%    64.8ms ± 0%   -0.25%  (p=0.000 n=9+8)
JSONDecode-96                262ms ± 1%     261ms ± 1%     ~     (p=0.579 n=10+10)
Mandelbrot200-96            18.0ms ± 0%    18.1ms ± 0%   +0.17%  (p=0.000 n=8+10)
GoParse-96                  14.0ms ± 0%    14.1ms ± 1%   +0.42%  (p=0.003 n=9+10)
RegexpMatchEasy0_32-96       644ns ± 2%     645ns ± 2%     ~     (p=0.836 n=10+10)
RegexpMatchEasy0_1K-96      3.70µs ± 0%    3.49µs ± 0%   -5.58%  (p=0.000 n=10+10)
RegexpMatchEasy1_32-96       662ns ± 2%     657ns ± 2%     ~     (p=0.137 n=10+10)
RegexpMatchEasy1_1K-96      4.47µs ± 0%    4.31µs ± 0%   -3.48%  (p=0.000 n=10+10)
RegexpMatchMedium_32-96      844ns ± 2%     849ns ± 1%     ~     (p=0.208 n=10+10)
RegexpMatchMedium_1K-96      179µs ± 0%     182µs ± 0%   +1.20%  (p=0.000 n=10+10)
RegexpMatchHard_32-96       10.0µs ± 0%    10.1µs ± 0%   +0.48%  (p=0.000 n=10+9)
RegexpMatchHard_1K-96        297µs ± 0%     297µs ± 0%   -0.14%  (p=0.000 n=10+10)
Revcomp-96                   3.08s ± 0%     3.13s ± 0%   +1.56%  (p=0.000 n=9+9)
Template-96                  276ms ± 2%     275ms ± 1%     ~     (p=0.393 n=10+10)
TimeParse-96                1.37µs ± 0%    1.36µs ± 0%   -0.53%  (p=0.000 n=10+7)
TimeFormat-96               1.40µs ± 0%    1.42µs ± 0%   +0.97%  (p=0.000 n=10+10)
[Geo mean]                   264µs          262µs        -0.77%

Change-Id: Ie54eee4b3092af53e6da3baa6d1755098f57f3a2
Reviewed-on: https://go-review.googlesource.com/55670
Run-TryBot: Philip Hofer <phofer@umich.edu>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agotest: add test case where gccgo incorrectly rejected aliases
Ian Lance Taylor [Sun, 18 Feb 2018 21:52:39 +0000 (13:52 -0800)]
test: add test case where gccgo incorrectly rejected aliases

Updates #23912

Change-Id: I50d06506a8ac91ed99a761a9ff3fd0b03d4c8121
Reviewed-on: https://go-review.googlesource.com/94995
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: use RFC 723x as normative reference in docs
David Url [Tue, 13 Feb 2018 21:03:05 +0000 (22:03 +0100)]
net/http: use RFC 723x as normative reference in docs

Replace references to the obsoleted RFC 2616 with references to RFC
7230 through 7235, to avoid unnecessary confusion.
Obvious inconsistencies are marked with todo comments.

Updates #21974

Change-Id: I8fb4fcdd1333fc5193b93a2f09598f18c45e7a00
Reviewed-on: https://go-review.googlesource.com/94095
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: reset branch prediction when deleting a branch
Keith Randall [Mon, 22 Jan 2018 17:43:27 +0000 (09:43 -0800)]
cmd/compile: reset branch prediction when deleting a branch

When we go from a branch block to a plain block, reset the
branch prediction bit. Downstream passes asssume that if the
branch prediction is set, then the block has 2 successors.

Fixes #23504

Change-Id: I2898ec002228b2e34fe80ce420c6939201c0a5aa
Reviewed-on: https://go-review.googlesource.com/88955
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agoencoding/xml: simplify slice-growing logic in rawToken
Alberto Donizetti [Mon, 8 Jan 2018 12:03:57 +0000 (13:03 +0100)]
encoding/xml: simplify slice-growing logic in rawToken

It appears that old code (from 2009) in xml.(*Decoder).rawToken
replicates append's slice-growing functionality by allocating a new,
bigger backing array and then calling copy.

Simplifying the code by replacing it with a single append call does
not seem to hurt performance:

name         old time/op    new time/op    delta
Marshal-4      11.2µs ± 1%    11.3µs ±10%    ~     (p=0.069 n=19+17)
Unmarshal-4    28.6µs ± 1%    28.4µs ± 1%  -0.60%  (p=0.000 n=20+18)

name         old alloc/op   new alloc/op   delta
Marshal-4      5.78kB ± 0%    5.78kB ± 0%    ~     (all equal)
Unmarshal-4    8.61kB ± 0%    8.27kB ± 0%  -3.90%  (p=0.000 n=20+20)

name         old allocs/op  new allocs/op  delta
Marshal-4        23.0 ± 0%      23.0 ± 0%    ~     (all equal)
Unmarshal-4       189 ± 0%       190 ± 0%  +0.53%  (p=0.000 n=20+20)

Change-Id: Ie580d1216a44760e611e63dee2c339af5465aea5
Reviewed-on: https://go-review.googlesource.com/86655
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoregexp: dont use builtin type as variable name
Ahmet Soormally [Mon, 15 Jan 2018 12:12:46 +0000 (12:12 +0000)]
regexp: dont use builtin type as variable name

The existing implementation declares a variable error which collides
with builting type error.

This change simply renames error variable to err.

Change-Id: Ib56c2530f37f53ec70fdebb825a432d4c550cd04
Reviewed-on: https://go-review.googlesource.com/87775
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>

7 years agotext/template: remove associate's error return
Daniel Martí [Mon, 19 Feb 2018 21:03:12 +0000 (21:03 +0000)]
text/template: remove associate's error return

It's always nil, so simplify its signature. Found with unparam.

