runtime: add bmap.setoverflow

bmap already has a overflow (getter) method.
Add a setoverflow (setter) method, for readability.

Updates #19931
Updates #19992

Change-Id: I00b3d94037c0d75508a7ebd51085c5c3857fb764
Reviewed-on: https://go-review.googlesource.com/40977
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

runtime: convert hmap.overflow into hmap.extra

Any change to how we allocate overflow buckets
will require some extra hmap storage,
but we don't want hmap to grow,
particular as small maps usually don't need overflow buckets.

This CL converts the existing hmap overflow field,
which is usually used for pointer-free maps,
into a generic extra field.

This extra field can be used to hold data that is optional.
If it is valuable enough to do have special
handling of overflow buckets, which are medium-sized,
it is valuable enough to pay an extra alloc and two extra words for.

Adding fields to extra would entail adding overhead to pointer-free maps;
any mapextra fields added would need to be weighed against that.
This CL is just rearrangement, though.

Updates #19931
Updates #19992

Change-Id: If8537a206905b9d4dc6cd9d886184ece671b3f80
Reviewed-on: https://go-review.googlesource.com/40976
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

runtime: refactor hmap setoverflow into newoverflow

This simplifies the code, as well as providing
a single place to modify to change the
allocation of new overflow buckets.

Updates #19931
Updates #19992

Change-Id: I77070619f5c8fe449bbc35278278bca5eda780f2
Reviewed-on: https://go-review.googlesource.com/40975
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

cmd/internal: remove duplicate pathToPrefix function

goobj.importPathToPrefix is 3x faster than gc.pathToPrefix so rename and
move it to cmd/internal/objabi which is already imported by both goobj and
gc.

Change-Id: I10eda5bce95ef6d5d888818c5c47258c2833ea45
Reviewed-on: https://go-review.googlesource.com/40875
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

go/format: quote bytes in comments for readability

Change-Id: I8e90012b3498c51aaeb2f70182debab52549afb4
Reviewed-on: https://go-review.googlesource.com/40930
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile/internal/ppc64: remove unused condOps map

The last use of condOps was removed in c644a76.

Change-Id: I5383d0e7a9078fc17ca12ed032ecf8e7f4aa95d7
Reviewed-on: https://go-review.googlesource.com/41030
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: enhance postorder computation and repair loop finder

Replace derecursed postorder computation with one that
mimics DFS traversal.

Corrected outerinner function in loopfinder

Leave enhanced checks in place.

Change-Id: I657ba5e89c88941028d6d4c72e9f9056e30f1ce8
Reviewed-on: https://go-review.googlesource.com/40872
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@golang.org>

cmd/compile: better variable names in internal/syntax/branches.go

Follow-up on https://go-review.googlesource.com/#/c/39998/.

Change-Id: I97f8e31ca923685198984ad64f952d6dc8208edf
Reviewed-on: https://go-review.googlesource.com/40982
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: report block start for gotos jumping into blocks

Follow-up on https://go-review.googlesource.com/#/c/39998/
which dropped this information.

The reported blocks are the innermost blocks containing a
label jumped to from outside, not the outermost block as
reported originally by cmd/compile.

We could report the outermost block with a slighly more
involved algorithm (need to track containing blocks for
all unresolved forward gotos), but since gccgo also reports
the innermost blocks, the current approach seems good enough.

Change-Id: Ic0235b8fafe8d5f99dc9872b58e90e8d9e72c5db
Reviewed-on: https://go-review.googlesource.com/40980
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

.gitignore: change path to zbootstrap.go

zbootstrap.go was moved in 1e3570a. This updates .gitignore to match.

Change-Id: I7500e0abc8e60b6c36c6ae056b2f68b9d021fc6f
Reviewed-on: https://go-review.googlesource.com/40993
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/compile: remove uses of types.Dclstack - not needed anymore

Follow-up on https://go-review.googlesource.com/#/c/39998/.

Change-Id: I8830eebd7ea7e02b7edda99e67b6d43529401201
Reviewed-on: https://go-review.googlesource.com/40974
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/compile: check labels and branches during parse time

Instead of a separate check control flow pass (checkcfg.go)
operating on nodes, perform this check at parse time on the
new syntax tree. Permits this check to be done concurrently,
and doesn't depend on the specifics of the symbol's dclstack
implementation anymore. The remaining dclstack uses will be
removed in a follow-up change.

- added CheckBranches Mode flag (so we can turn off the check
  if we only care about syntactic correctness, e.g. for tests)

- adjusted test/goto.go error messages: the new branches
  checker only reports if a goto jumps into a block, but not
  which block (we may want to improve this again, eventually)

- also, the new branches checker reports one variable that
  is being jumped over by a goto, but it may not be the first
  one declared (this is fine either way)

- the new branches checker reports additional errors for
  fixedbugs/issue14006.go (not crucial to avoid those errors)

- the new branches checker now correctly reports only
  variable declarations being jumped over, rather than
  all declarations (issue 8042). Added respective tests.

Fixes #8042.

Change-Id: I53b6e1bda189748e1e1fb5b765a8a64337c27d40
Reviewed-on: https://go-review.googlesource.com/39998
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/internal/objabi: extract shared functionality from obj

Now only cmd/asm and cmd/compile depend on cmd/internal/obj. Changing
the assembler backends no longer requires reinstalling cmd/link or
cmd/addr2line.

There's also now one canonical definition of the object file format in
cmd/internal/objabi/doc.go, with a warning to update all three
implementations.

objabi is still something of a grab bag of unrelated code (e.g., flag
and environment variable handling probably belong in a separate "tool"
package), but this is still progress.

Fixes #15165.
Fixes #20026.

Change-Id: Ic4b92fac7d0d35438e0d20c9579aad4085c5534c
Reviewed-on: https://go-review.googlesource.com/40972
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

cmd/cgo: read source files once

Now cgo reads source files twice: for c prefix generation and parsing
go code to an ast node. It can be narrowed down to single loop.

Change-Id: Ie05452a3a12106aaab863244727390037e69e8e6
Reviewed-on: https://go-review.googlesource.com/40939
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

cmd/vet: make tests safe for perl 5.26

Someday we should write errchk in Go. But not today.

Fixes #20007

Change-Id: I61ccf3cfaa66a710782f8a2212a6a2b4040698da
Reviewed-on: https://go-review.googlesource.com/40950
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>

cmd/cgo: reject references to builtin functions other than calls

Here we restrict using cgo builtin references because internally they're go functions
as opposed to C usafe.Pointer values.

Fixes #18889

Change-Id: I1e4332e4884063ccbaf9772c172d4462ec8f3d13
Reviewed-on: https://go-review.googlesource.com/40934
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: replace Segment's linked list of Sections with a slice

Just noticed this in passing.

Change-Id: I58fa828ef58598209ed4cbe4abc6f9f02ffc4844
Reviewed-on: https://go-review.googlesource.com/40896
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

io: correctly compute call depth in test

TestMultiReaderFlatten determines the call depth by counting PCs
returned by runtime.Callers. With inlining, this is incorrect because
a PC can represent multiple calls. Furthermore, runtime.Callers might
return an additional "skip" PC, which does not represent a real call.
This modifies the test to use CallersFrames to determine the call depth.
Now the test passes with -l=4.

Change-Id: I284f3b1e0b2d194bd08c230c616914503e5a370d
Reviewed-on: https://go-review.googlesource.com/40990
Run-TryBot: David Lazar <lazard@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: fix TestCaller with -l=4

Only the noinline pragma on testCallerFoo is needed to pass the test,
but the second pragma makes the test robust to future changes to the
inliner.

Change-Id: I80b384380c598f52e0382f53b59bb47ff196363d
Reviewed-on: https://go-review.googlesource.com/40877
Run-TryBot: David Lazar <lazard@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: make example independent of inlining

Otherwise, with -l=4, runtime.Callers gets inlined and the example
prints too many frames. Now the example passes with -l=4.

Change-Id: I9e420af9371724ac3ec89efafd76a658cf82bb4a
Reviewed-on: https://go-review.googlesource.com/40876
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

runtime: skip logical frames in runtime.Caller

This rewrites runtime.Caller in terms of stackExpander, which already
handles inlined frames and partially skipped frames. This also has the
effect of making runtime.Caller understand cgo frames if there is a cgo
symbolizer.

Updates #19348.

Change-Id: Icdf4df921aab5aa394d4d92e3becc4dd169c9a6e
Reviewed-on: https://go-review.googlesource.com/40270
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

encoding/hex: change lookup table back to string

CL 27254 changed hextable to a byte array for performance.
CL 28219 fixed the compiler so that that is no longer necessary.
As Kirill notes in #15808, a string is preferable
as the linker can easily de-dup it.
So go back. No performance changes.

Change-Id: Ibef7d21d0f2507968a0606602c5dd57ed4a85b1b
Reviewed-on: https://go-review.googlesource.com/40970
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj: un-embed FuncInfo field in LSym

Automated refactoring using github.com/mdempsky/unbed (to rewrite
s.Foo to s.FuncInfo.Foo) and then gorename (to rename the FuncInfo
field to just Func).

Passes toolstash-check -all.

