]> Cypherpunks repositories - gostls13.git/log
gostls13.git
5 years agonet: add (*Resolver).LookupIP
Ian Gudger [Fri, 17 Apr 2020 11:04:23 +0000 (04:04 -0700)]
net: add (*Resolver).LookupIP

Previously, looking up only IPv4 or IPv6 addresses was only possible
with DefaultResolver via ResolveIPAddr. Add this functionality to the
Resolver type with a new method, LookupIP. This largely brings Resolver
functionally to parity with the global functions. The name LookupIP is
used over ResolveIPAddr to be consistent with the other Resolver
methods.

There are two main benefits to (*Resolver).LookupIP over
(*Resolver).LookupHost. First is an ergonomic benefit. Wanting a
specific family of address is common enough to justify a method, evident
by the existence of ResolveIPAddr. Second, this opens the possibility of
not performing unnecessary DNS requests when only a specific family of
addresses are needed. This optimization is left to follow up work.

Updates #30452

Change-Id: I241f61019588022a39738f8920b0ddba900cecdd
Reviewed-on: https://go-review.googlesource.com/c/go/+/228641
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoreflect: reduce allocations for Select with up to 4 cases
Martin Möhrmann [Tue, 28 Apr 2020 19:58:46 +0000 (21:58 +0200)]
reflect: reduce allocations for Select with up to 4 cases

Allocate the runcases slice on the stack if the number
of select cases is small (up to 4).

Found while looking at production profiles of common
proto based RPC server framework code in Google which do
not have a large number of cases.

name      old time/op    new time/op    delta
Select/1     147ns ± 2%     120ns ± 6%  -18.32%  (p=0.000 n=7+10)
Select/4     316ns ± 5%     249ns ± 2%  -21.23%  (p=0.000 n=10+10)
Select/8     516ns ± 3%     515ns ± 3%     ~     (p=0.858 n=10+9)

name      old alloc/op   new alloc/op   delta
Select/1     96.0B ± 0%     64.0B ± 0%  -33.33%  (p=0.000 n=10+10)
Select/4      336B ± 0%      208B ± 0%  -38.10%  (p=0.000 n=10+10)
Select/8      672B ± 0%      672B ± 0%     ~     (all equal)

name      old allocs/op  new allocs/op  delta
Select/1      4.00 ± 0%      3.00 ± 0%  -25.00%  (p=0.000 n=10+10)
Select/4      7.00 ± 0%      6.00 ± 0%  -14.29%  (p=0.000 n=10+10)
Select/8      11.0 ± 0%      11.0 ± 0%     ~     (all equal)

Change-Id: I1687e74fc8e86606a27f03fa8a561bcfb68775d6
Reviewed-on: https://go-review.googlesource.com/c/go/+/230657
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agodatabase/sql: document Connect and Close may need a timeout
Daniel Theophanes [Tue, 28 Apr 2020 14:31:12 +0000 (07:31 -0700)]
database/sql: document Connect and Close may need a timeout

Opening a connection with Connect should still create a derived
context with a timeout because some clients will not use a timeout
and the connection pool may open a connection asynchronously.

Likewise, if a connection close makes a network operation it should
provide some type of sane timeout for the operation.

Fixes #38185

Change-Id: I9b7ce2996c81c486170dcc84b12672a99610fa27
Reviewed-on: https://go-review.googlesource.com/c/go/+/230438
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
5 years agostrings: simpler and slightly faster implementation of FieldsFunc
Robert Griesemer [Fri, 24 Apr 2020 05:17:58 +0000 (22:17 -0700)]
strings: simpler and slightly faster implementation of FieldsFunc

Removed the need for maintaining an extra variable in one of the inner loops,
leading to a slight speed-up for short strings.

Benchmarks run on a "quiet" MacBook Pro, 3.3GHz Dual-Core Intel Core i7,
with 16GB 2133MHz LPDDR3 RAM running macOS 10.15.4.

name                        old time/op    new time/op    delta
FieldsFunc/ASCII/16-4          147ns ± 0%     144ns ± 1%  -2.04%  (p=0.000 n=4+5)
FieldsFunc/ASCII/256-4        1.63µs ± 0%    1.59µs ± 1%  -2.50%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4       30.0µs ± 0%    29.3µs ± 2%    ~     (p=0.190 n=4+5)
FieldsFunc/ASCII/65536-4       491µs ± 5%     473µs ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/ASCII/1048576-4    8.02ms ± 7%    7.85ms ± 4%    ~     (p=0.548 n=5+5)
FieldsFunc/Mixed/16-4          182ns ± 1%     181ns ± 4%    ~     (p=0.357 n=5+5)
FieldsFunc/Mixed/256-4        1.74µs ± 1%    1.74µs ± 1%    ~     (p=0.881 n=5+5)
FieldsFunc/Mixed/4096-4       34.9µs ± 2%    34.7µs ± 0%    ~     (p=0.310 n=5+5)
FieldsFunc/Mixed/65536-4       595µs ± 1%     589µs ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/Mixed/1048576-4    10.1ms ± 3%     9.8ms ± 2%    ~     (p=0.095 n=5+5)

name                        old speed      new speed      delta
FieldsFunc/ASCII/16-4        109MB/s ± 1%   111MB/s ± 1%  +2.33%  (p=0.008 n=5+5)
FieldsFunc/ASCII/256-4       157MB/s ± 0%   161MB/s ± 1%  +2.57%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4      137MB/s ± 0%   140MB/s ± 2%    ~     (p=0.190 n=4+5)
FieldsFunc/ASCII/65536-4     134MB/s ± 4%   139MB/s ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/ASCII/1048576-4   131MB/s ± 6%   134MB/s ± 4%    ~     (p=0.548 n=5+5)
FieldsFunc/Mixed/16-4       87.8MB/s ± 1%  88.3MB/s ± 4%    ~     (p=0.421 n=5+5)
FieldsFunc/Mixed/256-4       147MB/s ± 1%   147MB/s ± 1%    ~     (p=0.841 n=5+5)
FieldsFunc/Mixed/4096-4      117MB/s ± 2%   118MB/s ± 0%    ~     (p=0.310 n=5+5)
FieldsFunc/Mixed/65536-4     110MB/s ± 1%   111MB/s ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/Mixed/1048576-4   104MB/s ± 3%   107MB/s ± 2%    ~     (p=0.095 n=5+5)

name                        old alloc/op   new alloc/op   delta
FieldsFunc/ASCII/16-4          32.0B ± 0%     32.0B ± 0%    ~     (all equal)
FieldsFunc/ASCII/256-4          352B ± 0%      352B ± 0%    ~     (all equal)
FieldsFunc/ASCII/4096-4       21.9kB ± 0%    21.9kB ± 0%    ~     (all equal)
FieldsFunc/ASCII/65536-4       448kB ± 0%     448kB ± 0%    ~     (all equal)
FieldsFunc/ASCII/1048576-4    8.85MB ± 0%    8.85MB ± 0%    ~     (p=0.738 n=5+5)
FieldsFunc/Mixed/16-4          48.0B ± 0%     48.0B ± 0%    ~     (all equal)
FieldsFunc/Mixed/256-4          416B ± 0%      416B ± 0%    ~     (all equal)
FieldsFunc/Mixed/4096-4       21.5kB ± 0%    21.5kB ± 0%    ~     (all equal)
FieldsFunc/Mixed/65536-4       448kB ± 0%     448kB ± 0%    ~     (all equal)
FieldsFunc/Mixed/1048576-4    8.85MB ± 0%    8.85MB ± 0%    ~     (p=0.690 n=5+5)

name                        old allocs/op  new allocs/op  delta
FieldsFunc/ASCII/16-4           1.00 ± 0%      1.00 ± 0%    ~     (all equal)
FieldsFunc/ASCII/256-4          1.00 ± 0%      1.00 ± 0%    ~     (all equal)
FieldsFunc/ASCII/4096-4         5.00 ± 0%      5.00 ± 0%    ~     (all equal)
FieldsFunc/ASCII/65536-4        12.0 ± 0%      12.0 ± 0%    ~     (all equal)
FieldsFunc/ASCII/1048576-4      24.0 ± 0%      24.0 ± 0%    ~     (all equal)
FieldsFunc/Mixed/16-4           1.00 ± 0%      1.00 ± 0%    ~     (all equal)
FieldsFunc/Mixed/256-4          1.00 ± 0%      1.00 ± 0%    ~     (all equal)
FieldsFunc/Mixed/4096-4         5.00 ± 0%      5.00 ± 0%    ~     (all equal)
FieldsFunc/Mixed/65536-4        12.0 ± 0%      12.0 ± 0%    ~     (all equal)
FieldsFunc/Mixed/1048576-4      24.0 ± 0%      24.0 ± 0%    ~     (all equal)

Change-Id: I06828d798ca1a624a26edd7f7b68c3bf2fc28f84
Reviewed-on: https://go-review.googlesource.com/c/go/+/229765
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
5 years agocmd/compile: simplify readonly sym checks in writebarrier pass
Josh Bleecher Snyder [Tue, 28 Apr 2020 17:20:55 +0000 (10:20 -0700)]
cmd/compile: simplify readonly sym checks in writebarrier pass

CL 220499 started marking readonly syms as SRODATA earlier,
so we can use that in the writebarrier pass now.

Passes toolstash-check.

Change-Id: Ic4d49714b8bffbe03c8e9a75ca96df4475bae732
Reviewed-on: https://go-review.googlesource.com/c/go/+/230559
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
5 years agoruntime: fix block leak due to race in span set
Michael Anthony Knyszek [Tue, 28 Apr 2020 15:14:56 +0000 (15:14 +0000)]
runtime: fix block leak due to race in span set

The span set data structure may leak blocks due to a race in the logic
to check whether it's safe to free a block. The simplest example of this
race is between two poppers:

1. Popper A claims slot spanSetEntries-2.
2. Popper B claims slot spanSetEntries-1.
3. Popper A gets descheduled before it subtracts from block.used.
4. Popper B subtracts from block.used, sees that claimed
   spanSetEntries-1, but also that block.used != 0, so it returns.
5. Popper A comes back and subtracts from block.used, but it didn't
   claim spanSetEntries-1 so it also returns.

The spine is left with a stale block pointer and the block later gets
overwritten by pushes, never to be re-used again.

The problem here is that we designate the claimer of slot
spanSetEntries-1 to be the one who frees the block, but that may not be
the thread that actually does the last subtraction from block.used.

Fixing this problem is tricky, and the fundamental problem there is that
block.used is not stable: it may be observed to be zero, but that
doesn't necessarily mean you're the last popper!