Change-Id: I45dd0f868ec2f5de98a970776be686417c8d73b6
Reviewed-on: https://go-review.googlesource.com/95235
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: add | operator to make rewrite rules more succinct
Keith Randall [Sun, 7 Jan 2018 21:23:59 +0000 (13:23 -0800)]
cmd/compile: add | operator to make rewrite rules more succinct

Instead of

(And64 x x) -> x
(And32 x x) -> x
(And16 x x) -> x
(And8  x x) -> x

we can now do:

(And(64|32|16|8) x x) -> x

Any part of an opcode can have a parenthesized, |-separated list of possibilites.
The rule is then expanded using each piece of the | combo.
If there are multiple | clauses, they get expanded in tandem.
(All the first positions, then all the second positions, etc.)
All places | opcodes appear must have the same count.

A more complicated example:

(MOV(L|SS)load [off1] {sym1} (LEAQ4 [off2] {sym2} ptr idx) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
(MOV(L|SS)loadidx4 [off1+off2] {mergeSym(sym1,sym2)} ptr idx mem)

This meta-rule generates 2 rules, a MOVL and a MOVSS rule.

This CL is carefully orchestrated to not change the generated rules file at all.
In some cases, this means we can't align the rules nicely because it changes
the whitespace in the generated code.  I'll clean that up as a separate step.

There are many more opportunites to compactify rules using this new mechanism.
I've just done some examples, there's more to do.

Change-Id: I8a5e748cd0761ccbb12d09b01925b2f1f4b2f608
Reviewed-on: https://go-review.googlesource.com/86595
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agoall: add more uses of stringer
Daniel Martí [Mon, 13 Nov 2017 09:43:17 +0000 (09:43 +0000)]
all: add more uses of stringer

By grepping for ]string{$, one can find many manual implementations of
stringer. The debug/dwarf ones needed the new -trimprefix flag, too.

html/template was fairly simple, just implementing the fallback as
stringer would. The changes there are trivial.

The ones in debug/dwarf needed a bit of extra logic since the GoString
wants to use its own format, depending on whether or not the value is
one of the known constants.

Change-Id: I501ea7deaa538fa425c8e9c2bb895f480169273f
Reviewed-on: https://go-review.googlesource.com/77253
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotext/template: differentiate nil from missing arg
Daniel Martí [Mon, 19 Feb 2018 20:48:58 +0000 (20:48 +0000)]
text/template: differentiate nil from missing arg

reflect.Value is a struct and does not have a kind nor any flag for
untyped nils. As a result, it is tricky to differentiate when we're
missing a value, from when we have one but it is untyped nil.

We could start using *reflect.Value instead, to add one level of
indirection, using nil for missing values and new(reflect.Value) for
untyped nils. However, that is a fairly invasive change, and would also
mean unnecessary allocations.

Instead, use a special reflect.Value that depicts when a value is
missing. This is the case for the "final" reflect.Value in multiple
scenarios, such as the start of a pipeline. Give it a specific,
unexported type too, to make sure it cannot be mistaken for any other
valid value.

Finally, replace "final.IsValid()" with "final != missingVal", since
final.IsValid() will be false when final is an untyped nil.

Also add a few test cases, all different variants of the untyped nil
versus missing value scenario.

Fixes #18716.

Change-Id: Ia9257a84660ead5a7007fd1cced7782760b62d9d
Reviewed-on: https://go-review.googlesource.com/95215
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
7 years agohtml/template: remove unused func
Daniel Martí [Mon, 19 Feb 2018 19:18:25 +0000 (19:18 +0000)]
html/template: remove unused func

Noticed while reading some code.

Change-Id: I63f8c7a453d4895583a8ffc13bec57385b6944aa
Reviewed-on: https://go-review.googlesource.com/95195
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: avoid clearing memory during byte slice allocation in gobytes
Martin Möhrmann [Sun, 18 Feb 2018 13:12:52 +0000 (14:12 +0100)]
runtime: avoid clearing memory during byte slice allocation in gobytes

Avoid using make in gobytes which clears the byte slice backing
array unnecessarily since the content is overwritten immediately again.

Check that the user provided length is positive and below the maximum
allowed allocation size explicitly in gobytes as this was done in makeslice
before this change.

Fixes #23634

Change-Id: Id852619e932aabfc468871c42ad07d34da91f45c
Reviewed-on: https://go-review.googlesource.com/94760
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc: improve clarity of map index examples
Yazen2017 [Sat, 17 Feb 2018 22:24:21 +0000 (17:24 -0500)]
doc: improve clarity of map index examples

The fourth example for map indexing states you have a map of type [K]V
and attempts to read in a variable of type T.  Further, the example
is meant to showcase the boolean return variable saying whether the
map contained a key, but overrides to type T.  This will not compile.

Changed last updated date to February 18

Fixes: #23895
Change-Id: I63c52adbcd989afd4855e329e6c727f4c01f7881
Reviewed-on: https://go-review.googlesource.com/94906
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoall: remove "the" duplications
Kunpei Sakai [Sun, 18 Feb 2018 17:20:57 +0000 (02:20 +0900)]
all: remove "the" duplications

Change-Id: I1f25b11fb9b7cd3c09968ed99913dc85db2025ef
Reviewed-on: https://go-review.googlesource.com/94976
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoos: make MkdirAll support path in extended-length form
Mansour Rahimi [Sun, 28 Jan 2018 21:19:13 +0000 (22:19 +0100)]
os: make MkdirAll support path in extended-length form

Calling MkdirAll on paths in extended-length form (\\?\-prefixed)
failed.

MkdirAll calls itself recursively with parent directory of given path in
its parameter. It finds parent directory by looking for delimiter in
the path, and taking the left part. When path is in extended-length form,
it finds empty path at the end.

Here is a sample of path in extended-length form:
\\?\c:\foo\bar

This change fixes that by passing trailing path separator to MkdirAll (so
it works for path like \\?\c:\).

Fixes #22230

Change-Id: I363660b262588c5382ea829773d3b6005ab8df3c
Reviewed-on: https://go-review.googlesource.com/86295
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: remove unused getrlimit function
Tobias Klauser [Fri, 16 Feb 2018 10:26:14 +0000 (11:26 +0100)]
runtime: remove unused getrlimit function

Follow CL 93655 which removed the (commented-out) usage of this
function.

Also remove unused constant _RLIMIT_AS and type rlimit.