Change-Id: I802c07a1239e0efea058a91a87c5efe12170083a
Reviewed-on: https://go-review.googlesource.com/40670
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

cmd/compile: eliminate dead code in if statements after typechecking

This is a more thorough and cleaner fix
than doing dead code elimination separately
during inlining, escape analysis, and export.

Unfortunately, it does add another full walk of the AST.
The performance impact is very small, but not non-zero.

If a label or goto is present in the dead code, it is not eliminated.
This restriction can be removed once label/goto checking occurs
much earlier in the compiler. In practice, it probably doesn't
matter much.

Updates #19699
Fixes #19705

name       old alloc/op      new alloc/op      delta
Template        39.2MB ± 0%       39.3MB ± 0%  +0.28%  (p=0.008 n=5+5)
Unicode         29.8MB ± 0%       29.8MB ± 0%    ~     (p=1.000 n=5+5)
GoTypes          113MB ± 0%        113MB ± 0%  -0.55%  (p=0.008 n=5+5)
SSA             1.25GB ± 0%       1.25GB ± 0%  +0.02%  (p=0.008 n=5+5)
Flate           25.3MB ± 0%       25.3MB ± 0%  -0.24%  (p=0.032 n=5+5)
GoParser        31.7MB ± 0%       31.8MB ± 0%  +0.31%  (p=0.008 n=5+5)
Reflect         78.2MB ± 0%       78.3MB ± 0%    ~     (p=0.421 n=5+5)
Tar             26.6MB ± 0%       26.7MB ± 0%  +0.21%  (p=0.008 n=5+5)
XML             42.2MB ± 0%       42.2MB ± 0%    ~     (p=0.056 n=5+5)

name       old allocs/op     new allocs/op     delta
Template          385k ± 0%         387k ± 0%  +0.51%  (p=0.016 n=5+5)
Unicode           321k ± 0%         321k ± 0%    ~     (p=1.000 n=5+5)
GoTypes          1.14M ± 0%        1.14M ± 0%    ~     (p=1.000 n=5+5)
SSA              9.71M ± 0%        9.72M ± 0%  +0.10%  (p=0.008 n=5+5)
Flate             234k ± 1%         234k ± 1%    ~     (p=0.690 n=5+5)
GoParser          315k ± 0%         317k ± 0%  +0.71%  (p=0.008 n=5+5)
Reflect           980k ± 0%         983k ± 0%  +0.30%  (p=0.032 n=5+5)
Tar               251k ± 0%         252k ± 0%  +0.55%  (p=0.016 n=5+5)
XML               392k ± 0%         393k ± 0%  +0.30%  (p=0.008 n=5+5)

Change-Id: Ia10ff4bbf5c6eae782582cc9cbc9785494d4fb83
Reviewed-on: https://go-review.googlesource.com/38773
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

reflect: document ptrdata field of rtype

This CL adds a simple explanation about what means the ptrdata field of
the reflect.rtype type.
Also document that rtype needs to be kept in sync with the runtime._type
type that rtype mirrors.

Change-Id: Icd9663a2e4bb94d922a2417cfe4537861d2ccc97
Reviewed-on: https://go-review.googlesource.com/40917
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

all: remove unnecessary ", _" from map reads

If the bool value isn't used, there is no need to assign to underscore -
there is a shorter form that only returns the value and behaves in the
exact same way.

Change-Id: Iaf801b8e966da6c2f565bc39e3bb028175c92d60
Reviewed-on: https://go-review.googlesource.com/40920
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

sync: improve Pool performance

Rewrite indexLocal to achieve higher performance.

Performance results on linux/amd64:

name            old time/op  new time/op  delta
Pool-4          19.1ns ± 2%  10.1ns ± 1%  -47.15%  (p=0.000 n=10+8)
PoolOverflow-4  3.11µs ± 1%  2.10µs ± 2%  -32.66%  (p=0.000 n=10+10)

Performance results on linux/386:

name            old time/op  new time/op  delta
Pool-4          20.0ns ± 2%  13.1ns ± 1%  -34.59%  (p=0.000 n=10+9)
PoolOverflow-4  3.51µs ± 1%  2.49µs ± 0%  -28.99%  (p=0.000 n=10+8)

Change-Id: I7d57a2d4cd47ec43d09ca1267bde2e3f05a9faa9
Reviewed-on: https://go-review.googlesource.com/40913
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: remove stray blank assignment

Left over from CL 39855.

Change-Id: I9df8b5c631d5afbdbf2fb306876648d8541931d3
Reviewed-on: https://go-review.googlesource.com/40941
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

net, internal/poll, net/internal/socktest: use accept4 and SOCK_{CLOEXEC,NONBLOCK} on dragonfly

Fixes #14222

Change-Id: I026fc9499fdefc33b8bb58b5963e2290adacbf63
Reviewed-on: https://go-review.googlesource.com/40895
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

syscall: add Accept4 and SOCK_{CLOEXEC,NONBLOCK} on dragonfly

This change just picks a few constants from DragonfFly BSD 4.6 kernel
and doesn't synchronize all the existing constants with the latest
DragonFly BSD kernels.

Updates #14222.

Change-Id: Ie107a8bee1a09393b3b42b6f82489532f5d13290
Reviewed-on: https://go-review.googlesource.com/40894
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: look up more runtime symbols before SSA begins

This avoids concurrent runtime package lookups.

Updates #15756

Change-Id: I9e2cbd042aba44923f0d03e6ca5b4eb60fa9e7ea
Reviewed-on: https://go-review.googlesource.com/40853
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj: unexport Link.Hash

A prior CL eliminated the last reference to Ctxt.Hash
from the compiler.

Change-Id: Ic97ff84ed1a14e0c93fb0e8ec0b2617c3397c0e8
Reviewed-on: https://go-review.googlesource.com/40699
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: relocate large comment to plive.go

It was a bit weird to have it at the top of pgen.go.
This does half of the TODO at the top of the comment.

Change-Id: I65140fa05673b2dbb6feddb8c1877f6d624a7844
Reviewed-on: https://go-review.googlesource.com/40698
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/internal/obj: rework gclocals handling

The compiler handled gcargs and gclocals LSyms unusually.
It generated placeholder symbols (makefuncdatasym),
filled them in, and then renamed them for content-addressability.
This is an important binary size optimization;
the same locals information occurs over and over.

This CL continues to treat these LSyms unusually,
but in a slightly more explicit way,
and importantly for concurrent compilation,
in a way that does not require concurrent
modification of Ctxt.Hash.

Instead of creating gcargs and gclocals in the usual way,
by creating a types.Sym and then an obj.LSym,
we add them directly to obj.FuncInfo,
initialize them in obj.InitTextSym,
and deduplicate and add them to ctxt.Data at the end.
Then the backend's job is simply to fill them in
and rename them appropriately.

Updates #15756

name       old alloc/op      new alloc/op      delta
Template        38.8MB ± 0%       38.7MB ± 0%  -0.22%  (p=0.016 n=5+5)
Unicode         29.8MB ± 0%       29.8MB ± 0%    ~     (p=0.690 n=5+5)
GoTypes          113MB ± 0%        113MB ± 0%  -0.24%  (p=0.008 n=5+5)
SSA             1.25GB ± 0%       1.24GB ± 0%  -0.39%  (p=0.008 n=5+5)
Flate           25.3MB ± 0%       25.2MB ± 0%  -0.43%  (p=0.008 n=5+5)
GoParser        31.7MB ± 0%       31.7MB ± 0%  -0.22%  (p=0.008 n=5+5)
Reflect         78.2MB ± 0%       77.6MB ± 0%  -0.80%  (p=0.008 n=5+5)
Tar             26.6MB ± 0%       26.3MB ± 0%  -0.85%  (p=0.008 n=5+5)
XML             42.4MB ± 0%       41.9MB ± 0%  -1.04%  (p=0.008 n=5+5)

name       old allocs/op     new allocs/op     delta
Template          378k ± 0%         377k ± 1%    ~     (p=0.151 n=5+5)
Unicode           321k ± 1%         321k ± 0%    ~     (p=0.841 n=5+5)
GoTypes          1.14M ± 0%        1.14M ± 0%  -0.47%  (p=0.016 n=5+5)
SSA              9.71M ± 0%        9.67M ± 0%  -0.33%  (p=0.008 n=5+5)
Flate             233k ± 1%         232k ± 1%    ~     (p=0.151 n=5+5)
GoParser          316k ± 0%         315k ± 0%  -0.49%  (p=0.016 n=5+5)
Reflect           979k ± 0%         972k ± 0%  -0.75%  (p=0.008 n=5+5)
Tar               250k ± 0%         247k ± 1%  -0.92%  (p=0.008 n=5+5)
XML               392k ± 1%         389k ± 0%  -0.67%  (p=0.008 n=5+5)

Change-Id: Idc36186ca9d2f8214b5f7720bbc27b6bb22fdc48
Reviewed-on: https://go-review.googlesource.com/40697
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/compile: implement non-constant rotates

Makes math/bits.Rotate{Left,Right} fast on amd64.

name              old time/op  new time/op  delta
RotateLeft-12     7.42ns ± 6%  5.45ns ± 6%  -26.54%   (p=0.000 n=9+10)
RotateLeft8-12    4.77ns ± 5%  3.42ns ± 7%  -28.25%   (p=0.000 n=8+10)
RotateLeft16-12   4.82ns ± 8%  3.40ns ± 7%  -29.36%  (p=0.000 n=10+10)
RotateLeft32-12   4.87ns ± 7%  3.48ns ± 7%  -28.51%    (p=0.000 n=8+9)
RotateLeft64-12   5.23ns ±10%  3.35ns ± 6%  -35.97%   (p=0.000 n=9+10)
RotateRight-12    7.59ns ± 8%  5.71ns ± 1%  -24.72%   (p=0.000 n=10+8)
RotateRight8-12   4.98ns ± 7%  3.36ns ± 9%  -32.55%  (p=0.000 n=10+10)
RotateRight16-12  5.12ns ± 2%  3.45ns ± 5%  -32.62%  (p=0.000 n=10+10)
RotateRight32-12  4.80ns ± 6%  3.42ns ±16%  -28.68%  (p=0.000 n=10+10)
RotateRight64-12  4.78ns ± 6%  3.42ns ± 6%  -28.50%  (p=0.000 n=10+10)

Update #18940

Change-Id: Ie79fb5581c489ed4d3b859314c5e669a134c119b
Reviewed-on: https://go-review.googlesource.com/39711
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

runtime: make internal CallersFrames-equivalent that doesn't escape PC slice

The Frames API forces the PC slice to escape to the heap because it
stores it in the Frames object. However, we'd like to use this API for
call stack expansion internally in the runtime in places where it
would be very good to avoid heap allocation.