Do something simpler: keep a counter of how many pops have happened to a
given block instead of block.used. This counter monotonically increases
when a pop is _completely done_.  Because this counter is monotonically
increasing, and only increases when a popper is done, then we know for
sure whichever popper is the last to increase it (i.e. its value is
spanSetBlockEntries) is also the last popper in the block. Because the
race described above still exists, the last popper may not be the one
which claimed the last slot in the block, but we know for certain nobody
else is popping from that block anymore so we can safely free it.
Finally, because pops serialize with pushes to the same slot, we need
not worry about concurrent pushers at all.

Updates #37487.

Change-Id: I6697219372774c8ca7d8ee6895eaa230a64ce9e1
Reviewed-on: https://go-review.googlesource.com/c/go/+/230497
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoruntime: flush mcaches to mcentral before reading memstats
Michael Anthony Knyszek [Tue, 28 Apr 2020 16:40:38 +0000 (16:40 +0000)]
runtime: flush mcaches to mcentral before reading memstats

Currently mcaches are flushed to mcentral after a bunch of memstats have
already been read. This is not safe (in the sense that it doesn't ensure
consisent memstats) since memstats may in general change when mcentral
data structures are manipulated.

Note that prior to the new mcentral implementation this was not a
problem because mcentral operations happened to never modify certain
memstats. As of the new mcentral implementation, we might for example
persistentalloc when uncaching a span, which would change memstats. This
can cause a skew between the value of sys (which currently is calculated
before mcaches are flushed) and the value of gc_sys and other_sys.

Fix this by moving mcache flushing to the very top of updatememstats.
Also leave a comment explaining that this must be done first, in
general, because mcentrals make no guarantee that they will not
influence memstats (and doing so would be unnecessarily restrictive).

Fixes #38712.

Change-Id: I15bacb313c54a46e380a945a71bb75db67169c1b
Reviewed-on: https://go-review.googlesource.com/c/go/+/230498
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
5 years agocmd/compile: port first part of arm64 opt rules to typed aux
Alberto Donizetti [Tue, 28 Apr 2020 13:32:43 +0000 (15:32 +0200)]
cmd/compile: port first part of arm64 opt rules to typed aux

Fairly minimal changes.

Passes

  GOARCH=arm64 gotip build -toolexec 'toolstash -cmp' -a std

Change-Id: I14b2e3ad4933ef14b1b6993f427bf84fe85b1818
Reviewed-on: https://go-review.googlesource.com/c/go/+/229939
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agodoc/go1.15: add notes for GOMODCACHE, modcacheunzipinplace
Jay Conrod [Tue, 28 Apr 2020 15:58:33 +0000 (11:58 -0400)]
doc/go1.15: add notes for GOMODCACHE, modcacheunzipinplace

For #36568
For #34527

Change-Id: Ieea4b4a7644e9c957f48d08d2e172e39b571502f
Reviewed-on: https://go-review.googlesource.com/c/go/+/230537
Reviewed-by: Michael Matloob <matloob@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agointernal/goversion: update to 1.15
Bryan C. Mills [Tue, 28 Apr 2020 01:20:36 +0000 (01:20 +0000)]
internal/goversion: update to 1.15

This is a second attempt at CL 230024, with
cmd/go/testdata/script/mod_retention.txt updated to perform a
version-independent comparison on the 'go' version added to a go.mod
file that lacks one.

Fixes #38708

Change-Id: I15dcd83b51ed5ec57946b419bcbaec41e85a46f9
Reviewed-on: https://go-review.googlesource.com/c/go/+/230382
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
5 years agodatabase/sql: fix incorrect function name in example_test
Vivian Liang [Tue, 28 Apr 2020 00:36:41 +0000 (00:36 +0000)]
database/sql: fix incorrect function name in example_test

fixes incorrect function name prefix in `example_test.go`

Change-Id: I0cce8afc89f6ee6dab3e5152246d28deb597fe53
GitHub-Last-Rev: 4e9a6067c8307e049ace2e35414a838bf6c8a8a4
GitHub-Pull-Request: golang/go#38715
Reviewed-on: https://go-review.googlesource.com/c/go/+/230381
Reviewed-by: Daniel Theophanes <kardianos@gmail.com>
5 years agocmd/compile: adopt strong aux typing for some s390x rules
Ruixin Bao [Mon, 27 Apr 2020 23:20:45 +0000 (16:20 -0700)]
cmd/compile: adopt strong aux typing for some s390x rules

Convert some optimizations rules to strongly-typed versions. So far, I
have only converted the rules that need no additional changes (i.e: only
need to change '->' to "=>").

This CL covers the rules from line 478 - line 800 in S390X.rules file.
Some compare and branch rules also fall in this range, but they were
already done previously in another CL.

Passes toolstash-check.
Change-Id: I9167c5f1a32f4fd6c29bacc13fff95e83b0533e0
Reviewed-on: https://go-review.googlesource.com/c/go/+/230338
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: rewrite some arm64 rules to use typed aux fields
fanzha02 [Wed, 22 Apr 2020 02:03:00 +0000 (10:03 +0800)]
cmd/compile: rewrite some arm64 rules to use typed aux fields

Passes toolstash-check -all.

Change-Id: Ibf8c2532b0de65901bf0dd9ef0d198dc54d56470
Reviewed-on: https://go-review.googlesource.com/c/go/+/229738
Reviewed-by: Keith Randall <khr@golang.org>
5 years agogo/types: improve errors for method calls with pointer receivers
Rebecca Stambler [Mon, 27 Apr 2020 23:41:14 +0000 (19:41 -0400)]
go/types: improve errors for method calls with pointer receivers

The compiler has better error messages for methods called without a
pointer receiver when one is expected. This change is similar to
CL 229801, but for method calls.

Also, added better error messages for functions called with the wrong
capitalization. I left the third TODO in this switch statement almost
as-is because I'm not sure that the extra complexity is worth it -
I adjusted the error to look like the one the compiler reports.

Fixes golang/go#38658

Change-Id: Ie0ca2503e12f3659f112f0135cc27db1b027fdcb
Reviewed-on: https://go-review.googlesource.com/c/go/+/230380
Run-TryBot: Rebecca Stambler <rstambler@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
5 years agoos, internal/poll, internal/syscall/unix: use copy_file_range on Linux
Andrei Tudor Călin [Tue, 21 Apr 2020 23:08:33 +0000 (02:08 +0300)]
os, internal/poll, internal/syscall/unix: use copy_file_range on Linux

Linux 4.5 introduced (and Linux 5.3 refined) the copy_file_range
system call, which allows file systems the opportunity to implement
copy acceleration techniques. This commit adds support for
copy_file_range(2) to the os package.

Introduce a new ReadFrom method on *os.File, which makes *os.File
implement the io.ReaderFrom interface. If dst and src are both files,
this enables io.Copy(dst, src) to call dst.ReadFrom(src), which, in
turn, will call copy_file_range(2) if possible. If copy_file_range(2)
is not supported by the host kernel, or if either of dst or src
refers to a non-regular file, ReadFrom falls back to the regular
io.Copy code path.

Add internal/poll.CopyFileRange, which acquires locks on the
appropriate poll.FDs and performs the actual work, as well as
internal/syscall/unix.CopyFileRange, which wraps the copy_file_range
system call itself at the lowest level.

Rework file layout in internal/syscall/unix to accomodate the
additional system call numbers needed for copy_file_range.
Merge these definitions with the ones used by getrandom(2) into
sysnum_linux_$GOARCH.go files.

A note on additional optimizations: if dst and src both refer to pipes
in the invocation dst.ReadFrom(src), we could, in theory, use the
existing splice(2) code in package internal/poll to splice directly
from src to dst. Attempting this runs into trouble with the poller,
however. If we call splice(src, dst) and see EAGAIN, we cannot know
if it came from src not being ready for reading or dst not being
ready for writing. The write end of src and the read end of dst are
not under our control, so we cannot reliably use the poller to wait
for readiness. Therefore, it seems infeasible to use the new ReadFrom
method to splice between pipes directly. In conclusion, for now, the
only optimization enabled by the new ReadFrom method on *os.File is
the copy_file_range optimization.

Fixes #36817.

Change-Id: I696372639fa0cdf704e3f65414f7321fc7d30adb
Reviewed-on: https://go-review.googlesource.com/c/go/+/229101
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agobufio: optimize bufio.Reader.ReadString to avoid an allocation and copy
Alex Gaynor [Mon, 27 Apr 2020 23:10:23 +0000 (23:10 +0000)]
bufio: optimize bufio.Reader.ReadString to avoid an allocation and copy

name                old time/op    new time/op    delta
ReaderReadString-4     226ns ±12%     161ns ±11%  -28.76%  (p=0.008 n=5+5)

name                old alloc/op   new alloc/op   delta
ReaderReadString-4      288B ± 0%      144B ± 0%  -50.00%  (p=0.008 n=5+5)

name                old allocs/op  new allocs/op  delta
ReaderReadString-4      2.00 ± 0%      1.00 ± 0%  -50.00%  (p=0.008 n=5+5)

Change-Id: I77f330b8340c2bfbfff1f6f1000170b65953a200
GitHub-Last-Rev: 65d65302a7b80504b4d37b81a3843fe1439e638a
GitHub-Pull-Request: golang/go#34706
Reviewed-on: https://go-review.googlesource.com/c/go/+/199257
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agoimage/draw: optimize paletted dst + uniform src
Nigel Tao [Sat, 25 Apr 2020 12:29:02 +0000 (22:29 +1000)]
image/draw: optimize paletted dst + uniform src

name            old time/op  new time/op  delta
PalettedFill-4  5.74ms ± 1%  0.01ms ± 1%  -99.78%  (p=0.008 n=5+5)
PalettedRGBA-4  3.34ms ± 3%  3.33ms ± 0%     ~     (p=0.690 n=5+5)

Fixes #35938

Thanks to pjbgtnj for the suggestion.

Change-Id: I07b494482cce918f556e196c5a4b481b4c16de3a
Reviewed-on: https://go-review.googlesource.com/c/go/+/230118
Run-TryBot: Nigel Tao <nigeltao@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
5 years agostrconv: remove redundant conversions to int
smasher164 [Mon, 27 Apr 2020 17:23:24 +0000 (13:23 -0400)]
strconv: remove redundant conversions to int

IntSize is an untyped constant that does not need explicit conversion.
Annotating IntSize as an int and running github.com/mdempsky/unconvert
reveals these two cases.

Fixes #38682.

Change-Id: I014646b7457ddcde32474810153229dcf0c269c6
Reviewed-on: https://go-review.googlesource.com/c/go/+/230306
Run-TryBot: Akhil Indurti <aindurti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agoruntime: ensure allocToCache updates searchAddr in a valid way
Michael Anthony Knyszek [Wed, 22 Apr 2020 21:36:11 +0000 (21:36 +0000)]
runtime: ensure allocToCache updates searchAddr in a valid way

Currently allocToCache assumes it can move the search address past the
block it allocated the cache from, which violates the property that
searchAddr should always point to mapped memory (i.e. memory represented
by pageAlloc.inUse).