Change-Id: Ifb6e6b2104f4c2555269f8ced72bfcae24f5d5e9
Reviewed-on: https://go-review.googlesource.com/94775
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/go: document 'go run' exit codes
Yury Smolsky [Fri, 16 Feb 2018 12:56:03 +0000 (14:56 +0200)]
cmd/go: document 'go run' exit codes

Updated docs that go run does not return the exit code of
the compiled binary.

Fixes #23716

Change-Id: Ib85459974c4c6d2760ddba957ef711628098661f
Reviewed-on: https://go-review.googlesource.com/94795
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: move map fast functions into type specific files
Martin Möhrmann [Sat, 27 Jan 2018 11:38:13 +0000 (12:38 +0100)]
runtime: move map fast functions into type specific files

Overall code is unchanged.

The functions for different types (32, 64, str) of map fast routines
are collected in map_fast.go that has grown to ~1300 lines.

Moving the functions for each map fast type into a separate file
allows for an easier overview and navigation within the map code.

Change-Id: Ic09e4212f9025a66a10b11ef8dac23ad49d1d5ae
Reviewed-on: https://go-review.googlesource.com/90335
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agocmd/compile: replace misleading variable name
Martin Möhrmann [Sun, 21 Jan 2018 09:25:07 +0000 (10:25 +0100)]
cmd/compile: replace misleading variable name

One of the variables declared in cleantempnopop named 'kill'
does not hold a OVARKILL node but an OVARLIVE node.
Rename that variable to 'live' to differentiate it from the other
variable named kill that holds a OVARKILL node.

Passes toolstash -cmp.

Change-Id: I34c8729e5c303b8cdabe44c9af980d4f16000e4b
Reviewed-on: https://go-review.googlesource.com/88816
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: rename map implementation and test files to use a common prefix
Martin Möhrmann [Sat, 27 Jan 2018 11:48:15 +0000 (12:48 +0100)]
runtime: rename map implementation and test files to use a common prefix

Rename all map implementation and test files to use "map"
as a file name prefix instead of "hashmap" for the implementation
and "map" for the test file names.

Change-Id: I7b317c1f7a660b95c6d1f1a185866f2839e69446
Reviewed-on: https://go-review.googlesource.com/90336
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agodoc: update devel/release.html to add Go 1.10
Andrew Bonventre [Sat, 17 Feb 2018 04:26:02 +0000 (23:26 -0500)]
doc: update devel/release.html to add Go 1.10

Change-Id: I869f1607b0209ad190bc02999879f0682965fdf7
Reviewed-on: https://go-review.googlesource.com/94915
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/go: permit pkg-config flags in any argument position
Ian Lance Taylor [Sat, 17 Feb 2018 00:17:44 +0000 (16:17 -0800)]
cmd/go: permit pkg-config flags in any argument position

Fixes #23875

Change-Id: I503af71f44d11cd6b787fef100246b55735614a0
Reviewed-on: https://go-review.googlesource.com/94896
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocompress/bzip2: use sort.Slice in huffman.go
Joe Kyo [Wed, 15 Nov 2017 08:48:49 +0000 (08:48 +0000)]
compress/bzip2: use sort.Slice in huffman.go

Change-Id: Ie4d23cdb81473a4c989a977a127479cf825084dc
Reviewed-on: https://go-review.googlesource.com/77850
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: fix command injection in VCS path
Arthur Khashaev [Mon, 12 Feb 2018 00:28:12 +0000 (03:28 +0300)]
cmd/go: fix command injection in VCS path

Fixes #23867, CVE-2018-7187

Change-Id: I5d0ba4923c9ed354ef76290e149c182447f9dfe2
Reviewed-on: https://go-review.googlesource.com/94656
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/go: restrict meta imports to valid schemes
Ian Lance Taylor [Thu, 15 Feb 2018 23:57:13 +0000 (15:57 -0800)]
cmd/go: restrict meta imports to valid schemes

Before this change, when using -insecure, we permitted any meta import
repo root as long as it contained "://". When not using -insecure, we
restrict meta import repo roots to be valid URLs. People may depend on
that somehow, so permit meta import repo roots to be invalid URLs, but
require them to have valid schemes per RFC 3986.

Fixes #23867

Change-Id: Iac666dfc75ac321bf8639dda5b0dba7c8840922d
Reviewed-on: https://go-review.googlesource.com/94603
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: increase timeout length for TestOnlyWriteTimeout
Richard Miller [Fri, 16 Feb 2018 17:01:52 +0000 (17:01 +0000)]
net/http: increase timeout length for TestOnlyWriteTimeout

This test was sometimes timing out on the plan9/arm builder
(raspberry pi) when run in parallel with other network intensive
tests. It appears that tcp on the loopback interface could do
with some tuning for better performance on Plan 9, but until
that's done, increasing the timeout from 5 to 10 seconds allows
this test to pass.  This should have no effect on other platforms
where 5 seconds was already enough.

Change-Id: If310ee569cae8ca8f56346d84ce23803feb23a41
Reviewed-on: https://go-review.googlesource.com/94796
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: don't ignore address hint for sysReserve in Plan 9
Richard Miller [Fri, 16 Feb 2018 15:20:04 +0000 (15:20 +0000)]
runtime: don't ignore address hint for sysReserve in Plan 9

On Plan 9, sysReserve was ignoring the address hint and allocating
memory wherever it is available.  This causes the new
TestArenaCollision test to fail on 32-bit Plan 9.  We now use the
address hint in the specific case where sysReserve is extending the
process address space at its end, and similarly we contract the
address space in the case where sysFree is releasing memory at
the end.

Fixes #23860

Change-Id: Ia5254779ba8f1698c999832720a88de400b5f91a
Reviewed-on: https://go-review.googlesource.com/94776
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: David du Colombier <0intro@gmail.com>
7 years agocmd/compile: make math.Ceil/Floor/Round/Trunc intrinsics on arm64
Chad Rosier [Fri, 16 Feb 2018 14:22:32 +0000 (09:22 -0500)]
cmd/compile: make math.Ceil/Floor/Round/Trunc intrinsics on arm64

name       old time/op  new time/op  delta
Ceil        550ns ± 0%   486ns ± 7%  -11.64%  (p=0.000 n=13+18)
Floor       495ns ±19%   512ns ±12%     ~     (p=0.164 n=20+20)
Round       550ns ± 0%   487ns ± 8%  -11.49%  (p=0.000 n=12+19)
Trunc       563ns ± 7%   488ns ±13%  -13.44%  (p=0.000 n=15+2)

Change-Id: I53f234b160b3c026a277506e2cf977d150379464
Reviewed-on: https://go-review.googlesource.com/88295
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agoruntime: only run TestArenaCollision if the target can exec
Elias Naur [Fri, 16 Feb 2018 09:26:00 +0000 (10:26 +0100)]
runtime: only run TestArenaCollision if the target can exec

Replace the test for nacl with testenv.MustHaveExec to also skip
test on iOS.