This commit makes this possible by pulling the bulk of the Frames
implementation into an internal frameExpander API. The key difference
between these APIs is that the frameExpander does not hold the PC
slice; instead, the caller is responsible for threading the PC slice
through the frameExpander API calls. This makes it possible to keep
the PC slice on the stack. The Frames API then becomes a thin shim
around the frameExpander that keeps the PC slice in the Frames object.

Change-Id: If6b2d0b9132a2a905a0cf5deced9feddce76fc0e
Reviewed-on: https://go-review.googlesource.com/40610
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Lazar <lazard@golang.org>

net/http: ignore extra space between response version and status code

Reading a response with a status line like "HTTP/1.0  401 Unauthorized"
(with two spaces after the version) has been returning an error. Now the
extra space will be ignored.

Fixes #19989

Change-Id: I0c88a6ef7562ba80e2e2635be2070dd1b5b671a7
Reviewed-on: https://go-review.googlesource.com/40933
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: only include the version info and export data in ABI hash

Previously the "ABI hash" for a package (used to determine if a loaded shared
library has the ABI expected by its loader) was the hash of the entire
__.PKGDEF file. But that means it depends on the build ID generated by the go
tool for the package, which means that if a file is added (even a .c or .h
file!) to the package, the ABI changes, perhaps uncessarily.

Fixes #19920

Change-Id: If919481e1a03afb350c8a9c7a0666bb90ee90270
Reviewed-on: https://go-review.googlesource.com/40401
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

spec: clarify use of fused-floating point operations

Added a paragraph and examples explaining when an implementation
may use fused floating-point operations (such as FMA) and how to
prevent operation fusion.

For #17895.

Change-Id: I64c9559fc1097e597525caca420cfa7032d67014
Reviewed-on: https://go-review.googlesource.com/40391
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/asm: detect invalid DS form offsets for ppc64x

While debugging a recent regression it was discovered that
the assembler for ppc64x was not always generating the correct
instruction for DS form loads and stores.  When an instruction
is DS form then the offset must be a multiple of 4, and if it
isn't then bits outside the offset field were being incorrectly
set resulting in unexpected and incorrect instructions.

This change adds a check to determine when the opcode is DS form
and then verifies that the offset is a multiple of 4 before
generating the instruction, otherwise logs an error.

This also changes a few asm files that were using unaligned offsets
for DS form loads and stores.  In the runtime package these were
instructions intended to cause a crash so using aligned or unaligned
offsets doesn't change that behavior.

Change-Id: Ie3a7e1e65dcc9933b54de7a46a054da8459cb56f
Reviewed-on: https://go-review.googlesource.com/40476
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>

syscall: make TestFcntlFlock more robust

Avoid the use of constant absolute temp files in tests.  This could
produce flaky results, for example on multiuser development machines.

Change-Id: Ia76157a0660fbe294bb31a46ded886cea5deec97
Reviewed-on: https://go-review.googlesource.com/40916
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

misc/ios: fix the Test386EndToEnd test on iOS

Some tests need the src/runtime/textflag.h file. Make sure it is
included in iOS test runs.

Change-Id: I5e0e7ebe85679686ef15a7d336f28ac9b68a587a
Reviewed-on: https://go-review.googlesource.com/40915
Run-TryBot: Elias Naur <elias.naur@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

io/ioutil: make TestTempFile more robust

The first part of this test tries to confirm that we can't create
a TempFile in a non-existent directory, but does not ensure that
the non-existent directory really does not exist.  Instead, let's
create an empty temp directory, and use a non-existent subdir of
that.

Change-Id: I176f14ed5f5a2d7a8c29d8f6949755db69d7dbb6
Reviewed-on: https://go-review.googlesource.com/40914
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

doc: dissuade people from using PGP for security reports

Change-Id: I7e4f22a2b6c80dd0787c011703f3f8586ff55a50
Reviewed-on: https://go-review.googlesource.com/40860
Reviewed-by: Chris Broadfoot <cbro@golang.org>

cmd/compile: remove underscore from variable names in constant comparison test

The leading underscores aren't necessary.

Change-Id: I1d4c4b681e2a29ef40a0a6cf705c3b17a49c9f65
Reviewed-on: https://go-review.googlesource.com/40873
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/go/internal/work: clean up after TestRespectGroupSticky

Use our own tempdir, to avoid having to Init (and somehow teardown)
Builder.  This way we don't leave behind any temp files.

Also, don't create a hardcoded path inside a testcase.

Followup to golang/go#18878.
Fixes golang/go#19449.

Change-Id: Ieb1ebeab24ae8a74a6fa058d9c23f72b3fc1c444
Reviewed-on: https://go-review.googlesource.com/40912
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/asm, cmd/internal/obj/s390x, math: add LGDR and LDGR instructions

The instructions allow moves between floating point and general
purpose registers without any conversion taking place.

Change-Id: I82c6f3ad9c841a83783b5be80dcf5cd538ff49e6
Reviewed-on: https://go-review.googlesource.com/38777
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

cmd/go: don't generate DWARF in the compiler for plan9

Follow-up to CL 40859.

Change-Id: I785b8b28f179a8f2fcd42b0a97ffcd41de4b21bc
Reviewed-on: https://go-review.googlesource.com/40865
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/go: disable compiler DWARF generation on some platforms

The linker disables DWARF for these platforms.
Given that, we can spare the compiler some work.

Change-Id: Ic5a6b675150aca199bdc1dd2cdf5eeb4e215bdff
Reviewed-on: https://go-review.googlesource.com/40859
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/go: don't generate DWARF in the compiler if the linker will discard it

Change-Id: I7610f968f5c22b51a6a8813d6f31ccb772a12ba9
Reviewed-on: https://go-review.googlesource.com/40858
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: add flag to disable DWARF generation

DWARF generation has non-trivial cost,
and in some cases is not necessary.
Provide an option to opt out.

Alloc impact of disabling DWARF generation:

name       old alloc/op      new alloc/op      delta
Template        38.7MB ± 0%       37.6MB ± 0%  -2.77%  (p=0.016 n=5+4)
Unicode         29.8MB ± 0%       29.8MB ± 0%  -0.16%  (p=0.032 n=5+5)
GoTypes          113MB ± 0%        110MB ± 0%  -2.38%  (p=0.008 n=5+5)
Compiler         463MB ± 0%        457MB ± 0%  -1.34%  (p=0.008 n=5+5)
SSA             1.25GB ± 0%       1.23GB ± 0%  -1.64%  (p=0.008 n=5+5)
Flate           25.3MB ± 0%       25.0MB ± 0%  -1.05%  (p=0.008 n=5+5)
GoParser        31.7MB ± 0%       30.9MB ± 0%  -2.74%  (p=0.008 n=5+5)
Reflect         78.2MB ± 0%       76.7MB ± 0%  -1.90%  (p=0.008 n=5+5)
Tar             26.5MB ± 0%       26.0MB ± 0%  -2.04%  (p=0.008 n=5+5)
XML             42.4MB ± 0%       41.1MB ± 0%  -2.86%  (p=0.008 n=5+5)

name       old allocs/op     new allocs/op     delta
Template          377k ± 0%         360k ± 1%  -4.46%  (p=0.008 n=5+5)
Unicode           321k ± 0%         320k ± 0%    ~     (p=0.151 n=5+5)
GoTypes          1.14M ± 0%        1.10M ± 0%  -4.13%  (p=0.008 n=5+5)
Compiler         4.26M ± 0%        4.13M ± 0%  -3.14%  (p=0.008 n=5+5)
SSA              9.70M ± 0%        9.33M ± 0%  -3.89%  (p=0.008 n=5+5)
Flate             233k ± 0%         228k ± 0%  -2.40%  (p=0.008 n=5+5)
GoParser          316k ± 0%         302k ± 0%  -4.48%  (p=0.008 n=5+5)
Reflect           980k ± 0%         945k ± 0%  -3.62%  (p=0.008 n=5+5)
Tar               249k ± 0%         241k ± 0%  -3.19%  (p=0.008 n=5+5)
XML               391k ± 0%         376k ± 0%  -3.95%  (p=0.008 n=5+5)

Change-Id: I97dbfb6b40195d1e0b91be097a4bf0e7f65b26af
Reviewed-on: https://go-review.googlesource.com/40857
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj: unify Setuintxx and WriteInt

They do basically the same work.