This bug was already fixed once for pageAlloc.alloc in the Go 1.14
release via CL 216697, but that changed failed to take into account
allocToCache.

Fixes #38605.

Change-Id: Id08180aa10d19dc0f9f551a1d9e327a295560dff
Reviewed-on: https://go-review.googlesource.com/c/go/+/229577
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
5 years agohash/crc32: simplify hasVX checking on s390x
Ruixin(Peter) Bao [Mon, 27 Apr 2020 19:23:37 +0000 (15:23 -0400)]
hash/crc32: simplify hasVX checking on s390x

Originally, we use an assembly function that returns a boolean result to
tell whether the machine has vector facility or not. It is now no longer
needed when we can directly use cpu.S390X.HasVX variable. This CL
also removes the last occurence of hasVectorFacility function on s390x.

Change-Id: Id20cb746c21eacac5e13344b362e2d87adfe4317
Reviewed-on: https://go-review.googlesource.com/c/go/+/230337
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agomath/big: simplify hasVX checking on s390x
Ruixin(Peter) Bao [Mon, 27 Apr 2020 13:51:01 +0000 (09:51 -0400)]
math/big: simplify hasVX checking on s390x

Originally, we use an assembly function that returns a boolean result to
tell whether the machine has vector facility or not. It is now no longer
needed when we can directly use cpu.S390X.HasVX variable.

Change-Id: Ic1dae851982532bcfd9a9453416c112347f21d87
Reviewed-on: https://go-review.googlesource.com/c/go/+/230318
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agomath: simplify hasVX checking on s390x
Ruixin Bao [Mon, 27 Apr 2020 19:02:52 +0000 (12:02 -0700)]
math: simplify hasVX checking on s390x

Originally, we use an assembly function that returns a boolean result to
tell whether the machine has vector facility or not. It is now no longer
needed when we can directly use cpu.S390X.HasVX variable.

Change-Id: Ic3ffeb9e63238ef41406d97cdc42502145ddb454
Reviewed-on: https://go-review.googlesource.com/c/go/+/230319
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocrypto/ecdsa: implement ecdsa on s390x for P256/P384/P521 using KDSA instruction
Ruixin Bao [Wed, 15 Apr 2020 20:50:50 +0000 (16:50 -0400)]
crypto/ecdsa: implement ecdsa on s390x for P256/P384/P521 using KDSA instruction

This CL revives CL 174437(also IBM CLA) and adds benchmarks and some simplifications.

The original commit message is as follows:

Utilize KDSA when available. This guarantees constant time operation on all three curves mentioned,
and is faster than conventional assembly.

Benchmarks:
name             old time/op    new time/op    delta
SignP256-8         15.2µs ±14%    14.1µs ±18%     ~     (p=0.356 n=9+10)
SignP384-8         4.28ms ±26%    0.02ms ±30%  -99.43%  (p=0.000 n=10+10)
VerifyP256-8       33.6µs ±13%    13.3µs ±38%  -60.32%  (p=0.000 n=9+10)

name             old alloc/op   new alloc/op   delta
SignP256-8         2.16kB ± 0%    1.60kB ± 0%  -25.63%  (p=0.000 n=9+10)
SignP384-8         1.75MB ± 0%    0.00MB ± 0%  -99.90%  (p=0.000 n=9+10)
VerifyP256-8       1.08kB ± 0%    0.18kB ± 0%  -83.70%  (p=0.000 n=9+10)

name             old allocs/op  new allocs/op  delta
SignP256-8           29.0 ± 0%      22.0 ± 0%  -24.14%  (p=0.000 n=10+10)
SignP384-8          14.4k ± 0%      0.0k ± 0%  -99.85%  (p=0.000 n=9+10)
VerifyP256-8         23.0 ± 0%       7.0 ± 0%  -69.57%  (p=0.000 n=10+10)

Change-Id: Ifa1fc5917fa7592dd592affa7549147dbc9b4169
Reviewed-on: https://go-review.googlesource.com/c/go/+/228580
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
5 years agocrypto/ed25519: implement ed25519 on s390x using KDSA instruction
Ruixin(Peter) Bao [Mon, 21 Oct 2019 17:18:31 +0000 (13:18 -0400)]
crypto/ed25519: implement ed25519 on s390x using KDSA instruction

This CL allows the usage of KDSA instruction when it is available.  The
instruction is designed to be resistant to side channel attacks and
offers performance improvement for ed25519.

Benchmarks:
name              old time/op    new time/op    delta
Signing-8            120µs ±20%      62µs ±12%   -48.40%  (p=0.000 n=10+10)
Verification-8       325µs ±17%      69µs ±10%   -78.80%  (p=0.000 n=10+10)

name              old alloc/op   new alloc/op   delta
Signing-8             448B ± 0%        0B       -100.00%  (p=0.000 n=10+10)
Verification-8        288B ± 0%        0B       -100.00%  (p=0.000 n=10+10)

name              old allocs/op  new allocs/op  delta
Signing-8             5.00 ± 0%      0.00       -100.00%  (p=0.000 n=10+10)
Verification-8        2.00 ± 0%      0.00       -100.00%  (p=0.000 n=10+10)

Change-Id: I0330ce83d807370b419ce638bc2cae4cb3c250dc
Reviewed-on: https://go-review.googlesource.com/c/go/+/202578
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
5 years agoRevert "go/types: add UsesCgo config to support _cgo_gotypes.go"
Andrew Bonventre [Mon, 27 Apr 2020 17:46:01 +0000 (17:46 +0000)]
Revert "go/types: add UsesCgo config to support _cgo_gotypes.go"

This reverts CL 33677.

Reason for revert: NetBSD is broken

Updates #38649

Change-Id: Id60e3c97d3cb4fb0053dea03b95dbbb0b850c883
Reviewed-on: https://go-review.googlesource.com/c/go/+/230038
Run-TryBot: Andrew Bonventre <andybons@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
5 years agoruntime: added several new lock-rank partial order edges
Dan Scales [Mon, 13 Apr 2020 15:37:03 +0000 (08:37 -0700)]
runtime: added several new lock-rank partial order edges

Several new ones came from my testing (long, repeated runs) and one (assistQueue ->
spine) came from the staticlockranking builder (filed as issue 38441).

Fixes #38441

Change-Id: I4268da0d8b8cc51251eba6bd936110c8ab4c4e61
Reviewed-on: https://go-review.googlesource.com/c/go/+/229480
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
5 years agoruntime: bound small object sweeping to 100 spans when allocating
Michael Anthony Knyszek [Fri, 24 Apr 2020 20:06:57 +0000 (20:06 +0000)]
runtime: bound small object sweeping to 100 spans when allocating

Currently, the small object sweeper will sweep until it finds a free
slot or there are no more spans of that size class to sweep. In dense
heaps, this can cause sweeping for a given size class to take
unbounded time, and gets worse with larger heaps.

This CL limits the small object sweeper to try at most 100 spans
before giving up and allocating a fresh span. Since it's already shown
that 100 spans are completely full at that point, the space overhead
of this fresh span is at most 1%.

This CL is based on an experimental CL by Austin Clements (CL 187817)
and is updated to be part of the mcentral implementation, gated by
go115NewMCentralImpl.

Updates #18155.

Change-Id: I37a72c2dcc61dd6f802d1d0eac3683e6642b6ef8
Reviewed-on: https://go-review.googlesource.com/c/go/+/229998
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Austin Clements <austin@google.com>
5 years agoruntime: add new mcentral implementation
Michael Anthony Knyszek [Thu, 20 Feb 2020 20:58:45 +0000 (20:58 +0000)]
runtime: add new mcentral implementation

Currently mcentral is implemented as a couple of linked lists of spans
protected by a lock. Unfortunately this design leads to significant lock
contention.

The span ownership model is also confusing and complicated. In-use spans
jump between being owned by multiple sources, generally some combination
of a gcSweepBuf, a concurrent sweeper, an mcentral or an mcache.

So first to address contention, this change replaces those linked lists
with gcSweepBufs which have an atomic fast path. Then, we change up the
ownership model: a span may be simultaneously owned only by an mcentral
and the page reclaimer. Otherwise, an mcentral (which now consists of
sweep bufs), a sweeper, or an mcache are the sole owners of a span at
any given time. This dramatically simplifies reasoning about span
ownership in the runtime.

As a result of this new ownership model, sweeping is now driven by
walking over the mcentrals rather than having its own global list of
spans. Because we no longer have a global list and we traditionally
haven't used the mcentrals for large object spans, we no longer have
anywhere to put large objects. So, this change also makes it so that we
keep large object spans in the appropriate mcentral lists.

In terms of the static lock ranking, we add the spanSet spine locks in
pretty much the same place as the mcentral locks, since they have the
potential to be manipulated both on the allocation and sweep paths, like
the mcentral locks.

This new implementation is turned on by default via a feature flag
called go115NewMCentralImpl.

Benchmark results for 1 KiB allocation throughput (5 runs each):

name \ MiB/s  go113       go114       gotip       gotip+this-patch
AllocKiB-1    1.71k ± 1%  1.68k ± 1%  1.59k ± 2%      1.71k ± 1%
AllocKiB-2    2.46k ± 1%  2.51k ± 1%  2.54k ± 1%      2.93k ± 1%
AllocKiB-4    4.27k ± 1%  4.41k ± 2%  4.33k ± 1%      5.01k ± 2%
AllocKiB-8    4.38k ± 3%  5.24k ± 1%  5.46k ± 1%      8.23k ± 1%
AllocKiB-12   4.38k ± 3%  4.49k ± 1%  5.10k ± 1%     10.04k ± 0%
AllocKiB-16   4.31k ± 1%  4.14k ± 3%  4.22k ± 0%     10.42k ± 0%
AllocKiB-20   4.26k ± 1%  3.98k ± 1%  4.09k ± 1%     10.46k ± 3%
AllocKiB-24   4.20k ± 1%  3.97k ± 1%  4.06k ± 1%     10.74k ± 1%
AllocKiB-28   4.15k ± 0%  4.00k ± 0%  4.20k ± 0%     10.76k ± 1%

Fixes #37487.

Change-Id: I92d47355acacf9af2c41bf080c08a8c1638ba210
Reviewed-on: https://go-review.googlesource.com/c/go/+/221182
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
5 years agoruntime: implement the spanSet data structure
Michael Anthony Knyszek [Thu, 20 Feb 2020 21:28:02 +0000 (21:28 +0000)]
runtime: implement the spanSet data structure

This change implements the spanSet data structure which is based off of
the gcSweepBuf data structure. While the general idea is the same (one
has two of these which one switches between every GC cycle; one to push
to and one to pop from), there are some key differences.