Change-Id: I6822714f6d71533d1b18bbb7894f6ad339d8aea1
Reviewed-on: https://go-review.googlesource.com/94755
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/compile: permit go:cgo_import_dynamic anywhere
Ian Lance Taylor [Wed, 14 Feb 2018 15:42:40 +0000 (07:42 -0800)]
cmd/compile: permit go:cgo_import_dynamic anywhere

It's used on Solaris to import symbols from shared libraries, e.g., in
golang.org/x/sys/unix and golang.org/x/net/internal/socket.
We could use a different directive but that would require build tags
in all the places that use it.

Updates #23672
Updates #23749

Change-Id: I47fcf72a6d2862e304204705979c2056c2f78ec5
Reviewed-on: https://go-review.googlesource.com/94018
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agodoc: add note about invalid flag errors to 1.10 release notes
Ian Lance Taylor [Thu, 15 Feb 2018 02:04:54 +0000 (18:04 -0800)]
doc: add note about invalid flag errors to 1.10 release notes

Updates #23672
Updates #23749

Change-Id: I85b6493bd1e4095398508b7ef395c35463dfbde1
Reviewed-on: https://go-review.googlesource.com/94158
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
7 years agocmd/compile: arm64 intrinsics for math/bits.OnesCount
Balaram Makam [Tue, 30 Jan 2018 17:16:52 +0000 (12:16 -0500)]
cmd/compile: arm64 intrinsics for math/bits.OnesCount

This adds math/bits intrinsics for OnesCount on arm64.

name         old time/op  new time/op  delta
OnesCount    3.81ns ± 0%  1.60ns ± 0%  -57.96%  (p=0.000 n=7+8)
OnesCount8   1.60ns ± 0%  1.60ns ± 0%     ~     (all equal)
OnesCount16  2.41ns ± 0%  1.60ns ± 0%  -33.61%  (p=0.000 n=8+8)
OnesCount32  4.17ns ± 0%  1.60ns ± 0%  -61.58%  (p=0.000 n=8+8)
OnesCount64  3.80ns ± 0%  1.60ns ± 0%  -57.84%  (p=0.000 n=8+8)

Update #18616

Conflicts:
src/cmd/compile/internal/gc/asm_test.go

Change-Id: I63ac2f63acafdb1f60656ab8a56be0b326eec5cb
Reviewed-on: https://go-review.googlesource.com/90835
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile/internal/gc: use functype instead of OTFUNC
Matthew Dempsky [Fri, 7 Apr 2017 22:39:36 +0000 (15:39 -0700)]
cmd/compile/internal/gc: use functype instead of OTFUNC

Slightly simpler.

Change-Id: Ic3a96675c56cc8c2e336b932536c2247f8cbb96d
Reviewed-on: https://go-review.googlesource.com/39996
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoruntime: replace _MaxMem with maxAlloc
Austin Clements [Tue, 2 Jan 2018 02:51:47 +0000 (21:51 -0500)]
runtime: replace _MaxMem with maxAlloc

Now that we have memLimit, also having _MaxMem is a bit confusing.

Replace it with maxAlloc, which better conveys what it limits. We also
define maxAlloc slightly differently: since it's now clear that it
limits allocation size, we can account for a subtle difference between
32-bit and 64-bit.

Change-Id: Iac39048018cc0dae7f0919e25185fee4b3eed529
Reviewed-on: https://go-review.googlesource.com/85890
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: move comment about address space sizes to malloc.go
Austin Clements [Mon, 1 Jan 2018 22:53:59 +0000 (17:53 -0500)]
runtime: move comment about address space sizes to malloc.go

Currently there's a detailed comment in lfstack_64bit.go about address
space limitations on various architectures. Since that's now relevant
to malloc, move it to a more prominent place in the documentation for
memLimitBits.

Updates #10460.

Change-Id: If9708291cf3a288057b8b3ba0ba6a59e3602bbd6
Reviewed-on: https://go-review.googlesource.com/85889
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: remove non-reserved heap logic
Austin Clements [Sun, 31 Dec 2017 00:35:46 +0000 (19:35 -0500)]
runtime: remove non-reserved heap logic

Currently large sysReserve calls on some OSes don't actually reserve
the memory, but just check that it can be reserved. This was important
when we called sysReserve to "reserve" many gigabytes for the heap up
front, but now that we map memory in small increments as we need it,
this complication is no longer necessary.

This has one curious side benefit: currently, on Linux, allocations
that are large enough to be rejected by mmap wind up freezing the
application for a long time before it panics. This happens because
sysReserve doesn't reserve the memory, so sysMap calls mmap_fixed,
which calls mmap, which fails because the mapping is too large.
However, mmap_fixed doesn't inspect *why* mmap fails, so it falls back
to probing every page in the desired region individually with mincore
before performing an (otherwise dangerous) MAP_FIXED mapping, which
will also fail. This takes a long time for a large region. Now this
logic is gone, so the mmap failure leads to an immediate panic.

Updates #10460.

Change-Id: I8efe88c611871cdb14f99fadd09db83e0161ca2e
Reviewed-on: https://go-review.googlesource.com/85888
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: use sparse mappings for the heap
Austin Clements [Wed, 20 Dec 2017 06:05:23 +0000 (22:05 -0800)]
runtime: use sparse mappings for the heap

This replaces the contiguous heap arena mapping with a potentially
sparse mapping that can support heap mappings anywhere in the address
space.

This has several advantages over the current approach:

* There is no longer any limit on the size of the Go heap. (Currently
  it's limited to 512GB.) Hence, this fixes #10460.