Setuintxx was only used in a single place,
so eliminate it in favor of WriteInt.

duintxxLSym's alignment rounding was not used in practice;
change it into alignment assertion.

Passes toolstash-check. No compiler performance changes.

Change-Id: I0f7410cf2ccffbdc02ad796eaf973ee6a83074f8
Reviewed-on: https://go-review.googlesource.com/40863
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj: pretty-print LSym.Type when debugging

We have a stringer for LSym.Type. Use it.

Before:

"".algarray t=31 size=224

After:

"".algarray SBSS size=224

Change-Id: Ib4c7d2bc1dbe9943cf2a5dfa5d9f2d7fbd50b7f2
Reviewed-on: https://go-review.googlesource.com/40862
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: improve output when TestAssembly build fails

Change-Id: Ibee84399d81463d3e7d5319626bb0d6b60b86bd9
Reviewed-on: https://go-review.googlesource.com/40861
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

regexp: Use []bool instead of map[uint32]bool in makeOnePass

Variable m is accessed like variable onePassRunes.
So, we can use array instead of map for variable m.

Onepass compile performance is improved 10~20%.

name                                      old time/op    new time/op    delta
CompileOnepass/^(?:(?:(?:.(?:$))?))...-4    4.60µs ± 1%    4.10µs ± 1%  -10.98%  (p=0.008 n=5+5)
CompileOnepass/^abcd$-4                     3.76µs ± 2%    3.33µs ± 1%  -11.43%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a{0,})*?)$-4          4.47µs ± 1%    4.14µs ± 1%   -7.49%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a+)*)$-4              4.07µs ± 1%    3.81µs ± 2%   -6.60%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a|(?:aa)))$-4         5.21µs ± 1%    4.80µs ± 2%   -7.90%  (p=0.008 n=5+5)
CompileOnepass/^(?:[^\s\S])$-4              3.26µs ± 1%    3.06µs ± 1%   -6.15%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:(?:a*)+))$-4          4.33µs ± 0%    4.04µs ± 3%   -6.78%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]+$-4                   3.33µs ± 1%    3.06µs ± 4%   -8.24%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]*$-4                   3.69µs ± 1%    3.39µs ± 4%   -8.26%  (p=0.008 n=5+5)
CompileOnepass/^(?:a*)$-4                   3.78µs ± 2%    3.36µs ± 0%  -11.09%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:aa)|a)$-4             5.11µs ± 1%    4.60µs ± 1%   -9.85%  (p=0.008 n=5+5)
CompileOnepass/^...$-4                      3.51µs ± 1%    3.25µs ± 1%   -7.37%  (p=0.008 n=5+5)
CompileOnepass/^(?:a|(?:aa))$-4             5.05µs ± 1%    4.59µs ± 1%   -9.13%  (p=0.008 n=5+5)
CompileOnepass/^a((b))c$-4                  5.24µs ± 1%    4.66µs ± 1%  -11.16%  (p=0.008 n=5+5)
CompileOnepass/^a.[l-nA-Cg-j]?e$-4          5.94µs ± 7%    5.02µs ± 1%  -15.62%  (p=0.008 n=5+5)
CompileOnepass/^a((b))$-4                   5.34µs ± 7%    4.21µs ± 2%  -21.05%  (p=0.008 n=5+5)
CompileOnepass/^a(?:(b)|(c))c$-4            7.25µs ± 7%    5.86µs ± 1%  -19.17%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b|c)$-4                 4.13µs ± 9%    3.35µs ± 1%  -18.99%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c)$-4                6.48µs ± 4%    5.05µs ± 1%  -22.16%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c+)$-4               7.05µs ± 7%    5.48µs ± 1%  -22.21%  (p=0.008 n=5+5)
CompileOnepass/^a(?:bc)+$-4                 4.71µs ± 4%    3.85µs ± 0%  -18.14%  (p=0.008 n=5+5)
CompileOnepass/^a(?:[bcd])+$-4              4.10µs ± 2%    3.69µs ± 1%   -9.98%  (p=0.008 n=5+5)
CompileOnepass/^a((?:[bcd])+)$-4            5.20µs ± 1%    4.59µs ± 0%  -11.72%  (p=0.008 n=5+5)
CompileOnepass/^a(:?b|c)*d$-4               9.29µs ± 2%    8.23µs ± 1%  -11.43%  (p=0.008 n=5+5)
CompileOnepass/^.bc(d|e)*$-4                6.33µs ± 3%    5.51µs ± 5%  -12.95%  (p=0.008 n=5+5)
CompileOnepass/^loooooooooooooooooo...-4     162µs ± 0%     138µs ± 2%  -14.70%  (p=0.008 n=5+5)

name                                      old alloc/op   new alloc/op   delta
CompileOnepass/^(?:(?:(?:.(?:$))?))...-4    3.38kB ± 0%    3.27kB ± 0%   -3.08%  (p=0.008 n=5+5)
CompileOnepass/^abcd$-4                     2.74kB ± 0%    2.59kB ± 0%   -5.54%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a{0,})*?)$-4          3.34kB ± 0%    3.23kB ± 0%   -3.12%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a+)*)$-4              2.95kB ± 0%    2.85kB ± 0%   -3.52%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a|(?:aa)))$-4         3.74kB ± 0%    3.58kB ± 0%   -4.07%  (p=0.008 n=5+5)
CompileOnepass/^(?:[^\s\S])$-4              2.45kB ± 0%    2.35kB ± 0%   -4.20%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:(?:a*)+))$-4          3.13kB ± 0%    3.02kB ± 0%   -3.32%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]+$-4                   2.48kB ± 0%    2.38kB ± 0%   -4.19%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]*$-4                   2.52kB ± 0%    2.42kB ± 0%   -4.13%  (p=0.008 n=5+5)
CompileOnepass/^(?:a*)$-4                   2.63kB ± 0%    2.53kB ± 0%   -3.95%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:aa)|a)$-4             3.62kB ± 0%    3.47kB ± 0%   -4.19%  (p=0.008 n=5+5)
CompileOnepass/^...$-4                      2.87kB ± 0%    2.78kB ± 0%   -3.34%  (p=0.008 n=5+5)
CompileOnepass/^(?:a|(?:aa))$-4             3.62kB ± 0%    3.47kB ± 0%   -4.19%  (p=0.008 n=5+5)
CompileOnepass/^a((b))c$-4                  4.38kB ± 0%    4.23kB ± 0%   -3.33%  (p=0.008 n=5+5)
CompileOnepass/^a.[l-nA-Cg-j]?e$-4          4.30kB ± 0%    4.15kB ± 0%   -3.35%  (p=0.008 n=5+5)
CompileOnepass/^a((b))$-4                   4.05kB ± 0%    3.90kB ± 0%   -3.56%  (p=0.008 n=5+5)
CompileOnepass/^a(?:(b)|(c))c$-4            5.30kB ± 0%    5.05kB ± 0%   -4.83%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b|c)$-4                 2.87kB ± 0%    2.77kB ± 0%   -3.62%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c)$-4                4.35kB ± 0%    4.21kB ± 0%   -3.31%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c+)$-4               4.58kB ± 0%    4.44kB ± 0%   -3.14%  (p=0.008 n=5+5)
CompileOnepass/^a(?:bc)+$-4                 3.15kB ± 0%    3.00kB ± 0%   -4.82%  (p=0.008 n=5+5)
CompileOnepass/^a(?:[bcd])+$-4              2.94kB ± 0%    2.84kB ± 0%   -3.53%  (p=0.008 n=5+5)
CompileOnepass/^a((?:[bcd])+)$-4            4.08kB ± 0%    3.94kB ± 0%   -3.53%  (p=0.008 n=5+5)
CompileOnepass/^a(:?b|c)*d$-4               6.10kB ± 0%    5.85kB ± 0%   -4.19%  (p=0.008 n=5+5)
CompileOnepass/^.bc(d|e)*$-4                4.46kB ± 0%    4.31kB ± 0%   -3.28%  (p=0.008 n=5+5)
CompileOnepass/^loooooooooooooooooo...-4     135kB ± 0%     127kB ± 0%   -5.88%  (p=0.008 n=5+5)