Firstly, we never have a need to iterate over this data structure so
delete numBlocks and block. Secondly, we want to be able to pop from the
front of the structure concurrently with pushes to the back. As a result
we need to maintain both a head and a tail and this change introduces an
atomic headTail structure similar to the one used by sync.Pool. It also
implements popfirst in a similar way.

As a result of this headTail, we need to be able to explicitly reset the
length, head, and tail when it goes empty at the end of sweep
termination, so add a reset method.

Updates #37487.

Change-Id: I5b8ad290ec32d591e3c8c05e496c5627018074f6
Reviewed-on: https://go-review.googlesource.com/c/go/+/221181
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
5 years agoruntime: manage a pool of spanSetBlocks and free them eagerly
Michael Anthony Knyszek [Thu, 13 Feb 2020 21:13:40 +0000 (21:13 +0000)]
runtime: manage a pool of spanSetBlocks and free them eagerly

This change adds a global pool of spanSetBlocks to the spanSet data
structure and adds support for eagerly freeing these blocks back to the
pool if the block goes empty.

This change prepares us to use this data structure in more places in the
runtime by allowing reuse of spanSetBlock.

Updates #37487.

Change-Id: I0752226e3667a9e3e1d87c9b66edaedeae1ac23f
Reviewed-on: https://go-review.googlesource.com/c/go/+/221180
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
5 years agoruntime: add spanSet data structure
Michael Anthony Knyszek [Thu, 20 Feb 2020 21:22:03 +0000 (21:22 +0000)]
runtime: add spanSet data structure

This change copies the gcSweepBuf data structure into a new file and
renames it spanSet. It will serve as the basis for a heavily modified
version of the gcSweepBuf data structure for the new mcentral
implementation.

We move it into a separate file now for two reasons:
1. We will need both implementations as they will coexist simultaneously
   for a time.
2. By creating it now in a new change it'll make future changes which
   modify it easier to review (rather than introducing the new file then).

Updates #37487.

Change-Id: If80603cab6e813a1ee2e5ecd49dcde5d8045a6c7
Reviewed-on: https://go-review.googlesource.com/c/go/+/221179
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
5 years agocmd/compile: implement multi-control branches for riscv64
Joel Sing [Mon, 30 Mar 2020 15:00:50 +0000 (02:00 +1100)]
cmd/compile: implement multi-control branches for riscv64

Implement multi-control branches for riscv64, switching to using the BNEZ
pseudo-instruction when rewriting conditionals. This will allow for further
branch optimisations to later be performed via rewrites.

Change-Id: I7f2c69f3c77494b403f26058c6bc8432d8070ad0
Reviewed-on: https://go-review.googlesource.com/c/go/+/226399
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Joel Sing <joel@sing.id.au>

5 years agodatabase/sql/driver: enhance driver package documentation
Daniel Theophanes [Mon, 27 Apr 2020 16:48:42 +0000 (09:48 -0700)]
database/sql/driver: enhance driver package documentation

Change-Id: I455acdb71354f14e0bc4104cbd7abd5c89958dd9
Reviewed-on: https://go-review.googlesource.com/c/go/+/230214
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
5 years agocmd/compile: convert more generic rules to typed aux
Josh Bleecher Snyder [Sun, 26 Apr 2020 20:02:32 +0000 (13:02 -0700)]
cmd/compile: convert more generic rules to typed aux

Passes toolstash-check.

Change-Id: I07d79d809babfeda883a49f0b21ab27ede5381cb
Reviewed-on: https://go-review.googlesource.com/c/go/+/230211
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: eliminate some array equality alg loops
Josh Bleecher Snyder [Sun, 26 Apr 2020 01:14:42 +0000 (18:14 -0700)]
cmd/compile: eliminate some array equality alg loops

type T [3]string

Prior to this change, we generated this equality alg for T:

func eqT(p, q *T) (r bool) {
    for i := range *p {
        if len(p[i]) == len(q[i]) {
        } else {
            return
        }
    }
    for j := range *p {
        if runtime.memeq(p[j].ptr, q[j].ptr, len(p[j])) {
        } else {
            return
        }
    }
    return true
}

That first loop can be profitably eliminated;
it's cheaper to spell out 3 length equality checks.

We now generate:

func eqT(p, q *T) (r bool) {
    if len(p[0]) == len(q[0]) &&
        len(p[1]) == len(q[1]) &&
        len(p[2]) == len(q[2]) {
    } else {
        return
    }
    for i := 0; i < len(p); i++ {
        if runtime.memeq(p[j].ptr, q[j].ptr, len(p[j])) {
        } else {
            return
        }
    }
    return true
}

We now also eliminate loops for small float arrays as well,
and for any array of size 1.

These cutoffs were selected to minimize code size on amd64
at this moment, for lack of a more compelling methodology.
Any smallish number would do.

The switch from range loops to plain for loops allowed me
to use a temp instead of a named var, which eliminated
a pointless argument to checkAll.
The code to construct them is also a bit clearer, in my opinion.

Change-Id: I1bdd8ee4a2739d00806e66b17a4e76b46e71231a
Reviewed-on: https://go-review.googlesource.com/c/go/+/230210
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: improve equality algs for arrays of strings
Josh Bleecher Snyder [Sat, 25 Apr 2020 22:14:35 +0000 (15:14 -0700)]
cmd/compile: improve equality algs for arrays of strings

type T [8]string

Prior to this change, we generated this equality algorithm for T:

func eqT(p, q *T) (r bool) {
    for i := range *p {
        if p[i] == q[i] {
} else {
            return
        }
    }
    return true
}

This change splits this into two loops, so that we can do the
cheap (length) half early and only then do the expensive (contents) half.

We now generate:

func eqT(p, q *T) (r bool) {
    for i := range *p {
        if len(p[i]) == len(q[i]) {
        } else {
            return
        }
    }
    for j := range *p {
        if runtime.memeq(p[j].ptr, q[j].ptr, len(p[j])) {
        } else {
            return
        }
    }
    return true
}

The generated code is typically ~17% larger because it contains
two loops instead of one. In the future, we might want to unroll
the first loop when the array is small.

Change-Id: I26b2793b90ec6aff21766a411b15a4ff1096c03f
Reviewed-on: https://go-review.googlesource.com/c/go/+/230209
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: improve equality algs for arrays of interfaces
Josh Bleecher Snyder [Fri, 24 Apr 2020 22:59:17 +0000 (15:59 -0700)]
cmd/compile: improve equality algs for arrays of interfaces

type T [8]interface{}

Prior to this change, we generated this equality algorithm for T:

func eqT(p, q *T) bool {
    for i := range *p {
        if p[i] != q[i] {
            return false
        }
    }
    return true
}

This change splits this into two loops, so that we can do the
cheap (type) half early and only then do the expensive (data) half.

We now generate:

func eqT(p, q *T) (r bool) {
    for i := range *p {
        if p[i].type == q[i].type {
        } else {
            return
        }
    }
    for j := range *p {
        if runtime.efaceeq(p[j].type, p[j].data, q[j].data) {
        } else {
            return
        }
    }
    return true
}

The use of a named return value and a bare return is to work
around some typechecking problems that stymied me.

The structure of using equals and else (instead of not equals and then)
was for implementation convenience and clarity. As a bonus,
it generates slightly shorter code on AMD64, because zeroing a register
to return is cheaper than writing $1 to it.

The generated code is typically ~17% larger because it contains
two loops instead of one. In the future, we might want to unroll
the first loop when the array is small.

Change-Id: I5b2c8dd3384852f085c4f3e1f6ad20bc5ae59062
Reviewed-on: https://go-review.googlesource.com/c/go/+/230208
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: improve generated eq algs for structs containing interfaces
Josh Bleecher Snyder [Fri, 24 Apr 2020 18:01:38 +0000 (11:01 -0700)]
cmd/compile: improve generated eq algs for structs containing interfaces

type T struct {
    s interface{}
    i int
}

Prior to this change, we generated this equality algorithm for T:

func eqT(p, q *T) bool {
    return p.s.type == q.s.type &&
      runtime.efaceeq(p.s.type, p.s.data, q.s.data) &&
      p.i == q.i
}

This change splits the two halves of the interface equality,
so that we can do the cheap (type) half early and the expensive
(data) half late. We now generate:

func eqT(p, q *T) bool {
    return p.s.type == q.s.type &&
      p.i == q.i &&
      runtime.efaceeq(p.s.type, p.s.data, q.s.data)
}

The generated code tends to be a bit smaller. Examples:

go/ast
.eq."".ForStmt 306 -> 304  (-0.65%)
.eq."".TypeAssertExpr 221 -> 219  (-0.90%)
.eq."".TypeSwitchStmt 228 -> 226  (-0.88%)
.eq."".ParenExpr 150 -> 148  (-1.33%)
.eq."".IndexExpr 221 -> 219  (-0.90%)
.eq."".SwitchStmt 228 -> 226  (-0.88%)
.eq."".RangeStmt 334 -> 332  (-0.60%)

Change-Id: Iec9e24f214ca772416202b9fb9252e625c22380e
Reviewed-on: https://go-review.googlesource.com/c/go/+/230207
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: refactor out eqinterface
Josh Bleecher Snyder [Fri, 24 Apr 2020 18:00:44 +0000 (11:00 -0700)]
cmd/compile: refactor out eqinterface

Refactor out creating the two Nodes needed to check interface equality.
Preliminary work to other optimizations.

Passes toolstash-check.

Change-Id: Id6b39e8e78f07289193423d0ef905d70826acf89
Reviewed-on: https://go-review.googlesource.com/c/go/+/230206
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: improve generated eq algs for structs containing strings
Josh Bleecher Snyder [Fri, 24 Apr 2020 16:54:13 +0000 (09:54 -0700)]
cmd/compile: improve generated eq algs for structs containing strings

type T struct {
    s string
    i int
}

Prior to this change, we generated this equality algorithm for T:

func eqT(p, q *T) bool {
    return len(p.s) == len(q.s) &&
      runtime.memequal(p.s.ptr, q.s.ptr, len(p.s)) &&
      p.i == q.i
}

This change splits the two halves of the string equality,
so that we can do the cheap (length) half early and the expensive
(contents) half late. We now generate:

func eqT(p, q *T) bool {
    return len(p.s) == len(q.s) &&
      p.i == q.i &&
      runtime.memequal(p.s.ptr, q.s.ptr, len(p.s))
}

The generated code for these functions tends to be a bit shorter. Examples:

runtime
.eq."".Frame 274 -> 272  (-0.73%)
.eq."".funcinl 249 -> 247  (-0.80%)
.eq."".modulehash 207 -> 205  (-0.97%)

Change-Id: I4efac9f7d410f0a11a94dcee2bf9c0b49b60e301
Reviewed-on: https://go-review.googlesource.com/c/go/+/230205
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: refactor out eqstring
Josh Bleecher Snyder [Fri, 24 Apr 2020 16:43:49 +0000 (09:43 -0700)]
cmd/compile: refactor out eqstring

Refactor out creating the two Nodes needed to check string equality.
Preliminary work to other optimizations.