* It eliminates many failures modes of heap initialization and
  growing. In particular it eliminates any possibility of panicking
  with an address space conflict. This can happen for many reasons and
  even causes a low but steady rate of TSAN test failures because of
  conflicts with the TSAN runtime. See #16936 and #11993.

* It eliminates the notion of "non-reserved" heap, which was added
  because creating huge address space reservations (particularly on
  64-bit) led to huge process VSIZE. This was at best confusing and at
  worst conflicted badly with ulimit -v. However, the non-reserved
  heap logic is complicated, can race with other mappings in non-pure
  Go binaries (e.g., #18976), and requires that the entire heap be
  either reserved or non-reserved. We currently maintain the latter
  property, but it's quite difficult to convince yourself of that, and
  hence difficult to keep correct. This logic is still present, but
  will be removed in the next CL.

* It fixes problems on 32-bit where skipping over parts of the address
  space leads to mapping huge (and never-to-be-used) metadata
  structures. See #19831.

This also completely rewrites and significantly simplifies
mheap.sysAlloc, which has been a source of many bugs. E.g., #21044,
 #20259, #18651, and #13143 (and maybe #23222).

This change also makes it possible to allocate individual objects
larger than 512GB. As a result, a few tests that expected huge
allocations to fail needed to be changed to make even larger
allocations. However, at the moment attempting to allocate a humongous
object may cause the program to freeze for several minutes on Linux as
we fall back to probing every page with addrspace_free. That logic
(and this failure mode) will be removed in the next CL.

Fixes #10460.
Fixes #22204 (since it rewrites the code involved).

This slightly slows down compilebench and the x/benchmarks garbage
benchmark.

name       old time/op     new time/op     delta
Template       184ms ± 1%      185ms ± 1%    ~     (p=0.065 n=10+9)
Unicode       86.9ms ± 3%     86.3ms ± 1%    ~     (p=0.631 n=10+10)
GoTypes        599ms ± 0%      602ms ± 0%  +0.56%  (p=0.000 n=10+9)
Compiler       2.87s ± 1%      2.89s ± 1%  +0.51%  (p=0.002 n=9+10)
SSA            7.29s ± 1%      7.25s ± 1%    ~     (p=0.182 n=10+9)
Flate          118ms ± 2%      118ms ± 1%    ~     (p=0.113 n=9+9)
GoParser       147ms ± 1%      148ms ± 1%  +1.07%  (p=0.003 n=9+10)
Reflect        401ms ± 1%      404ms ± 1%  +0.71%  (p=0.003 n=10+9)
Tar            175ms ± 1%      175ms ± 1%    ~     (p=0.604 n=9+10)
XML            209ms ± 1%      210ms ± 1%    ~     (p=0.052 n=10+10)

(https://perf.golang.org/search?q=upload:20171231.4)

name                       old time/op  new time/op  delta
Garbage/benchmem-MB=64-12  2.23ms ± 1%  2.25ms ± 1%  +0.84%  (p=0.000 n=19+19)

(https://perf.golang.org/search?q=upload:20171231.3)

Relative to the start of the sparse heap changes (starting at and
including "runtime: fix various contiguous bitmap assumptions"),
overall slowdown is roughly 1% on GC-intensive benchmarks:

name        old time/op     new time/op     delta
Template        183ms ± 1%      185ms ± 1%  +1.32%  (p=0.000 n=9+9)
Unicode        84.9ms ± 2%     86.3ms ± 1%  +1.65%  (p=0.000 n=9+10)
GoTypes         595ms ± 1%      602ms ± 0%  +1.19%  (p=0.000 n=9+9)
Compiler        2.86s ± 0%      2.89s ± 1%  +0.91%  (p=0.000 n=9+10)
SSA             7.19s ± 0%      7.25s ± 1%  +0.75%  (p=0.000 n=8+9)
Flate           117ms ± 1%      118ms ± 1%  +1.10%  (p=0.000 n=10+9)
GoParser        146ms ± 2%      148ms ± 1%  +1.48%  (p=0.002 n=10+10)
Reflect         398ms ± 1%      404ms ± 1%  +1.51%  (p=0.000 n=10+9)
Tar             173ms ± 1%      175ms ± 1%  +1.17%  (p=0.000 n=10+10)
XML             208ms ± 1%      210ms ± 1%  +0.62%  (p=0.011 n=10+10)
[Geo mean]      369ms           373ms       +1.17%

(https://perf.golang.org/search?q=upload:20180101.2)

name                       old time/op  new time/op  delta
Garbage/benchmem-MB=64-12  2.22ms ± 1%  2.25ms ± 1%  +1.51%  (p=0.000 n=20+19)

(https://perf.golang.org/search?q=upload:20180101.3)

Change-Id: I5daf4cfec24b252e5a57001f0a6c03f22479d0f0
Reviewed-on: https://go-review.googlesource.com/85887
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: eliminate most uses of mheap_.arena_*
Austin Clements [Tue, 19 Dec 2017 04:35:34 +0000 (20:35 -0800)]
runtime: eliminate most uses of mheap_.arena_*

This replaces all uses of the mheap_.arena_* fields outside of
mallocinit and sysAlloc. These fields fundamentally assume a
contiguous heap between two bounds, so eliminating these is necessary
for a sparse heap.

Many of these are replaced with checks for non-nil spans at the test
address (which in turn checks for a non-nil entry in the heap arena
array). Some of them are just for debugging and somewhat meaningless
with a sparse heap, so those we just delete.

Updates #10460.

Change-Id: I8345b95ffc610aed694f08f74633b3c63506a41f
Reviewed-on: https://go-review.googlesource.com/85886
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: make span map sparse
Austin Clements [Wed, 13 Dec 2017 21:09:02 +0000 (16:09 -0500)]
runtime: make span map sparse

This splits the span map into separate chunks for every 64MB of the
heap. The span map chunks now live in the same indirect structure as
the bitmap.

Updates #10460.