name                                      old allocs/op  new allocs/op  delta
CompileOnepass/^(?:(?:(?:.(?:$))?))...-4      47.0 ± 0%      46.0 ± 0%   -2.13%  (p=0.008 n=5+5)
CompileOnepass/^abcd$-4                       41.0 ± 0%      40.0 ± 0%   -2.44%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a{0,})*?)$-4            49.0 ± 0%      48.0 ± 0%   -2.04%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a+)*)$-4                44.0 ± 0%      43.0 ± 0%   -2.27%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:a|(?:aa)))$-4           54.0 ± 0%      53.0 ± 0%   -1.85%  (p=0.008 n=5+5)
CompileOnepass/^(?:[^\s\S])$-4                33.0 ± 0%      32.0 ± 0%   -3.03%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:(?:a*)+))$-4            46.0 ± 0%      45.0 ± 0%   -2.17%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]+$-4                     36.0 ± 0%      35.0 ± 0%   -2.78%  (p=0.008 n=5+5)
CompileOnepass/^[a-c]*$-4                     41.0 ± 0%      40.0 ± 0%   -2.44%  (p=0.008 n=5+5)
CompileOnepass/^(?:a*)$-4                     42.0 ± 0%      41.0 ± 0%   -2.38%  (p=0.008 n=5+5)
CompileOnepass/^(?:(?:aa)|a)$-4               53.0 ± 0%      52.0 ± 0%   -1.89%  (p=0.008 n=5+5)
CompileOnepass/^...$-4                        39.0 ± 0%      38.0 ± 0%   -2.56%  (p=0.008 n=5+5)
CompileOnepass/^(?:a|(?:aa))$-4               53.0 ± 0%      52.0 ± 0%   -1.89%  (p=0.008 n=5+5)
CompileOnepass/^a((b))c$-4                    53.0 ± 0%      52.0 ± 0%   -1.89%  (p=0.008 n=5+5)
CompileOnepass/^a.[l-nA-Cg-j]?e$-4            56.0 ± 0%      55.0 ± 0%   -1.79%  (p=0.008 n=5+5)
CompileOnepass/^a((b))$-4                     47.0 ± 0%      46.0 ± 0%   -2.13%  (p=0.008 n=5+5)
CompileOnepass/^a(?:(b)|(c))c$-4              65.0 ± 0%      64.0 ± 0%   -1.54%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b|c)$-4                   40.0 ± 0%      39.0 ± 0%   -2.50%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c)$-4                  57.0 ± 0%      56.0 ± 0%   -1.75%  (p=0.008 n=5+5)
CompileOnepass/^a(?:b?|c+)$-4                 63.0 ± 0%      62.0 ± 0%   -1.59%  (p=0.008 n=5+5)
CompileOnepass/^a(?:bc)+$-4                   46.0 ± 0%      45.0 ± 0%   -2.17%  (p=0.008 n=5+5)
CompileOnepass/^a(?:[bcd])+$-4                43.0 ± 0%      42.0 ± 0%   -2.33%  (p=0.008 n=5+5)
CompileOnepass/^a((?:[bcd])+)$-4              49.0 ± 0%      48.0 ± 0%   -2.04%  (p=0.008 n=5+5)
CompileOnepass/^a(:?b|c)*d$-4                  101 ± 0%       100 ± 0%   -0.99%  (p=0.008 n=5+5)
CompileOnepass/^.bc(d|e)*$-4                  60.0 ± 0%      59.0 ± 0%   -1.67%  (p=0.008 n=5+5)
CompileOnepass/^loooooooooooooooooo...-4     1.09k ± 0%     1.08k ± 0%   -0.74%  (p=0.008 n=5+5)

Fixes #19984

Change-Id: I2268b28d068926a057c62751528de15b6de61a7b
Reviewed-on: https://go-review.googlesource.com/40890
Reviewed-by: Ralph Corderoy <ralph@inputplus.co.uk>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

vendor: update golang_org/x/net/route package

Update the route package to git rev 6b27048a.

Introduce the following changes:
 - 6b27048 route: drop support for go1.5
 - b7fd658 route: fix typo
 - 41bba8d route: add support for the manipulation of routing informaion

Updates #19967

Change-Id: Id2bb93df97a45254a2df2b048db0143e3e52bbdf
Reviewed-on: https://go-review.googlesource.com/40830
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: remove flag_largemodel

It was added in 2013 in CL 7064048.
All uses of it in the compiler disappeared with
(or possibly before) the SSA backend.
Several releases have gone by without it,
from which I conclude that it is now not needed.

Change-Id: I2095f4ac05d4d7ab998168993a7fd5d954aeee88
Reviewed-on: https://go-review.googlesource.com/40856
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

vendor: update golang_org/x/net/http2 packages

Update the http2/hpack package to rev 05d3205.

Introduce the following changes:
 - 05d3205 http2/hpack: fix memory leak in headerFieldTable lookup maps
 - bce15e7 http2/hpack: speedup Encoder.searchTable
 - dd2d9a6 http2/hpack: remove hpack's constant time string comparison
 - 357296a all: single space after period
 - 71a0359 x/net/http2: Fix various typos in doc comments.

Updates #19967

Change-Id: Ie2c8edcaaf96abde515cb995dfa503b54776abfe
Reviewed-on: https://go-review.googlesource.com/40833
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

vendor: update golang_org/x/net/lif package

Update the lif package to rev 7bf7a75.

Introduce the following changes:
 - 7bf7a75 lif: use of nativeEndian to make API endian agnostic
 - adc6ba9 lif: drop support for go1.5

Updates #19967

Change-Id: Iaba893b5ee9af4c54bf5ba4244ce5752ce9f6ad3
Reviewed-on: https://go-review.googlesource.com/40831
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: Budget for inlining is now 80 nodes, not 40

This comment was out of date since the bump to 80 done as the same time
as inlining transitive functions in:

  commit 77ccb16eb12f461eaea5fdf652a2e929dc154192
  Author: Russ Cox <rsc@golang.org>
  Date:   Tue Feb 24 12:19:01 2015 -0500

      cmd/internal/gc: transitive inlining

Adjust the comment at the top of the file accordingly.

Change-Id: Ia6d7397c874e3b85396e82dc9678e56aab9ad728
Reviewed-on: https://go-review.googlesource.com/40910
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: delay IP protocol stack-snooping system calls

This change delays IP protocol stack-snooping system calls until the
start of connection setup for the better experience with some system
call auditing, such as seccomp on Linux. See #16789 for examples.

Also updates the documentation on favoriteAddrFamily, which is the
owner of stack-snooping system calls.

Fixes #16789.

Change-Id: I4af27bc1ed06ffb1f657b6f6381c328c1f41c66c
Reviewed-on: https://go-review.googlesource.com/40750
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

testing: add TB.Helper to better support test helpers

This CL implements the proposal at
https://github.com/golang/proposal/blob/master/design/4899-testing-helper.md.

It's based on Josh's CL 79890043 from a few years ago:
https://codereview.appspot.com/79890043 but makes several changes,
most notably by using the new CallersFrames API so that it works with
mid-stack inlining.

Another detail came up while I was working on this: I didn't want the
user to be able to call t.Helper from inside their TestXxx function
directly (which would mean we'd print a file:line from inside the
testing package itself), so I explicitly prevented this from working.

Fixes #4899.

Change-Id: I37493edcfb63307f950442bbaf993d1589515310
Reviewed-on: https://go-review.googlesource.com/38796
Run-TryBot: Caleb Spare <cespare@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/compile: add test for Issue 12536

The fixedbugs/issue12536.go file was erroneously deleted just before
committing the patch that fixed the issue (CL 14400).

That's an easy test and there's a small reproducer in the issue, add
it back.

Updates #12536

Change-Id: Ib7b0cd245588299e9a5469e1d75805fd0261ce1a
Reviewed-on: https://go-review.googlesource.com/40712
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

vendor: update golang_org/x/crypto packages

Update the poly1305 and curve25519 packages to the current state of /x/crypto.

Updates #19967

Change-Id: Ib71534f78040f31bfd5debb06f3c4a54a77955b3
Reviewed-on: https://go-review.googlesource.com/40711
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime/trace: iterate over frames instead of PCs

Now the runtime/trace tests pass with -l=4.

This also gets rid of the frames cache for multiple reasons:

1) The frames cache was used to avoid repeated calls to funcname and
funcline. Now these calls happen inside the CallersFrames iterator.

2) Maintaining a frames cache is harder: map[uintptr]traceFrame
doesn't work since each PC can map to multiple traceFrames.

3) It's not clear that the cache is important.

Change-Id: I2914ac0b3ba08e39b60149d99a98f9f532b35bbb
Reviewed-on: https://go-review.googlesource.com/40591
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

runtime/trace: better output when test fails

Change-Id: I108d15eb4cd25904bb76de4ed7548c039c69d1a3
Reviewed-on: https://go-review.googlesource.com/40590
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

net: validate network in Dial{,IP} and Listen{Packet,IP} for IP networks

The argument of the first parameter for connection setup functions on
IP networks must contain a protocol name or number. This change adds
validation for arguments of IP networks to connection setup functions.

Fixes #18185.

Change-Id: I6aaedd7806e3ed1043d4b1c834024f350b99361d
Reviewed-on: https://go-review.googlesource.com/40512
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

net: don't enclose non-literal IPv6 addresses in square brackets

The net package uses various textual representations for network
identifiers and locators on the Internet protocol suite as API.
In fact, the representations are the composition of subset of multple
RFCs: RFC 3986, RFC 4007, RFC 4632, RFC 4291 and RFC 5952.

RFC 4007 describes guidelines for the use of textual representation of
IPv6 addressing/routing scope zone and doesn't prohibit the format for
implementation dependent purposes, as in, specifying a literal IPv6
address and its connected region of routing topology as application
user interface. However, a non-literal IPv6 address, for example, a
host name, with a zone enclosed in square brackets confuses us because
a zone is basically for non-global IPv6 addresses and a pair of square
brackets is used as a set of delimiters between a literal IPv6 address
and a service name or transport port number.