Passes toolstash-check.

Change-Id: I72e824dac904e579b8ba9a3669a94fa1471112d2
Reviewed-on: https://go-review.googlesource.com/c/go/+/230204
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: make runtime calls last in eq algs
Josh Bleecher Snyder [Fri, 24 Apr 2020 16:49:35 +0000 (09:49 -0700)]
cmd/compile: make runtime calls last in eq algs

type T struct {
    f float64
    a [64]uint64
    g float64
}

Prior to this change, the generated equality algorithm for T was:

func eqT(p, q *T) bool {
    return p.f == q.f && runtime.memequal(p.a, q.a, 512) && p.g == q.g
}

In handwritten code, we would normally put the cheapest checks first.
This change takes a step in that direction. We now generate:

func eqT(p, q *T) bool {
    return p.f == q.f && p.g == q.g && runtime.memequal(p.a, q.a, 512)
}

For most types, this also generates considerably shorter code. Examples:

runtime
.eq."".mstats 406 -> 391  (-3.69%)
.eq.""._func 114 -> 101  (-11.40%)
.eq."".itab 115 -> 102  (-11.30%)
.eq."".scase 125 -> 116  (-7.20%)
.eq."".traceStack 119 -> 102  (-14.29%)
.eq."".gcControllerState 169 -> 161  (-4.73%)
.eq."".sweepdata 121 -> 112  (-7.44%)

However, for types in which we make unwise choices about inlining
memory-only comparisons (#38494), this generates longer code.

Example:

cmd/internal/obj
.eq."".objWriter 211 -> 214  (+1.42%)
.eq."".Addr 185 -> 187  (+1.08%)

Fortunately, such cases are not common.

Change-Id: I47a27da93c1f88ec71fa350c192f36b29548a217
Reviewed-on: https://go-review.googlesource.com/c/go/+/230203
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: remove If type in rulegen
Daniel Martí [Sun, 26 Apr 2020 20:44:36 +0000 (21:44 +0100)]
cmd/compile: remove If type in rulegen

We only generate if statements via CondBreak, which is nice as the
control flow is simple and easy to work with. It seems like the If type
was added but never used, so remove it to avoid confusion.

We had a TODO about replacing CondBreak with If instead. I gave that a
try, but it doesn't seem worth the effort. The code gets more complex
and we don't really win anything in return.

While at it, don't use op strings as format strings in exprf. This
doesn't cause any issue at the moment, but it's best to be explicit
about the operator not containing any formatting verbs.

Change-Id: Ib59ad72d3628bf91594efc609e222232ad1e8748
Reviewed-on: https://go-review.googlesource.com/c/go/+/230257
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/go/internal/web: use url.Redacted
Baokun Lee [Tue, 17 Mar 2020 01:26:07 +0000 (09:26 +0800)]
cmd/go/internal/web: use url.Redacted

Updates #37873

Change-Id: I2228f31fc7bd7daef086cd05d365fa7c68e60a83
Reviewed-on: https://go-review.googlesource.com/c/go/+/223757
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: rewrite decArgs rules to use typed aux field
Cuong Manh Le [Sat, 25 Apr 2020 07:37:33 +0000 (14:37 +0700)]
cmd/compile: rewrite decArgs rules to use typed aux field

Passes toolstash-check.

Change-Id: I386fb9d52709c4844b313f59f62a4f47c10e490d
Reviewed-on: https://go-review.googlesource.com/c/go/+/230117
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/link: use definition from debug/elf for ldelf
Meng Zhuo [Sun, 26 Apr 2020 07:47:18 +0000 (15:47 +0800)]
cmd/link: use definition from debug/elf for ldelf

Change-Id: I92d0fb3a244d0151fcc4b25a20913ad69a89f198
Reviewed-on: https://go-review.googlesource.com/c/go/+/224977
Run-TryBot: Meng Zhuo <mengzhuo1203@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agocmd/compile: convert more arm64 lowering rules to typed aux
Alberto Donizetti [Sat, 25 Apr 2020 16:46:29 +0000 (18:46 +0200)]
cmd/compile: convert more arm64 lowering rules to typed aux

Passes

  GOARCH=arm64 gotip build -toolexec 'toolstash -cmp' -a std

Change-Id: Idc0ac2638c7a9b840ba2d6f4bba2e9c5df24c807
Reviewed-on: https://go-review.googlesource.com/c/go/+/230177
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoimage/png: fix some 32-bit int overflows
Nigel Tao [Mon, 27 Apr 2020 02:23:26 +0000 (12:23 +1000)]
image/png: fix some 32-bit int overflows

Fixes #38435

Change-Id: Ib9ae3cf7f338b2860a5688e448a125f257fe624e
Reviewed-on: https://go-review.googlesource.com/c/go/+/230219
Reviewed-by: Andrew Ekstedt <andrew.ekstedt@gmail.com>
Reviewed-by: Rob Pike <r@golang.org>
5 years agocmd/compile: adopt strong aux typing for some s390x rules
Michael Munday [Thu, 23 Apr 2020 06:33:14 +0000 (23:33 -0700)]
cmd/compile: adopt strong aux typing for some s390x rules

Apply strong aux typing to lowering rules that do not require
modification beyond substituting -> for =>. Other lowering rules
and all the optimization rules will follow. I'm breaking it up
to allow toolstash-check to pass on the big CLs.

Passes toolstash-check -all.

Change-Id: I6f1340058a8eb5a1390411e59fcbea9d7f777e58
Reviewed-on: https://go-review.googlesource.com/c/go/+/229400
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: adopt strong aux typing for some s390x rules
Ruixin(Peter) Bao [Fri, 24 Apr 2020 18:52:20 +0000 (14:52 -0400)]
cmd/compile: adopt strong aux typing for some s390x rules

Convert some Move and Zero Lowering rules to strongly-typed versions.

Passes toolstash-check.

Change-Id: Icaabe05e206d59798e5883a90e9a33bb30270b13
Reviewed-on: https://go-review.googlesource.com/c/go/+/229919
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: minor rulegen simplifications
Daniel Martí [Sat, 18 Apr 2020 16:34:00 +0000 (17:34 +0100)]
cmd/compile: minor rulegen simplifications

The commuteDepth variable is no longer necessary; remove it.

Else branches after a log.Fatal call are unnecessary.

Also make the unbalanced return an integer, so we can differentiate
positive from negative cases. We only want to continue a rule with the
following lines if this balance is positive, for example.

While at it, make the balance loop stop when it goes negative, to not
let ")(" seem balanced.

Change-Id: I8aa313343ca5a2f07f638b62a0398fdf108fc9eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/228822
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
5 years agonet/http/httputil: make Switching Protocol requests (e.g. Websockets) cancelable
Pierre Carru [Sun, 26 Apr 2020 09:11:35 +0000 (09:11 +0000)]
net/http/httputil: make Switching Protocol requests (e.g. Websockets) cancelable

Ensures that a canceled client request for Switching Protocols
(e.g. h2c, Websockets) will cause the underlying connection to
be terminated.

Adds a goroutine in handleUpgradeResponse in order to select on
the incoming client request's context and appropriately cancel it.

Fixes #35559

Change-Id: I1238e18fd4cce457f034f78d9cdce0e7f93b8bf6
GitHub-Last-Rev: 3629c78493f667703ea99f9f4db5e63ddaaa0e6b
GitHub-Pull-Request: golang/go#38021
Reviewed-on: https://go-review.googlesource.com/c/go/+/224897
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
5 years agonet/http/httputil: fix typo in unit test name
Tyson Andre [Sat, 25 Apr 2020 21:39:19 +0000 (21:39 +0000)]
net/http/httputil: fix typo in unit test name

Everywhere else is using "cancellation"

The reasoning is mentioned in 170060

> Though there is variation in the spelling of canceled,
> cancellation is always spelled with a double l.
>
> Reference: https://www.grammarly.com/blog/canceled-vs-cancelled/

Change-Id: Ifc97c6785afb401814af77c377c2e2745ce53c5a
GitHub-Last-Rev: 05edd7477da46ca5c02703e139dbfc0cd05408ac
GitHub-Pull-Request: golang/go#38662
Reviewed-on: https://go-review.googlesource.com/c/go/+/230200
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoA: Qais Patankar (individual CLA)
Qais Patankar [Sat, 25 Apr 2020 19:37:24 +0000 (19:37 +0000)]
A: Qais Patankar (individual CLA)

Change-Id: I329a27d73166b70c5c13918a695d58958dbf8604
GitHub-Last-Rev: ca7be2b287c5732952f8b90f43888d38aa9ed316
GitHub-Pull-Request: golang/go#38659
Reviewed-on: https://go-review.googlesource.com/c/go/+/230197
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agomath/cmplx: fix typo in code comment
Tyson Andre [Sat, 25 Apr 2020 20:53:42 +0000 (20:53 +0000)]
math/cmplx: fix typo in code comment

Everywhere else is using "cancellation" as of 2019

The reasoning is mentioned in 170060.

> Though there is variation in the spelling of canceled,
> cancellation is always spelled with a double l.
>
> Reference: https://www.grammarly.com/blog/canceled-vs-cancelled/

Change-Id: I933ea68d7251986ce582b92c33b7cb13cee1d207
GitHub-Last-Rev: fc3d5ada2bd0087ea9cfb3f105689876e7a2ee4f
GitHub-Pull-Request: golang/go#38661
Reviewed-on: https://go-review.googlesource.com/c/go/+/230199
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agotesting: give short package variable a longer name
Brad Fitzpatrick [Sat, 25 Apr 2020 14:01:45 +0000 (07:01 -0700)]
testing: give short package variable a longer name

(Update to CL 229837)

Change-Id: Ieab46bd384f76f678ef0d6a38dc043bc4b0c458a
Reviewed-on: https://go-review.googlesource.com/c/go/+/230157
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agocmd/go/internal/cache: fix typing error in errVerifyMode
Mukesh Sharma [Sat, 25 Apr 2020 05:42:10 +0000 (05:42 +0000)]
cmd/go/internal/cache: fix typing error in errVerifyMode

This change fixes the typing mistake in errVerifyMode error message in cache.