This causes a slight improvement in compilebench and the x/benchmarks
garbage benchmark. I'm not sure why it improves performance.

name       old time/op     new time/op     delta
Template       185ms ± 1%      184ms ± 1%    ~            (p=0.315 n=9+10)
Unicode       86.9ms ± 1%     86.9ms ± 3%    ~            (p=0.356 n=9+10)
GoTypes        602ms ± 1%      599ms ± 0%  -0.59%         (p=0.002 n=9+10)
Compiler       2.89s ± 0%      2.87s ± 1%  -0.50%          (p=0.003 n=9+9)
SSA            7.25s ± 0%      7.29s ± 1%    ~            (p=0.400 n=9+10)
Flate          118ms ± 1%      118ms ± 2%    ~            (p=0.065 n=10+9)
GoParser       147ms ± 2%      147ms ± 1%    ~            (p=0.549 n=10+9)
Reflect        403ms ± 1%      401ms ± 1%  -0.47%         (p=0.035 n=9+10)
Tar            176ms ± 1%      175ms ± 1%  -0.59%         (p=0.013 n=10+9)
XML            211ms ± 1%      209ms ± 1%  -0.83%        (p=0.011 n=10+10)

(https://perf.golang.org/search?q=upload:20171231.1)

name                       old time/op  new time/op  delta
Garbage/benchmem-MB=64-12  2.24ms ± 1%  2.23ms ± 1%  -0.36%  (p=0.001 n=20+19)

(https://perf.golang.org/search?q=upload:20171231.2)

Change-Id: I2563f8704ab9812434947faf293c5327f9b0d07a
Reviewed-on: https://go-review.googlesource.com/85885
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: abstract remaining mheap.spans access
Austin Clements [Wed, 13 Dec 2017 21:03:23 +0000 (16:03 -0500)]
runtime: abstract remaining mheap.spans access

This abstracts the remaining direct accesses to mheap.spans into new
mheap.setSpan and mheap.setSpans methods.

For #10460.

Change-Id: Id1db8bc5e34a77a9221032aa2e62d05322707364
Reviewed-on: https://go-review.googlesource.com/85884
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: make the heap bitmap sparse
Austin Clements [Sat, 9 Dec 2017 03:57:53 +0000 (22:57 -0500)]
runtime: make the heap bitmap sparse

This splits the heap bitmap into separate chunks for every 64MB of the
heap and introduces an index mapping from virtual address to metadata.
It modifies the heapBits abstraction to use this two-level structure.
Finally, it modifies heapBitsSetType to unroll the bitmap into the
object itself and then copy it out if the bitmap would span
discontiguous bitmap chunks.

This is a step toward supporting general sparse heaps, which will
eliminate address space conflict failures as well as the limit on the
heap size.

It's also advantageous for 32-bit. 32-bit already supports
discontiguous heaps by always starting the arena at address 0.
However, as a result, with a contiguous bitmap, if the kernel chooses
a high address (near 2GB) for a heap mapping, the runtime is forced to
map up to 128MB of heap bitmap. Now the runtime can map sections of
the bitmap for just the parts of the address space used by the heap.

Updates #10460.

This slightly slows down the x/garbage and compilebench benchmarks.
However, I think the slowdown is acceptably small.

name        old time/op     new time/op     delta
Template        178ms ± 1%      180ms ± 1%  +0.78%    (p=0.029 n=10+10)
Unicode        85.7ms ± 2%     86.5ms ± 2%    ~       (p=0.089 n=10+10)
GoTypes         594ms ± 0%      599ms ± 1%  +0.70%    (p=0.000 n=9+9)
Compiler        2.86s ± 0%      2.87s ± 0%  +0.40%    (p=0.001 n=9+9)
SSA             7.23s ± 2%      7.29s ± 2%  +0.94%    (p=0.029 n=10+10)
Flate           116ms ± 1%      117ms ± 1%  +0.99%    (p=0.000 n=9+9)
GoParser        146ms ± 1%      146ms ± 0%    ~       (p=0.193 n=10+7)
Reflect         399ms ± 0%      403ms ± 1%  +0.89%    (p=0.001 n=10+10)
Tar             173ms ± 1%      174ms ± 1%  +0.91%    (p=0.013 n=10+9)
XML             208ms ± 1%      210ms ± 1%  +0.93%    (p=0.000 n=10+10)
[Geo mean]      368ms           371ms       +0.79%

name                       old time/op  new time/op  delta
Garbage/benchmem-MB=64-12  2.17ms ± 1%  2.21ms ± 1%  +2.15%  (p=0.000 n=20+20)

Change-Id: I037fd283221976f4f61249119d6b97b100bcbc66
Reviewed-on: https://go-review.googlesource.com/85883
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: fix various contiguous bitmap assumptions
Austin Clements [Sat, 9 Dec 2017 03:24:59 +0000 (22:24 -0500)]
runtime: fix various contiguous bitmap assumptions

There are various places that assume the heap bitmap is contiguous and
scan it sequentially. We're about to split up the heap bitmap. This
commit modifies all of these except heapBitsSetType to use the
heapBits abstractions so they can transparently switch to a
discontiguous bitmap.

Updates #10460. This is a step toward supporting sparse heaps.

Change-Id: I2f3994a5785e4dccb66602fb3950bbd290d9392c
Reviewed-on: https://go-review.googlesource.com/85882
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: lay out heap bitmap forward in memory
Austin Clements [Thu, 23 Jun 2016 20:25:50 +0000 (14:25 -0600)]
runtime: lay out heap bitmap forward in memory

Currently the heap bitamp is laid in reverse order in memory relative
to the heap itself. This was originally done out of "excessive
cleverness" so that computing a bitmap pointer could load only the
arena_start field and so that heaps could be more contiguous by
growing the arena and the bitmap out from a common center point.

However, this appears to have no actual performance benefit, it
complicates nearly every use of the bitmap, and it makes already
confusing code more confusing. Furthermore, it's still possible to use
a single field (the new bitmap_delta) for the bitmap pointer
computation by employing slightly different excessive cleverness.

Hence, this CL puts the bitmap into forward order.

This is a (very) updated version of CL 9404.

Change-Id: I743587cc626c4ecd81e660658bad85b54584108c
Reviewed-on: https://go-review.googlesource.com/85881
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: use spanOf* more widely
Austin Clements [Mon, 4 Dec 2017 16:02:59 +0000 (11:02 -0500)]
runtime: use spanOf* more widely

The logic in the spanOf* functions is open-coded in a lot of places
right now. Replace these with calls to the spanOf* functions.