To mitigate such confusion, this change makes JoinHostPort not enclose
non-literal IPv6 addresses in square brackets and SplitHostPort accept
the form "host%zone:port" to recommend that anything enclosed in
square brackets should be a literal IPv6 address.

Before this change:
JoinHostPort("name%zone", "80") = "[name%zone]:80"
JoinHostPort("[::1%zone]", "80") = "[::1%zone]:80"
SplitHostPort("name%zone:80") = "", "", "address name%zone:80: missing brackets in address"
SplitHostPort("[name%zone]:80") = "name%zone", "80", nil
SplitHostPort("[::1%zone]:80") = "::1%zone", "80", nil

After this change:
JoinHostPort("name%zone", "80") = "name%zone:80"
JoinHostPort("[::1%zone]", "80") = "[::1%zone]:80"
SplitHostPort("name%zone:80") = "name%zone", "80", nil
SplitHostPort("[name%zone]:80") = "name%zone", "80", nil // for backwards compatibility
SplitHostPort("[::1%zone]:80") = "::1%zone", "80", nil

Also updates docs and test cases on SplitHostPort and JoinHostPort for
clarification.

Fixes #18059.
Fixes #18060.

Change-Id: I5c3ccce4fa0fbdd58f698fc280635ea4a14d2a37
Reviewed-on: https://go-review.googlesource.com/40510
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/dist: fix incorrect platform string shared by all tests

all tests currently share the same platform string and fail to
vet expected platforms

Fixes #19958

Change-Id: I2801e1e84958e31975769581e27ea5ca6a0edf5b
Reviewed-on: https://go-review.googlesource.com/40511
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

syscall: fix TestGroupCleanup{UserNamespace} on Alpine

This updates TestGroupCleanup and TestGroupCleanupUserNamespace to pass in the
Alpine builder.

Updates #19938

Change-Id: Iacbfd73782eccd57f872f9e85726c6024529c277
Reviewed-on: https://go-review.googlesource.com/40692
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>

cmd/compile: fix s390x unsigned comparison constant merging rules

On s390x unsigned integer comparisons with immediates require the immediate
to be an unsigned 32-bit integer. The rule was checking that the immediate
was a signed 32-bit integer.

This CL also adds a test for comparisons that could be turned into compare
with immediate or equivalent instructions (depending on architecture and
optimizations applied).

Fixes #19940.

Change-Id: Ifd6aa989fd3d50e282f7d30fec9db462c28422b1
Reviewed-on: https://go-review.googlesource.com/40433
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>

test: disable flaky test/fixedbugs/issue10958.go

Updates #18589

Change-Id: I2c3bbc8257c68295051bd2e63e1e11794d0609c3
Reviewed-on: https://go-review.googlesource.com/40651
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: free workbufs during sweeping

This extends the sweeper to free workbufs back to the heap between GC
cycles, allowing this memory to be reused for GC'd allocations or
eventually returned to the OS.

This helps for applications that have high peak heap usage relative to
their regular heap usage (for example, a high-memory initialization
phase). Workbuf memory is roughly proportional to heap size and since
we currently never free workbufs, it's proportional to *peak* heap
size. By freeing workbufs, we can release and reuse this memory for
other purposes when the heap shrinks.

This is somewhat complicated because this costs ~1–2 µs per workbuf
span, so for large heaps it's too expensive to just do synchronously
after mark termination between starting the world and dropping the
worldsema. Hence, we do it asynchronously in the sweeper. This adds a
list of "free" workbuf spans that can be returned to the heap. GC
moves all workbuf spans to this list after mark termination and the
background sweeper drains this list back to the heap. If the sweeper
doesn't finish, that's fine, since getempty can directly reuse any
remaining spans to allocate more workbufs.

Performance impact is negligible. On the x/benchmarks, this reduces
GC-bytes-from-system by 6–11%.

Fixes #19325.

Change-Id: Icb92da2196f0c39ee984faf92d52f29fd9ded7a8
Reviewed-on: https://go-review.googlesource.com/38582
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: allocate GC workbufs from manually-managed spans

Currently the runtime allocates workbufs from persistent memory, which
means they can never be freed.

Switch to allocating them from manually-managed heap spans. This
doesn't free them yet, but it puts us in a position to do so.

For #19325.

Change-Id: I94b2512a2f2bbbb456cd9347761b9412e80d2da9
Reviewed-on: https://go-review.googlesource.com/38581
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: eliminate write barriers from alloc/mark bitmaps

This introduces a new type, *gcBits, to use for alloc/mark bitmap
allocations instead of *uint8. This type is marked go:notinheap, so
uses of it correctly eliminate write barriers. Since we now have a
type, this also extracts some common operations to methods both for
convenience and to avoid (*uint8) casts at most use sites.

For #19325.

Change-Id: Id51f734fb2e96b8b7715caa348c8dcd4aef0696a
Reviewed-on: https://go-review.googlesource.com/38580
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: rename gcBits -> gcBitsArena

This clarifies that the gcBits type is actually an arena of gcBits and
will let us introduce a new gcBits type representing a single
mark/alloc bitmap allocated from the arena.

For #19325.

Change-Id: Idedf76d202d9174a17c61bcca9d5539e042e2445
Reviewed-on: https://go-review.googlesource.com/38579
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: don't count manually-managed spans from heap_{inuse,sys}

Currently, manually-managed spans are included in memstats.heap_inuse
and memstats.heap_sys, but when we export these stats to the user, we
subtract out how much has been allocated for stack spans from both.
This works for now because stacks are the only manually-managed spans
we have.

However, we're about to use manually-managed spans for more things
that don't necessarily have obvious stats we can use to adjust the
user-presented numbers. Prepare for this by changing the accounting so
manually-managed spans don't count toward heap_inuse or heap_sys. This
makes these fields align with the fields presented to the user and
means we don't have to track more statistics just so we can adjust
these statistics.

For #19325.

Change-Id: I5cb35527fd65587ff23339276ba2c3969e2ad98f
Reviewed-on: https://go-review.googlesource.com/38577
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: generalize {alloc,free}Stack to {alloc,free}Manual

We're going to start using manually-managed spans for GC workbufs, so
rename the allocate/free methods and pass in a pointer to the stats to
use instead of using the stack stats directly.

For #19325.

Change-Id: I37df0147ae5a8e1f3cb37d59c8e57a1fcc6f2980
Reviewed-on: https://go-review.googlesource.com/38576
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: rename mspan.stackfreelist -> manualFreeList

We're going to use this free list for other types of manually-managed
memory in the heap.

For #19325.

Change-Id: Ib7e682295133eabfddf3a84f44db43d937bfdd9c
Reviewed-on: https://go-review.googlesource.com/38575
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: rename _MSpanStack -> _MSpanManual

We're about to generalize _MSpanStack to be used for other forms of
in-heap manual memory management in the runtime. This is an automated
rename of _MSpanStack to _MSpanManual plus some comment fix-ups.

For #19325.

Change-Id: I1e20a57bb3b87a0d324382f92a3e294ffc767395
Reviewed-on: https://go-review.googlesource.com/38574
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

cmd/compile: emit string symbols during walk

This avoids needing a mutex to protect stringsym,
and preserves a consistent ctxt.Data ordering
in the face of a concurrent backend.

Updates #15756

Change-Id: I775daae11db5db1269533a00f5249e3a03086ffc
Reviewed-on: https://go-review.googlesource.com/40509
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

spec: clarify size hint for make of maps

For #19903.

Change-Id: Ib28d08d45bfad653bcc1446f160b7b4a485529af
Reviewed-on: https://go-review.googlesource.com/40393
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/compile: remove last c-isms from typecheck.go comments

Change-Id: I0b1ae9d296115000fb30aab39f9eac1200ae68d0
Reviewed-on: https://go-review.googlesource.com/40451
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj: cache dwarfSym

Follow-up to review feedback from
mdempsky on CL 40507.

Reduces mutex contention by about 1%.

Change-Id: I540ea6772925f4a59e58f55a3458eff15880c328
Reviewed-on: https://go-review.googlesource.com/40575
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: refactor Linksym

Extract a helper function, linksymname.
This simplifies Linksym,
and linksymname will be useful in future work.

Change-Id: Ic5ff8b704a16d5020f6931e008e2f630f687cbd3
Reviewed-on: https://go-review.googlesource.com/40550
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj: generate function DWARF symbols early

This removes a concurrent access of ctxt.Data.

Updates #15756

Change-Id: Id017e90e47e093cd8825907f3853bb3d3bf8280d
Reviewed-on: https://go-review.googlesource.com/40507
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/vet/all: print all unparseable lines

In my experience, this usually happens when vet panics.
Dumping all unparseable lines should help diagnosis.

Inspired by the trybot failures in CL 40511.

Change-Id: Ib73e8c8b2942832589c3cc5d33ef35fdafe9965a
Reviewed-on: https://go-review.googlesource.com/40508
Reviewed-by: Rob Pike <r@golang.org>

hash/crc32: optimize arm64 crc32 implementation

ARMv8 defines crc32 instruction.

Comparing to the original crc32 calculation, this patch makes use of
crc32 instructions to do crc32 calculation instead of the multiple
lookup table algorithms.