Change-Id: I10c405a06e3396f9932db72d9de418d7f8aa013c
GitHub-Last-Rev: 14ea7c693cdff599f0c6f64906fe46c2db62a87b
GitHub-Pull-Request: golang/go#38655
Reviewed-on: https://go-review.googlesource.com/c/go/+/230097
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agotesting: replace all GOOS-specific path separators in TempDir
Tobias Klauser [Fri, 24 Apr 2020 10:17:01 +0000 (12:17 +0200)]
testing: replace all GOOS-specific path separators in TempDir

For GOOS=windows the path separator characters '\' and ':' also need be
replaced.

Updates #38465

Change-Id: If7c8cf93058c87d7df6cda140e82fd76578fe699
Reviewed-on: https://go-review.googlesource.com/c/go/+/229837
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: convert another tranch of generic rules to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 23:26:50 +0000 (16:26 -0700)]
cmd/compile: convert another tranch of generic rules to typed aux

Passes toolstash-check.

Change-Id: I27c01f03c5303876bbf6878a81ffdc4e95de3e3e
Reviewed-on: https://go-review.googlesource.com/c/go/+/230031
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoos, net: define and use os.ErrDeadlineExceeded
Ian Lance Taylor [Fri, 17 Apr 2020 22:42:12 +0000 (15:42 -0700)]
os, net: define and use os.ErrDeadlineExceeded

If an I/O operation fails because a deadline was exceeded,
return os.ErrDeadlineExceeded. We used to return poll.ErrTimeout,
an internal error, and told users to check the Timeout method.
However, there are other errors with a Timeout method that returns true,
notably syscall.ETIMEDOUT which is returned for a keep-alive timeout.
Checking errors.Is(err, os.ErrDeadlineExceeded) should permit code
to reliably tell why it failed.

This change does not affect the handling of net.Dialer.Deadline,
nor does it change the handling of net.DialContext when the context
deadline is exceeded. Those cases continue to return an error
reported as "i/o timeout" for which Timeout is true, but that error
is not os.ErrDeadlineExceeded.

Fixes #31449

Change-Id: I0323f42e944324c6f2578f00c3ac90c24fe81177
Reviewed-on: https://go-review.googlesource.com/c/go/+/228645
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
5 years agocmd/compile: avoid double-zeroing
Josh Bleecher Snyder [Thu, 23 Apr 2020 22:20:56 +0000 (15:20 -0700)]
cmd/compile: avoid double-zeroing

This triggers in 131 functions in std+cmd.
In those functions, it often helps considerably
(2-10% text size reduction).

Noticed while working on #38554.

Change-Id: Id0dbb8e7cb21d469ec08ec3d5be9beb9e8291e9c
Reviewed-on: https://go-review.googlesource.com/c/go/+/229707
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: optimize Move with all-zero ro sym src to Zero
Josh Bleecher Snyder [Thu, 23 Apr 2020 20:28:14 +0000 (13:28 -0700)]
cmd/compile: optimize Move with all-zero ro sym src to Zero

We set up static symbols during walk that
we later make copies of to initialize local variables.
It is difficult to ascertain at that time exactly
when copying a symbol is profitable vs locally
initializing an autotmp.

During SSA, we are much better placed to optimize.
This change recognizes when we are copying from a
global readonly all-zero symbol and replaces it with
direct zeroing.

This often allows the all-zero symbol to be
deadcode eliminated at link time.
This is not ideal--it makes for large object files,
and longer link times--but it is the cleanest fix I could find.

This makes the final binary for the program in #38554
shrink from >500mb to ~2.2mb.

It also shrinks the standard binaries:

file      before    after     Δ       %
addr2line 4412496   4404304   -8192   -0.186%
buildid   2893816   2889720   -4096   -0.142%
cgo       4841048   4832856   -8192   -0.169%
compile   19926480  19922432  -4048   -0.020%
cover     5281816   5277720   -4096   -0.078%
link      6734648   6730552   -4096   -0.061%
nm        4366240   4358048   -8192   -0.188%
objdump   4755968   4747776   -8192   -0.172%
pprof     14653060  14612100  -40960  -0.280%
trace     11805940  11777268  -28672  -0.243%
vet       7185560   7181416   -4144   -0.058%
total     113588440 113465560 -122880 -0.108%

And not just by removing unnecessary symbols;
the program text shrinks a bit as well.

Fixes #38554

Change-Id: I8381ae6084ae145a5e0cd9410c451e52c0dc51c8
Reviewed-on: https://go-review.googlesource.com/c/go/+/229704
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: move duffcopy auxint calculation out of rewrite rules
Josh Bleecher Snyder [Thu, 23 Apr 2020 20:11:00 +0000 (13:11 -0700)]
cmd/compile: move duffcopy auxint calculation out of rewrite rules

Package amd64 is a more natural home for it.
It also makes it easier to see how many bytes
are being copied in ssa.html.

Passes toolstash-check.

Change-Id: I5ecf0f0f18e8db2faa2caf7a05028c310952bd94
Reviewed-on: https://go-review.googlesource.com/c/go/+/229703
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert race cleanup rule to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 06:11:08 +0000 (23:11 -0700)]
cmd/compile: convert race cleanup rule to typed aux

Passes toolstash-check.

Change-Id: I3005210cc156d01a6ac1ccaafb4311c607681bf0
Reviewed-on: https://go-review.googlesource.com/c/go/+/229691
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert Move and Zero optimizations to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 06:08:59 +0000 (23:08 -0700)]
cmd/compile: convert Move and Zero optimizations to typed aux

Passes toolstash-check.

Change-Id: I9bd722ce19b2ef39931658a02663aeb7db575939
Reviewed-on: https://go-review.googlesource.com/c/go/+/229690
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert devirtualization rule to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:46:11 +0000 (22:46 -0700)]
cmd/compile: convert devirtualization rule to typed aux

Passes toolstash-check.

Change-Id: Iebcdc35f1a2112d5384c70eb3fdbd92ebb3d248e
Reviewed-on: https://go-review.googlesource.com/c/go/+/229689
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert inlineable memmove rules to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:41:18 +0000 (22:41 -0700)]
cmd/compile: convert inlineable memmove rules to typed aux

Passes toolstash-check.

Change-Id: I0190c5403040f930895a083181da2092a5c297e4
Reviewed-on: https://go-review.googlesource.com/c/go/+/229688
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert pointer and address comparisons to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:31:32 +0000 (22:31 -0700)]
cmd/compile: convert pointer and address comparisons to typed aux

Passes toolstash-check.

Change-Id: Id4c4d341e5733673eb8a899e881d70b193f76580
Reviewed-on: https://go-review.googlesource.com/c/go/+/229687
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert nilcheck elim rules to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:27:23 +0000 (22:27 -0700)]
cmd/compile: convert nilcheck elim rules to typed aux

Passes toolstash-check.

Change-Id: Ic7efb0e4778844366f581c6310a1a2f3bfc1868a
Reviewed-on: https://go-review.googlesource.com/c/go/+/229686
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert floating point optimizations to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:22:00 +0000 (22:22 -0700)]
cmd/compile: convert floating point optimizations to typed aux

Passes toolstash-check.

Change-Id: I1318ede351da4cf769f7b9d87b275720fc278159
Reviewed-on: https://go-review.googlesource.com/c/go/+/229685
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert reassociation optimizations to typed aux, part two
Josh Bleecher Snyder [Fri, 24 Apr 2020 05:13:29 +0000 (22:13 -0700)]
cmd/compile: convert reassociation optimizations to typed aux, part two

Passes toolstash-check.

Change-Id: Ia8fad6973983eebe6d78d9dd8de8c99b8edcecdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/229684
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert reassociation optimizations to typed aux
Josh Bleecher Snyder [Fri, 24 Apr 2020 02:46:49 +0000 (19:46 -0700)]
cmd/compile: convert reassociation optimizations to typed aux

Passes toolstash-check.

Change-Id: I85a161c4e1c29e06e3f7e6ad4a59ff8cc51c7296
Reviewed-on: https://go-review.googlesource.com/c/go/+/229683
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoruntime: sleep in TestSegv program to let signal be delivered
Ian Lance Taylor [Wed, 22 Apr 2020 23:03:30 +0000 (16:03 -0700)]
runtime: sleep in TestSegv program to let signal be delivered

Since we're sleeping rather than waiting for the goroutines,
let the goroutines run forever.

Fixes #38595

Change-Id: I4cd611fd7565f6e8d91e50c9273d91c514825314
Reviewed-on: https://go-review.googlesource.com/c/go/+/229484
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
5 years agocmd/compile: use typed aux for first half of arm64 lowering
Alberto Donizetti [Fri, 24 Apr 2020 15:24:16 +0000 (17:24 +0200)]
cmd/compile: use typed aux for first half of arm64 lowering

Passes

  GOARCH=arm64 gotip build -toolexec 'toolstash -cmp' -a std

Change-Id: Icb530d8d128d9938ab44a9c716c8dd09a34ededf
Reviewed-on: https://go-review.googlesource.com/c/go/+/229937
Reviewed-by: Keith Randall <khr@golang.org>
5 years agodoc/go1.15: include changes in net/http/pprof and runtime/pprof
Hana (Hyang-Ah) Kim [Fri, 24 Apr 2020 20:57:07 +0000 (16:57 -0400)]
doc/go1.15: include changes in net/http/pprof and runtime/pprof

net/http/pprof: delta profile support
runtime/pprof: profile labels plumbing for goroutine profiles

Change-Id: I92e750dc894c8c6b3c3ba10f7be58bb541d3c289
Reviewed-on: https://go-review.googlesource.com/c/go/+/230023
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agointernal/goversion: revert "update to 1.15"
Heschi Kreinick [Fri, 24 Apr 2020 22:00:21 +0000 (22:00 +0000)]
internal/goversion: revert "update to 1.15"

This reverts CL 230024, commit 5e10ba996915f25e21808eee6baff950d2895634.

Reason for revert: breaks cmd/go TestScript/mod_retention

Change-Id: I2044beff3008156dd11d7bd8154a6208ae692c57
Reviewed-on: https://go-review.googlesource.com/c/go/+/230029
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agointernal/goversion: update to 1.15
Heschi Kreinick [Fri, 24 Apr 2020 20:57:40 +0000 (16:57 -0400)]
internal/goversion: update to 1.15

Tests tagged +build go1.15 are currently not running. They should.

Change-Id: Ib97ec57a7a35cea65e2d14fb2b067e5fe49ee284
Reviewed-on: https://go-review.googlesource.com/c/go/+/230024
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
5 years agocmd/compile: add more non-ID comparisons to schedule
Josh Bleecher Snyder [Thu, 23 Apr 2020 23:22:37 +0000 (16:22 -0700)]
cmd/compile: add more non-ID comparisons to schedule

These comparisons are fairly arbitrary,
but they should be more stable in the face
of other compiler changes than value ID.