Change-Id: I3cc996aceb9a529b60fea7ec6fef22008c012978
Reviewed-on: https://go-review.googlesource.com/85880
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: consolidate mheap.lookup* and spanOf*
Austin Clements [Mon, 4 Dec 2017 15:58:15 +0000 (10:58 -0500)]
runtime: consolidate mheap.lookup* and spanOf*

I think we'd forgotten about the mheap.lookup APIs when we introduced
spanOf*, but, at any rate, the spanOf* functions are used far more
widely at this point, so this CL eliminates the mheap.lookup*
functions in favor of spanOf*.

Change-Id: I15facd0856e238bb75d990e838a092b5bef5bdfc
Reviewed-on: https://go-review.googlesource.com/85879
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: split object finding out of heapBitsForObject
Austin Clements [Tue, 12 Dec 2017 00:40:12 +0000 (19:40 -0500)]
runtime: split object finding out of heapBitsForObject

heapBitsForObject does two things: it finds the base of the object and
it creates the heapBits for the base of the object. There are several
places where we just care about the base of the object. Furthermore,
greyobject only needs the heapBits in the checkmark path and can
easily compute them only when needed. Once we eliminate passing the
heap bits to grayobject, almost all uses of heapBitsForObject don't
need the heap bits.

Hence, this splits heapBitsForObject into findObject and
heapBitsForAddr (the latter already exists), removes the hbits
argument to grayobject, and replaces all heapBitsForObject calls with
calls to findObject.

In addition to making things cleaner overall, heapBitsForAddr is going
to get more expensive shortly, so it's important that we don't do it
needlessly.

Note that there's an interesting performance pitfall here. I had
originally moved findObject to mheap.go, since it made more sense
there. However, that leads to a ~2% slow down and a whopping 11%
increase in L1 icache misses on both the x/garbage and compilebench
benchmarks. This suggests we may want to be more principled about
this, but, for now, let's just leave findObject in mbitmap.go.

(I tried to make findObject small enough to inline by splitting out
the error case, but, sadly, wasn't quite able to get it under the
inlining budget.)

Change-Id: I7bcb92f383ade565d22a9f2494e4c66fd513fb10
Reviewed-on: https://go-review.googlesource.com/85878
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: replace mlookup and findObject with heapBitsForObject
Austin Clements [Mon, 4 Dec 2017 15:43:11 +0000 (10:43 -0500)]
runtime: replace mlookup and findObject with heapBitsForObject

These functions all serve essentially the same purpose. mlookup is
used in only one place and findObject in only three. Use
heapBitsForObject instead, which is the most optimized implementation.

(This may seem slightly silly because none of these uses care about
the heap bits, but we're about to split up the functionality of
heapBitsForObject anyway. At that point, findObject will rise from the
ashes.)

Change-Id: I906468c972be095dd23cf2404a7d4434e802f250
Reviewed-on: https://go-review.googlesource.com/85877
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: validate lfnode addresses
Austin Clements [Mon, 4 Dec 2017 00:28:28 +0000 (19:28 -0500)]
runtime: validate lfnode addresses

Change-Id: Ic8c506289caaf6218494e5150d10002e0232feaa
Reviewed-on: https://go-review.googlesource.com/85876
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agoruntime: expand/update lfstack address space assumptions
Austin Clements [Sun, 3 Dec 2017 23:08:57 +0000 (18:08 -0500)]
runtime: expand/update lfstack address space assumptions

I was spelunking Linux's address space code and found that some of the
information about maximum virtual addresses in lfstack's comments was
out of date. This expands and updates the comment.

Change-Id: I9f54b23e6b266b3c5cc20259a849231fb751f6e7
Reviewed-on: https://go-review.googlesource.com/85875
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agocmd/compile: improve absorb shifts optimization for arm64
Chad Rosier [Thu, 15 Feb 2018 19:49:03 +0000 (14:49 -0500)]
cmd/compile: improve absorb shifts optimization for arm64

Current absorb shifts optimization can generate dead Value nodes which increase
use count of other live nodes. It will impact other optimizations (such as
combined loads) which are enabled based on specific use count. This patch fixes
the issue by decreasing the use count of nodes referenced by dead Value nodes
generated by absorb shifts optimization.

Performance impacts on go1 benchmarks (data collected on A57@2GHzx8):