ARMv8 provides IEEE and Castagnoli polynomials for crc32 calculation
so that the perfomance of these two types of crc32 get significant
improved.

name                                        old time/op   new time/op    delta
CRC32/poly=IEEE/size=15/align=0-32            117ns ± 0%      38ns ± 0%   -67.44%
CRC32/poly=IEEE/size=15/align=1-32            117ns ± 0%      38ns ± 0%   -67.52%
CRC32/poly=IEEE/size=40/align=0-32            129ns ± 0%      41ns ± 0%   -68.37%
CRC32/poly=IEEE/size=40/align=1-32            129ns ± 0%      41ns ± 0%   -68.29%
CRC32/poly=IEEE/size=512/align=0-32           828ns ± 0%     246ns ± 0%   -70.29%
CRC32/poly=IEEE/size=512/align=1-32           828ns ± 0%     132ns ± 0%   -84.06%
CRC32/poly=IEEE/size=1kB/align=0-32          1.58µs ± 0%    0.46µs ± 0%   -70.98%
CRC32/poly=IEEE/size=1kB/align=1-32          1.58µs ± 0%    0.46µs ± 0%   -70.92%
CRC32/poly=IEEE/size=4kB/align=0-32          6.06µs ± 0%    1.74µs ± 0%   -71.27%
CRC32/poly=IEEE/size=4kB/align=1-32          6.10µs ± 0%    1.74µs ± 0%   -71.44%
CRC32/poly=IEEE/size=32kB/align=0-32         48.3µs ± 0%    13.7µs ± 0%   -71.61%
CRC32/poly=IEEE/size=32kB/align=1-32         48.3µs ± 0%    13.7µs ± 0%   -71.60%
CRC32/poly=Castagnoli/size=15/align=0-32      116ns ± 0%      38ns ± 0%   -67.07%
CRC32/poly=Castagnoli/size=15/align=1-32      116ns ± 0%      38ns ± 0%   -66.90%
CRC32/poly=Castagnoli/size=40/align=0-32      127ns ± 0%      40ns ± 0%   -68.11%
CRC32/poly=Castagnoli/size=40/align=1-32      127ns ± 0%      40ns ± 0%   -68.11%
CRC32/poly=Castagnoli/size=512/align=0-32     828ns ± 0%     132ns ± 0%   -84.06%
CRC32/poly=Castagnoli/size=512/align=1-32     827ns ± 0%     132ns ± 0%   -84.04%
CRC32/poly=Castagnoli/size=1kB/align=0-32    1.59µs ± 0%    0.22µs ± 0%   -85.89%
CRC32/poly=Castagnoli/size=1kB/align=1-32    1.58µs ± 0%    0.22µs ± 0%   -85.79%
CRC32/poly=Castagnoli/size=4kB/align=0-32    6.14µs ± 0%    0.77µs ± 0%   -87.40%
CRC32/poly=Castagnoli/size=4kB/align=1-32    6.06µs ± 0%    0.77µs ± 0%   -87.25%
CRC32/poly=Castagnoli/size=32kB/align=0-32   48.3µs ± 0%     5.9µs ± 0%   -87.71%
CRC32/poly=Castagnoli/size=32kB/align=1-32   48.4µs ± 0%     6.0µs ± 0%   -87.69%
CRC32/poly=Koopman/size=15/align=0-32         104ns ± 0%     104ns ± 0%    +0.00%
CRC32/poly=Koopman/size=15/align=1-32         104ns ± 0%     104ns ± 0%    +0.00%
CRC32/poly=Koopman/size=40/align=0-32         235ns ± 0%     235ns ± 0%    +0.00%
CRC32/poly=Koopman/size=40/align=1-32         235ns ± 0%     235ns ± 0%    +0.00%
CRC32/poly=Koopman/size=512/align=0-32       2.71µs ± 0%    2.71µs ± 0%    -0.07%
CRC32/poly=Koopman/size=512/align=1-32       2.71µs ± 0%    2.71µs ± 0%    -0.04%
CRC32/poly=Koopman/size=1kB/align=0-32       5.40µs ± 0%    5.39µs ± 0%    -0.06%
CRC32/poly=Koopman/size=1kB/align=1-32       5.40µs ± 0%    5.40µs ± 0%    +0.02%
CRC32/poly=Koopman/size=4kB/align=0-32       21.5µs ± 0%    21.5µs ± 0%    -0.16%
CRC32/poly=Koopman/size=4kB/align=1-32       21.5µs ± 0%    21.5µs ± 0%    -0.05%
CRC32/poly=Koopman/size=32kB/align=0-32       172µs ± 0%     172µs ± 0%    -0.07%
CRC32/poly=Koopman/size=32kB/align=1-32       172µs ± 0%     172µs ± 0%    -0.01%

name                                        old speed     new speed      delta
CRC32/poly=IEEE/size=15/align=0-32          128MB/s ± 0%   394MB/s ± 0%  +207.95%
CRC32/poly=IEEE/size=15/align=1-32          128MB/s ± 0%   394MB/s ± 0%  +208.09%
CRC32/poly=IEEE/size=40/align=0-32          310MB/s ± 0%   979MB/s ± 0%  +216.07%
CRC32/poly=IEEE/size=40/align=1-32          310MB/s ± 0%   979MB/s ± 0%  +216.16%
CRC32/poly=IEEE/size=512/align=0-32         618MB/s ± 0%  2074MB/s ± 0%  +235.72%
CRC32/poly=IEEE/size=512/align=1-32         618MB/s ± 0%  3852MB/s ± 0%  +523.55%
CRC32/poly=IEEE/size=1kB/align=0-32         646MB/s ± 0%  2225MB/s ± 0%  +244.57%
CRC32/poly=IEEE/size=1kB/align=1-32         647MB/s ± 0%  2225MB/s ± 0%  +243.87%
CRC32/poly=IEEE/size=4kB/align=0-32         676MB/s ± 0%  2352MB/s ± 0%  +248.02%
CRC32/poly=IEEE/size=4kB/align=1-32         672MB/s ± 0%  2352MB/s ± 0%  +250.15%
CRC32/poly=IEEE/size=32kB/align=0-32        678MB/s ± 0%  2387MB/s ± 0%  +252.17%
CRC32/poly=IEEE/size=32kB/align=1-32        678MB/s ± 0%  2388MB/s ± 0%  +252.11%
CRC32/poly=Castagnoli/size=15/align=0-32    129MB/s ± 0%   393MB/s ± 0%  +205.51%
CRC32/poly=Castagnoli/size=15/align=1-32    129MB/s ± 0%   390MB/s ± 0%  +203.41%
CRC32/poly=Castagnoli/size=40/align=0-32    314MB/s ± 0%   988MB/s ± 0%  +215.04%
CRC32/poly=Castagnoli/size=40/align=1-32    314MB/s ± 0%   987MB/s ± 0%  +214.68%
CRC32/poly=Castagnoli/size=512/align=0-32   618MB/s ± 0%  3860MB/s ± 0%  +524.32%
CRC32/poly=Castagnoli/size=512/align=1-32   619MB/s ± 0%  3859MB/s ± 0%  +523.66%
CRC32/poly=Castagnoli/size=1kB/align=0-32   645MB/s ± 0%  4568MB/s ± 0%  +608.56%
CRC32/poly=Castagnoli/size=1kB/align=1-32   650MB/s ± 0%  4567MB/s ± 0%  +602.94%
CRC32/poly=Castagnoli/size=4kB/align=0-32   667MB/s ± 0%  5297MB/s ± 0%  +693.81%
CRC32/poly=Castagnoli/size=4kB/align=1-32   676MB/s ± 0%  5297MB/s ± 0%  +684.00%
CRC32/poly=Castagnoli/size=32kB/align=0-32  678MB/s ± 0%  5519MB/s ± 0%  +713.83%
CRC32/poly=Castagnoli/size=32kB/align=1-32  677MB/s ± 0%  5497MB/s ± 0%  +712.04%
CRC32/poly=Koopman/size=15/align=0-32       143MB/s ± 0%   144MB/s ± 0%    +0.27%
CRC32/poly=Koopman/size=15/align=1-32       143MB/s ± 0%   144MB/s ± 0%    +0.33%
CRC32/poly=Koopman/size=40/align=0-32       169MB/s ± 0%   170MB/s ± 0%    +0.12%
CRC32/poly=Koopman/size=40/align=1-32       170MB/s ± 0%   170MB/s ± 0%    +0.08%
CRC32/poly=Koopman/size=512/align=0-32      189MB/s ± 0%   189MB/s ± 0%    +0.07%
CRC32/poly=Koopman/size=512/align=1-32      189MB/s ± 0%   189MB/s ± 0%    +0.04%
CRC32/poly=Koopman/size=1kB/align=0-32      190MB/s ± 0%   190MB/s ± 0%    +0.05%
CRC32/poly=Koopman/size=1kB/align=1-32      190MB/s ± 0%   190MB/s ± 0%    -0.01%
CRC32/poly=Koopman/size=4kB/align=0-32      190MB/s ± 0%   190MB/s ± 0%    +0.15%
CRC32/poly=Koopman/size=4kB/align=1-32      190MB/s ± 0%   191MB/s ± 0%    +0.05%
CRC32/poly=Koopman/size=32kB/align=0-32     191MB/s ± 0%   191MB/s ± 0%    +0.06%
CRC32/poly=Koopman/size=32kB/align=1-32     191MB/s ± 0%   191MB/s ± 0%    +0.02%

Also fix a bug of arm64 assembler

The optimization is mainly contributed by Fangming.Fang <fangming.fang@arm.com>

Change-Id: I900678c2e445d7e8ad9e2a9ab3305d649230905f
Reviewed-on: https://go-review.googlesource.com/40074
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

net/http/fcgi: expose cgi env vars in request context

The current interface can't access all environment
variables directly or via cgi.RequestFromMap, which
only reads variables on its "white list" to be set on
the http.Request it returns. If an fcgi variable is
not on the "white list" - e.g. REMOTE_USER - the old
code has no access to its value.