This reduces the number of value ID
comparisons in schedule while running
make.bash from 542,442 to 99,703.

There are lots of changes to generated code
from this change, but they appear to
be overall neutral.

It is possible to further reduce the
number of comparisons in schedule;
I have changes locally that reduce the
number to about 25,000 during make.bash.
However, the changes are increasingly
complex and arcane, and reduce in much less
code churn. Given that the goal is stability,
that suggests that this is a reasonable
place to stop, at least for now.

Change-Id: Ie3a75f84fd3f3fdb102fcd0b29299950ea66b827
Reviewed-on: https://go-review.googlesource.com/c/go/+/229799
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: add Value.Uses comparison during scheduling
Josh Bleecher Snyder [Thu, 23 Apr 2020 22:30:19 +0000 (15:30 -0700)]
cmd/compile: add Value.Uses comparison during scheduling

Falling back to comparing Value.ID during scheduling
is undesirable: Not only are we simply hoping for a good
outcome, but the decision we make will be easily perturbed
by other compiler changes, leading to random fluctuations.

This change adds another decision point to the scheduler
by scheduling Values with many uses earlier.
Values with fewer uses are less likely to be spilled for
other reasons, so we should issue them as late as possible
in the hope of avoiding a spill.

This reduces the number of Value ID comparisons
in schedule while running make.bash
from 1,000,844 to 542,442.

As you would expect, this changes a lot of functions,
but the overall trend is positive:

file    before    after     Δ       %
api     5237184   5233088   -4096   -0.078%
compile 19926480  19918288  -8192   -0.041%
cover   5281816   5277720   -4096   -0.078%
dist    3711608   3707512   -4096   -0.110%
total   113588440 113567960 -20480  -0.018%

Change-Id: Ic99ebc4c614d4ae3807ce44473ec6b04684388ec
Reviewed-on: https://go-review.googlesource.com/c/go/+/229798
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agogo/types: improve error message for pointer receiver errors
Rebecca Stambler [Fri, 24 Apr 2020 05:01:48 +0000 (01:01 -0400)]
go/types: improve error message for pointer receiver errors

The compiler produces high quality error messages when an interface is
implemented by *T, rather than T. This change improves the analogous
error messages in go/types, from "missing method X" to "missing method
X (X has pointer receiver)".

I am open to improving this message further - I didn't copy the compiler
error message exactly because, at one of the call sites of
(*check).missingMethod, we no longer have access to the name of the
interface.

Fixes golang/go#36336

Change-Id: Ic4fc38b13fff9e5d9a69cc750c21e0b0c34d85a8
Reviewed-on: https://go-review.googlesource.com/c/go/+/229801
Run-TryBot: Rebecca Stambler <rstambler@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
5 years agoMerge "cmd: merge branch 'dev.link' into master"
Gerrit Code Review [Fri, 24 Apr 2020 17:13:12 +0000 (17:13 +0000)]
Merge "cmd: merge branch 'dev.link' into master"

5 years agocmd/compile: CSE the RHS of rewrite rules
Josh Bleecher Snyder [Fri, 24 Apr 2020 02:22:26 +0000 (19:22 -0700)]
cmd/compile: CSE the RHS of rewrite rules

Keep track of all expressions encountered while
generating a rewrite result, and re-use them whenever possible.
Named expressions may still be used for clarity when desired.

Change-Id: I640dca108763eb8baeff8f9a4169300af3445b82
Reviewed-on: https://go-review.googlesource.com/c/go/+/229800
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
5 years agomath/big: rewrite subVW to use fast path on s390x
Ruixin(Peter) Bao [Tue, 24 Mar 2020 15:51:22 +0000 (11:51 -0400)]
math/big: rewrite subVW to use fast path on s390x

This CL replaces the original subVW implementation with a implementation
that uses a similar idea as CL 164968.

When we know the borrow bit is zero, we can copy the rest of words as
they will not be updated. Also, since we are copying vector of a words,
a faster implementation of copy is written in this CL to copy a word or
multiple words at a time.

Benchmarks:
name             old time/op    new time/op     delta
SubVW/1-18         4.43ns ± 0%     3.82ns ± 0%   -13.85%  (p=0.000 n=20+20)
SubVW/2-18         5.39ns ± 0%     4.25ns ± 0%   -21.23%  (p=0.000 n=20+20)
SubVW/3-18         6.29ns ± 0%     4.65ns ± 0%   -26.07%  (p=0.000 n=16+19)
SubVW/4-18         6.08ns ± 2%     4.84ns ± 0%   -20.43%  (p=0.000 n=20+20)
SubVW/5-18         7.06ns ± 1%     4.93ns ± 0%   -30.18%  (p=0.000 n=20+20)
SubVW/10-18        10.3ns ± 2%      7.2ns ± 0%   -30.35%  (p=0.000 n=20+19)
SubVW/100-18       48.0ns ± 4%     17.6ns ± 0%   -63.32%  (p=0.000 n=18+19)
SubVW/1000-18       448ns ±10%      236ns ± 1%   -47.24%  (p=0.000 n=20+20)
SubVW/10000-18     4.83µs ± 5%     2.96µs ± 0%   -38.73%  (p=0.000 n=20+19)
SubVW/100000-18    46.6µs ± 3%     30.6µs ± 1%   -34.30%  (p=0.000 n=20+20)
[Geo mean]         56.3ns          37.0ns        -34.24%

name             old speed      new speed       delta
SubVW/1-18       1.80GB/s ± 0%   2.10GB/s ± 0%   +16.16%  (p=0.000 n=20+20)
SubVW/2-18       2.97GB/s ± 0%   3.77GB/s ± 0%   +26.95%  (p=0.000 n=20+20)
SubVW/3-18       3.82GB/s ± 0%   5.16GB/s ± 0%   +35.26%  (p=0.000 n=20+19)
SubVW/4-18       5.26GB/s ± 1%   6.61GB/s ± 0%   +25.59%  (p=0.000 n=20+20)
SubVW/5-18       5.67GB/s ± 1%   8.11GB/s ± 0%   +43.12%  (p=0.000 n=20+20)
SubVW/10-18      7.79GB/s ± 2%  11.17GB/s ± 0%   +43.52%  (p=0.000 n=20+19)
SubVW/100-18     16.7GB/s ± 4%   45.5GB/s ± 0%  +172.61%  (p=0.000 n=18+20)
SubVW/1000-18    17.9GB/s ± 9%   33.9GB/s ± 1%   +89.25%  (p=0.000 n=20+20)
SubVW/10000-18   16.6GB/s ± 5%   27.0GB/s ± 0%   +63.08%  (p=0.000 n=20+19)
SubVW/100000-18  17.2GB/s ± 2%   26.1GB/s ± 1%   +52.18%  (p=0.000 n=20+20)
[Geo mean]       7.25GB/s       11.03GB/s        +52.01%

Change-Id: I32e99cbab3260054a96231d02b87049c833ab77e
Reviewed-on: https://go-review.googlesource.com/c/go/+/227297
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agodoc: remove The Go Project page (moved to x/website)
Dmitri Shuralyov [Wed, 22 Apr 2020 23:31:13 +0000 (19:31 -0400)]
doc: remove The Go Project page (moved to x/website)

This page has moved to the x/website repo in CL 229482 (commit
golang/website@70f4ee8c7e31e1c90ae44d835ff6214dc0496205).
Remove the old copy in this repo since it's no longer used.

For #29206.

Change-Id: Ief093ed8c5dfec43e06d473e4282275f61da74a5
Reviewed-on: https://go-review.googlesource.com/c/go/+/229485
Reviewed-by: Alexander Rakoczy <alex@golang.org>
5 years agocmd: merge branch 'dev.link' into master
Cherry Zhang [Fri, 24 Apr 2020 14:30:33 +0000 (10:30 -0400)]
cmd: merge branch 'dev.link' into master

In the dev.link branch we continued developing the new object
file format support and the linker improvements described in
https://golang.org/s/better-linker . Since the last merge, more
progress has been made to improve the new linker.

This is a clean merge.

Change-Id: I57c510b651a39354d78478a9a4499f770eef2eb1

5 years agomath/big: rewrite addVW to use fast path on s390x
Ruixin(Peter) Bao [Tue, 3 Dec 2019 21:06:18 +0000 (16:06 -0500)]
math/big: rewrite addVW to use fast path on s390x

Rewrite addVW to use a fast path and remove the original
vector and non vector implementation of addVW in assembly. This CL uses
a similar idea as CL 164968, where we copy the rest of words when we
know carry bit is zero.

In addition, since we are copying vector of words, a faster
implementation of copy is written in this CL to copy a word or multiple
words at a time.

Benchmarks:
name             old time/op    new time/op     delta
AddVW/1-18         4.56ns ± 0%     4.01ns ± 6%   -12.14%  (p=0.000 n=18+20)
AddVW/2-18         5.54ns ± 0%     4.42ns ± 5%   -20.20%  (p=0.000 n=18+20)
AddVW/3-18         6.55ns ± 0%     4.61ns ± 0%   -29.62%  (p=0.000 n=16+18)
AddVW/4-18         6.11ns ± 2%     5.12ns ± 6%   -16.19%  (p=0.000 n=20+20)
AddVW/5-18         7.32ns ± 4%     5.14ns ± 0%   -29.77%  (p=0.000 n=20+19)
AddVW/10-18        10.6ns ± 2%      7.2ns ± 1%   -31.47%  (p=0.000 n=20+20)
AddVW/100-18       49.6ns ± 2%     18.0ns ± 0%   -63.63%  (p=0.000 n=20+20)
AddVW/1000-18       465ns ± 3%      244ns ± 0%   -47.54%  (p=0.000 n=20+20)
AddVW/10000-18     4.99µs ± 4%     2.97µs ± 0%   -40.54%  (p=0.000 n=20+20)
AddVW/100000-18    48.3µs ± 3%     30.8µs ± 1%   -36.29%  (p=0.000 n=20+20)
[Geo mean]         58.1ns          38.0ns        -34.57%

name             old speed      new speed       delta
AddVW/1-18       1.76GB/s ± 0%   2.00GB/s ± 6%   +14.04%  (p=0.000 n=20+20)
AddVW/2-18       2.89GB/s ± 0%   3.63GB/s ± 5%   +25.55%  (p=0.000 n=18+20)
AddVW/3-18       3.66GB/s ± 0%   5.21GB/s ± 0%   +42.25%  (p=0.000 n=18+19)
AddVW/4-18       5.24GB/s ± 2%   6.27GB/s ± 6%   +19.61%  (p=0.000 n=20+20)
AddVW/5-18       5.47GB/s ± 4%   7.78GB/s ± 0%   +42.28%  (p=0.000 n=20+18)
AddVW/10-18      7.55GB/s ± 2%  11.04GB/s ± 1%   +46.09%  (p=0.000 n=20+20)
AddVW/100-18     16.1GB/s ± 2%   44.3GB/s ± 0%  +174.77%  (p=0.000 n=20+20)
AddVW/1000-18    17.2GB/s ± 3%   32.8GB/s ± 1%   +90.58%  (p=0.000 n=20+20)
AddVW/10000-18   16.0GB/s ± 4%   26.9GB/s ± 0%   +68.11%  (p=0.000 n=20+20)
AddVW/100000-18  16.6GB/s ± 3%   26.0GB/s ± 1%   +56.94%  (p=0.000 n=20+20)
[Geo mean]       7.03GB/s       10.75GB/s        +52.93%

Change-Id: Idbb73f3178311bd2b18a93bdc1e48f26869d2f6a
Reviewed-on: https://go-review.googlesource.com/c/go/+/209679
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: rewrite Wasm rules to use typed aux fields
Richard Musiol [Fri, 17 Apr 2020 21:06:49 +0000 (23:06 +0200)]
cmd/compile: rewrite Wasm rules to use typed aux fields

Passes toolstash-check -all.