name                     old time/op    new time/op    delta
BinaryTree17-8              6.28s ± 2%     6.24s ± 1%     ~     (p=0.065 n=10+9)
Fannkuch11-8                6.32s ± 0%     6.33s ± 0%   +0.17%  (p=0.000 n=10+10)
FmtFprintfEmpty-8          98.9ns ± 0%    99.2ns ± 0%   +0.34%  (p=0.000 n=9+7)
FmtFprintfString-8          183ns ± 1%     182ns ± 1%   -1.01%  (p=0.005 n=9+10)
FmtFprintfInt-8             199ns ± 1%     202ns ± 1%   +1.41%  (p=0.000 n=10+9)
FmtFprintfIntInt-8          272ns ± 1%     276ns ± 3%   +1.36%  (p=0.015 n=10+10)
FmtFprintfPrefixedInt-8     367ns ± 1%     369ns ± 1%   +0.68%  (p=0.042 n=10+10)
FmtFprintfFloat-8           491ns ± 1%     493ns ± 1%     ~     (p=0.064 n=10+10)
FmtManyArgs-8              1.31µs ± 1%    1.32µs ± 1%   +0.39%  (p=0.042 n=8+9)
GobDecode-8                17.0ms ± 2%    16.2ms ± 2%   -4.74%  (p=0.000 n=10+10)
GobEncode-8                13.7ms ± 2%    13.4ms ± 1%   -2.40%  (p=0.000 n=10+9)
Gzip-8                      844ms ± 0%     737ms ± 0%  -12.70%  (p=0.000 n=10+10)
Gunzip-8                   84.4ms ± 1%    83.9ms ± 0%   -0.55%  (p=0.000 n=10+8)
HTTPClientServer-8          122µs ± 1%     124µs ± 1%   +1.75%  (p=0.000 n=10+9)
JSONEncode-8               34.9ms ± 1%    32.4ms ± 0%   -7.11%  (p=0.000 n=10+9)
JSONDecode-8                150ms ± 0%     146ms ± 1%   -2.84%  (p=0.000 n=7+10)
Mandelbrot200-8            10.0ms ± 0%    10.0ms ± 0%     ~     (p=0.529 n=10+10)
GoParse-8                  8.18ms ± 1%    8.03ms ± 0%   -1.93%  (p=0.000 n=10+10)
RegexpMatchEasy0_32-8       209ns ± 0%     209ns ± 0%     ~     (p=0.248 n=10+9)
RegexpMatchEasy0_1K-8       789ns ± 1%     790ns ± 0%     ~     (p=0.361 n=10+10)
RegexpMatchEasy1_32-8       202ns ± 0%     202ns ± 1%     ~     (p=0.137 n=8+10)
RegexpMatchEasy1_1K-8      1.12µs ± 2%    1.12µs ± 1%     ~     (p=0.810 n=10+10)
RegexpMatchMedium_32-8      298ns ± 0%     298ns ± 0%     ~     (p=0.443 n=10+9)
RegexpMatchMedium_1K-8     83.0µs ± 5%    78.6µs ± 0%   -5.37%  (p=0.000 n=10+10)
RegexpMatchHard_32-8       4.32µs ± 0%    4.26µs ± 0%   -1.47%  (p=0.000 n=10+10)
RegexpMatchHard_1K-8        132µs ± 4%     126µs ± 0%   -4.41%  (p=0.000 n=10+9)
Revcomp-8                   1.11s ± 0%     1.11s ± 0%   +0.14%  (p=0.017 n=10+9)
Template-8                  155ms ± 1%     155ms ± 1%     ~     (p=0.796 n=10+10)
TimeParse-8                 774ns ± 1%     785ns ± 1%   +1.41%  (p=0.001 n=10+10)
TimeFormat-8                788ns ± 1%     806ns ± 1%   +2.24%  (p=0.000 n=10+9)

name                     old speed      new speed      delta
GobDecode-8              45.2MB/s ± 2%  47.5MB/s ± 2%   +4.96%  (p=0.000 n=10+10)
GobEncode-8              56.0MB/s ± 2%  57.4MB/s ± 1%   +2.44%  (p=0.000 n=10+9)
Gzip-8                   23.0MB/s ± 0%  26.3MB/s ± 0%  +14.55%  (p=0.000 n=10+10)
Gunzip-8                  230MB/s ± 1%   231MB/s ± 0%   +0.55%  (p=0.000 n=10+8)
JSONEncode-8             55.6MB/s ± 1%  59.9MB/s ± 0%   +7.65%  (p=0.000 n=10+9)
JSONDecode-8             12.9MB/s ± 0%  13.3MB/s ± 1%   +2.94%  (p=0.000 n=7+10)
GoParse-8                7.08MB/s ± 1%  7.22MB/s ± 0%   +1.95%  (p=0.000 n=10+10)
RegexpMatchEasy0_32-8     153MB/s ± 0%   153MB/s ± 0%   -0.16%  (p=0.023 n=10+10)
RegexpMatchEasy0_1K-8    1.30GB/s ± 1%  1.30GB/s ± 0%     ~     (p=0.393 n=10+10)
RegexpMatchEasy1_32-8     158MB/s ± 0%   158MB/s ± 0%     ~     (p=0.684 n=10+10)
RegexpMatchEasy1_1K-8     915MB/s ± 2%   918MB/s ± 1%     ~     (p=0.796 n=10+10)
RegexpMatchMedium_32-8   3.35MB/s ± 0%  3.35MB/s ± 0%     ~     (p=1.000 n=10+9)
RegexpMatchMedium_1K-8   12.3MB/s ± 5%  13.0MB/s ± 0%   +5.56%  (p=0.000 n=10+10)
RegexpMatchHard_32-8     7.40MB/s ± 0%  7.51MB/s ± 0%   +1.50%  (p=0.000 n=10+10)
RegexpMatchHard_1K-8     7.75MB/s ± 4%  8.10MB/s ± 0%   +4.52%  (p=0.000 n=10+8)
Revcomp-8                 229MB/s ± 0%   228MB/s ± 0%   -0.14%  (p=0.017 n=10+9)
Template-8               12.5MB/s ± 1%  12.5MB/s ± 1%     ~     (p=0.780 n=10+10)

Change-Id: I103389f168eac79f6af44e8fef93acc2a7a4ac96
Reviewed-on: https://go-review.googlesource.com/88415
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocompiler: honor //line directives in DWARF variable file/line attrs
Than McIntosh [Tue, 6 Feb 2018 14:36:13 +0000 (09:36 -0500)]
compiler: honor //line directives in DWARF variable file/line attrs

During DWARF debug generation, the DW_AT_decl_line / DW_AT_decl_file
attributes for variable DIEs were being computed without taking into
account the possibility of "//line" directives. Fix things up to use
the correct src.Pos methods to pick up this info.

Fixes #23704.

Change-Id: I88c21a0e0a9602392be229252d856a6d665868e2
Reviewed-on: https://go-review.googlesource.com/92255
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agointernal/trace: link user span start and end events
Hana Kim [Tue, 13 Feb 2018 22:18:08 +0000 (17:18 -0500)]
internal/trace: link user span start and end events

Also add testdata for version 1.11 including UserTaskSpan test trace.

Change-Id: I673fb29bb3aee96a14fadc0ab860d4f5832143f5
Reviewed-on: https://go-review.googlesource.com/93795
Reviewed-by: Heschi Kreinick <heschi@google.com>
7 years agocmd/cgo: delete double space in comment
zaq1tomo [Thu, 15 Feb 2018 16:12:24 +0000 (16:12 +0000)]
cmd/cgo: delete double space in comment

delete double space from comment

Change-Id: I71af5c1149941575016f79a91269f128b1fc16af
GitHub-Last-Rev: aba8874bd362d05d6c29c8647049369dfcd796f5
GitHub-Pull-Request: golang/go#23851
Reviewed-on: https://go-review.googlesource.com/94415
Reviewed-by: Ian Lance Taylor <iant@golang.org>