This passes variables in the Request context that aren't
used to add data to the Request itself and adds a method
that parses those env vars from the Request's context.

Fixes #16546

Change-Id: Ibf933a768b677ece1bb93d7bf99a14cef36ec671
Reviewed-on: https://go-review.googlesource.com/40012
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

internal/poll: rename RecvFrom to ReadFrom for consistency

Also adds missing docs.

Change-Id: Ibd8dbe8441bc7a41f01ed2e2033db98e479a5176
Reviewed-on: https://go-review.googlesource.com/40412
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/compile: make TestAssembly resilient to output ordering

To preserve reproducible builds, the text entries
during compilation will be sorted before being printed.
TestAssembly currently assumes that function init
comes after all user-defined functions.
Remove that assumption.
Instead of looking for "TEXT" to tell you where
a function ends--which may now yield lots of
non-function-code junk--look for a line beginning
with non-whitespace.

Updates #15756

Change-Id: Ibc82dba6143d769ef4c391afc360e523b1a51348
Reviewed-on: https://go-review.googlesource.com/39853
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/internal/obj: build ctxt.Text during Sym init

Instead of constructing ctxt.Text in Flushplist,
which will be called concurrently,
do it in InitTextSym, which must be called serially.
This allows us to avoid a mutex for ctxt.Text,
and preserves the existing ordering of functions
for debug output.

Passes toolstash-check.

Updates #15756

Change-Id: I6322b4da24f9f0db7ba25e5b1b50e8d3be2deb37
Reviewed-on: https://go-review.googlesource.com/40502
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/dist: don't compile unneeded GOARCH SSA rewrite rules during bootstrap

Speeds up build (the bootstrap phase) by ~6 seconds.

Bootstrap goes from ~18 seconds to ~12 seconds.

Change-Id: I7e2ec8f5fc668bf6168d90098eaf70390b16e479
Reviewed-on: https://go-review.googlesource.com/40503
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

hash/fnv: add 128-bit FNV hash support

The 128bit FNV hash will be used e.g. in QUIC.

The algorithm is described at
https://en.wikipedia.org/wiki/Fowler%E2%80%93Noll%E2%80%93Vo_hash_function

Change-Id: I13f3ec39b0e12b7a5008824a6619dff2e708ee81
Reviewed-on: https://go-review.googlesource.com/38356
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

encoding/asn1: support 31 bit identifiers with OID

The current implementation uses a max of 28 bits when decoding an
ObjectIdentifier.  This change makes it so that an int64 is used to
accumulate up to 35 bits.  If the resulting data would not overflow
an int32, it is used as an int.  Thus up to 31 bits may be used to
represent each subidentifier of an ObjectIdentifier.

Fixes #19933

Change-Id: I95d74b64b24cdb1339ff13421055bce61c80243c
Reviewed-on: https://go-review.googlesource.com/40436
Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>

os: handle relative symlinks starting with slash in Stat on windows

https://go-review.googlesource.com/c/39932/ handles relative symlinks.
But that change is incomplete.
We also have to handle relative symlinks starting with slash too.

Fixes #19937

Change-Id: I50dbccbaf270cb48a08fa57e5f450e5da18a7701
Reviewed-on: https://go-review.googlesource.com/40410
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: move Text.From.Sym initialization earlier

The initialization of an ATEXT Prog's From.Sym
can race with the assemblers in a concurrent compiler.
CL 40254 contains an initial, failed attempt to
fix that race.

This CL takes a different approach: Rather than
expose an API to initialize the Prog,
expose an API to initialize the Sym.

The initialization of the Sym can then be
moved earlier in the compiler, avoiding the race.

The growth of gc.Func has negligible
performance impact; see below.

Passes toolstash -cmp.

Updates #15756

name       old alloc/op      new alloc/op      delta
Template        38.8MB ± 0%       38.8MB ± 0%    ~     (p=0.968 n=9+10)
Unicode         29.8MB ± 0%       29.8MB ± 0%    ~     (p=0.684 n=10+10)
GoTypes          113MB ± 0%        113MB ± 0%    ~     (p=0.912 n=10+10)
SSA             1.25GB ± 0%       1.25GB ± 0%    ~     (p=0.481 n=10+10)
Flate           25.3MB ± 0%       25.3MB ± 0%    ~     (p=0.105 n=10+10)
GoParser        31.7MB ± 0%       31.8MB ± 0%  +0.09%  (p=0.016 n=8+10)
Reflect         78.3MB ± 0%       78.2MB ± 0%    ~     (p=0.190 n=10+10)
Tar             26.5MB ± 0%       26.6MB ± 0%  +0.13%  (p=0.011 n=10+10)
XML             42.4MB ± 0%       42.4MB ± 0%    ~     (p=0.971 n=10+10)

name       old allocs/op     new allocs/op     delta
Template          378k ± 1%         378k ± 0%    ~     (p=0.315 n=10+9)
Unicode           321k ± 1%         321k ± 0%    ~     (p=0.436 n=10+10)
GoTypes          1.14M ± 0%        1.14M ± 0%    ~     (p=0.079 n=10+9)
SSA              9.70M ± 0%        9.70M ± 0%  -0.04%  (p=0.035 n=10+10)
Flate             233k ± 1%         234k ± 1%    ~     (p=0.529 n=10+10)
GoParser          315k ± 0%         316k ± 0%    ~     (p=0.095 n=9+10)
Reflect           980k ± 0%         980k ± 0%    ~     (p=0.436 n=10+10)
Tar               249k ± 1%         250k ± 0%    ~     (p=0.280 n=10+10)
XML               391k ± 1%         391k ± 1%    ~     (p=0.481 n=10+10)

Change-Id: I3c93033dddd2e1df8cc54a106a6e615d27859e71
Reviewed-on: https://go-review.googlesource.com/40496
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

net: delete TestDialTimeoutFDLeak

It's flaky and distracting.

I'm not sure what it's testing, either. It hasn't saved us before.

Somebody can resurrect it if they have time.

Updates #15157

Change-Id: I27bbfe51e09b6259bba0f73d60d03a4d38711951
Reviewed-on: https://go-review.googlesource.com/40498
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

cmd/internal/obj: stop storing Text flags in From3

Prior to this CL, flags such as NOSPLIT
on ATEXT Progs were stored in From3.Offset.
Some but not all of those flags were also
duplicated into From.Sym.Attribute.

This CL migrates all of those flags into
From.Sym.Attribute and stops creating a From3.

A side-effect of this is that printing an
ATEXT Prog can no longer simply dump From3.Offset.
That's kind of good, since the raw flag value
wasn't very informative anyway, but it did
necessitate a bunch of updates to the cmd/asm tests.

The reason I'm doing this work now is that
avoiding storing flags in both From.Sym and From3.Offset
simplifies some other changes to fix the data
race first described in CL 40254.

This CL almost passes toolstash-check -all.
The only changes are in cases where the assembler
has decided that a function's flags may be altered,
e.g. to make a function with no calls in it NOSPLIT.
Prior to this CL, that information was not printed.

Sample before:

"".Ctz64 t=1 size=63 args=0x10 locals=0x0
0x0000 00000 (/Users/josh/go/tip/src/runtime/internal/sys/intrinsics.go:35) TEXT "".Ctz64(SB), $0-16
0x0000 00000 (/Users/josh/go/tip/src/runtime/internal/sys/intrinsics.go:35) FUNCDATA $0, gclocals·f207267fbf96a0178e8758c6e3e0ce28(SB)

Sample after:

"".Ctz64 t=1 nosplit size=63 args=0x10 locals=0x0
0x0000 00000 (/Users/josh/go/tip/src/runtime/internal/sys/intrinsics.go:35) TEXT "".Ctz64(SB), NOSPLIT, $0-16
0x0000 00000 (/Users/josh/go/tip/src/runtime/internal/sys/intrinsics.go:35) FUNCDATA $0, gclocals·f207267fbf96a0178e8758c6e3e0ce28(SB)

Observe the additional "nosplit" in the first line
and the additional "NOSPLIT" in the second line.

Updates #15756

Change-Id: I5c59bd8f3bdc7c780361f801d94a261f0aef3d13
Reviewed-on: https://go-review.googlesource.com/40495
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile/internal/ssa: refactor ARM64 address folding

These patterns are the only uses of isArg and isAuto, and they all
follow a common pattern too. Extract out so that we can more easily
tweak the interface for isArg/isAuto.

Passes toolstash -cmp for linux/arm64.

Change-Id: I9c509dabdc123c93cb1ad2f34fe8c12a9f313f6d
Reviewed-on: https://go-review.googlesource.com/40490
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>