Change-Id: Ib731a59eadfffa81914848005b0f757649affa6f
Reviewed-on: https://go-review.googlesource.com/c/go/+/228819
Run-TryBot: Richard Musiol <neelance@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: convert remaining mips rules to typed aux
Alberto Donizetti [Thu, 23 Apr 2020 09:12:04 +0000 (11:12 +0200)]
cmd/compile: convert remaining mips rules to typed aux

Passes

  GOARCH=mips gotip build -toolexec 'toolstash -cmp' -a std
  GOARCH=mipsle gotip build -toolexec 'toolstash -cmp' -a std

Change-Id: I35df0522e299aa755491cd25f47f1f1bf447848c
Reviewed-on: https://go-review.googlesource.com/c/go/+/229637
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocrypto/x509: add x509omitbundledroots build tag to not embed roots
Brad Fitzpatrick [Thu, 23 Apr 2020 20:58:46 +0000 (13:58 -0700)]
crypto/x509: add x509omitbundledroots build tag to not embed roots

On darwin/arm64, the copy of the system roots takes 256 KiB of disk
and 560 KiB of memory after parsing them (which is retained forever in
a package global by x509/root.go). In constrained environments like
iOS NetworkExtensions where total disk+RAM is capped at 15 MiB, these
certs take 5.3% of the total allowed memory.

It turns out you can get down from 816 KiB to 110 KiB by instead
storing compressed x509 certs in the binary and lazily inflating just
the needed certs at runtime as a function of the certs presented to
you by the server, then building a custom root CertPool in the
crypto/tls.Config.VerifyPeerCertificate hook.

This then saves 706 KiB.

Arguably that should be the default Go behavior, but involves
cooperation between x509 and tls, and adds a dependency to
compress/gzip. Also, it may not be the right trade-off for everybody,
as it involves burning more CPU on new TLS connections. Most iOS apps
don't run in a NetworkExtension context limiting them to 15 MiB.

The build tag is chosen to match the existing "nethttpomithttp2".

Change-Id: I7b1c845de08b22674f81dd546e7fadc7dda68bd7
Reviewed-on: https://go-review.googlesource.com/c/go/+/229762
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
5 years agocmd/compile: fix misalignment in sources column of generated ssa.html
Bradford Lamson-Scribner [Fri, 24 Apr 2020 00:46:43 +0000 (18:46 -0600)]
cmd/compile: fix misalignment in sources column of generated ssa.html

Fix regression where line numbers in the sources column of generated
ssa.html output became misaligned with the source code. This was due
to some new margins applied to certain h2 elements during the work
to combine identical columns.

Fixes #38612

Change-Id: I067ccbfa30d5de5be29aab9863bc1e21f6ded128
Reviewed-on: https://go-review.googlesource.com/c/go/+/229766
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/internal/obj: add IsAsm flag
David Chase [Thu, 23 Apr 2020 23:52:31 +0000 (19:52 -0400)]
cmd/internal/obj: add IsAsm flag

This allows more exciting changes to compiler-generated assembly
language that might not be correct for tricky hand-crafted
assembly (e.g., nop padding breaking tables of call or branch
instructions).

Updates #35881

Change-Id: I842b811796076c160180a364564f2844604df3fb
Reviewed-on: https://go-review.googlesource.com/c/go/+/229708
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agostrings: remove an obsolete doc note for FieldsFunc
Praveen Kumar [Thu, 23 Apr 2020 21:49:53 +0000 (21:49 +0000)]
strings: remove an obsolete doc note for FieldsFunc

Fixes #38630

Change-Id: I0b2b693dd88821dcfc035cf552b687565bb55ef6
GitHub-Last-Rev: 291b1b4dcf074860492bb67008f4e8228c655652
GitHub-Pull-Request: golang/go#38631
Reviewed-on: https://go-review.googlesource.com/c/go/+/229763
Reviewed-by: Robert Griesemer <gri@golang.org>
5 years agocmd/compile: remove ODDDARG
Matthew Dempsky [Thu, 23 Apr 2020 18:37:33 +0000 (11:37 -0700)]
cmd/compile: remove ODDDARG

No longer needed after the last CL. Separate commit because
renumbering Ops causes toolstash to complain.

Change-Id: I6223a790cc341f8184eccb503f95a1dfc32a81e4
Reviewed-on: https://go-review.googlesource.com/c/go/+/229760
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: use fixVariadicCall in escape analysis
Matthew Dempsky [Wed, 22 Apr 2020 02:48:02 +0000 (19:48 -0700)]
cmd/compile: use fixVariadicCall in escape analysis

This CL uses fixVariadicCall before escape analyzing function calls.
This has a number of benefits, though also some minor obstacles:

Most notably, it allows us to remove ODDDARG along with the logic
involved in setting it up, manipulating EscHoles, and later copying
its escape analysis flags to the actual slice argument. Instead, we
uniformly handle all variadic calls the same way. (E.g., issue31573.go
is updated because now f() and f(nil...) are handled identically.)

It also allows us to simplify handling of builtins and generic
function calls. Previously handling of calls was hairy enough to
require multiple dispatches on n.Op, whereas now the logic is uniform
enough that we can easily handle it with a single dispatch.

The downside is handling //go:uintptrescapes is now somewhat clumsy.
(It used to be clumsy, but it still is, too.) The proper fix here is
probably to stop using escape analysis tags for //go:uintptrescapes
and unsafe-uintptr, and have an earlier pass responsible for them.

Finally, note that while we now call fixVariadicCall in Escape, we
still have to call it in Order, because we don't (yet) run Escape on
all compiler-generated functions. In particular, the generated "init"
function for initializing package-level variables can contain calls to
variadic functions and isn't escape analyzed.

Passes toolstash-check -race.

Change-Id: I4cdb92a393ac487910aeee58a5cb8c1500eef881
Reviewed-on: https://go-review.googlesource.com/c/go/+/229759
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
5 years agogo/types: add UsesCgo config to support _cgo_gotypes.go
Matthew Dempsky [Wed, 30 Nov 2016 01:27:15 +0000 (17:27 -0800)]
go/types: add UsesCgo config to support _cgo_gotypes.go

This CL adds a UsesCgo config setting to go/types to specify that the
_cgo_gotypes.go file generated by cmd/cgo has been provided as a
source file. The type checker then internally resolves C.bar qualified
identifiers to _Cfoo_bar as appropriate.

It also adds support to srcimporter to automatically run cgo.
Unfortunately, this functionality is not compatible with overriding
OpenFile, because cmd/cgo and gcc will directly open files.

Updates #16623.
Updates #35721.

Change-Id: I1e1965fe41b765b7a9da3431f2a86cc16025dee2
Reviewed-on: https://go-review.googlesource.com/c/go/+/33677
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
5 years agocmd/compile: clean up PPC64.rules typed aux changes
Lynn Boger [Thu, 23 Apr 2020 18:00:33 +0000 (14:00 -0400)]
cmd/compile: clean up PPC64.rules typed aux changes

This adds a few minor changes from the first review.

Passes toolstash-check

Change-Id: I00f6f1b0235d0a8c686aa8793d0473b8fc6b1495
Reviewed-on: https://go-review.googlesource.com/c/go/+/229699
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: remove dead values after flagalloc
Josh Bleecher Snyder [Mon, 30 Dec 2019 21:16:28 +0000 (13:16 -0800)]
cmd/compile: remove dead values after flagalloc

Fix a longstanding TODO.

Provides widespread, minor improvements.
Negligible compiler cost.

Because the freeze nears, put in a safety flag to easily disable.

Change-Id: I338812181ab6d806fecf22afd3c3502e2c94f7a0
Reviewed-on: https://go-review.googlesource.com/c/go/+/229600
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/compile: splitload (CMPconst [0] x) into (TEST x x) on amd64
Josh Bleecher Snyder [Thu, 23 Apr 2020 19:00:14 +0000 (12:00 -0700)]
cmd/compile: splitload (CMPconst [0] x) into (TEST x x) on amd64

Provides minor widespread benefit to generated code.

Removes one source of random fluctuation when changing
other aspects of the compiler.

Change-Id: I16db6f5e240a97d27f05dc1ba5b8b729af3adb12
Reviewed-on: https://go-review.googlesource.com/c/go/+/229702
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: allow named values on RHS of rewrite rules
Josh Bleecher Snyder [Thu, 23 Apr 2020 07:30:14 +0000 (00:30 -0700)]
cmd/compile: allow named values on RHS of rewrite rules

Fixes #38621

Change-Id: Idbffdcc70903290dc58e5abb4867718bd5449fe1
Reviewed-on: https://go-review.googlesource.com/c/go/+/229701
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocmd/asm,cmd/internal/obj/ppc64: update instructions and tests
Lynn Boger [Wed, 22 Apr 2020 17:23:06 +0000 (13:23 -0400)]
cmd/asm,cmd/internal/obj/ppc64: update instructions and tests

This change adds some instructions that were missing from the
ppc64 assembler, mostly power9 but a few others from earlier.

Tests in cmd/asm for ppc64 were updated: ppc64.s includes the
new instructions, and ppc64enc.s now includes not only the
new instructions but most ppc64 opcodes to provide a more
complete test of the ppc64 assembler.

The ppc64 instruction set is used for linux/ppc64le,
linux/ppc64, and aix/ppc64.

Change-Id: I8695f89dbca06174847963f4ef869f2e584d5bbf
Reviewed-on: https://go-review.googlesource.com/c/go/+/229479
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>