Clément Chigot [Wed, 20 Feb 2019 15:48:43 +0000 (16:48 +0100)]
cmd/link: fix trampoline generation for aix/ppc64
This commit fixes trampoline generation on aix/ppc64 which must use TOC
symbols.
It also adds a size to runtime.text.X symbols to prevent ld from moving
them, like runtime.text.
Change-Id: Ida033ec20ad8d7b7fb3faeb0ec4fa7bc4ce86b7e
Reviewed-on: https://go-review.googlesource.com/c/go/+/164009
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Ian Lance Taylor [Thu, 14 Mar 2019 01:56:37 +0000 (18:56 -0700)]
runtime: introduce and consistently use setNsec for timespec
The general code for setting a timespec value sometimes used set_nsec
and sometimes used a combination of set_sec and set_nsec. Standardize
on a setNsec function that takes a number of nanoseconds and splits
them up to set the tv_sec and tv_nsec fields. Consistently mark
setNsec as go:nosplit, since it has to be that way on some systems
including Darwin and GNU/Linux. Consistently use timediv on 32-bit
systems to help stay within split-stack limits on processors that
don't have a 64-bit division instruction.
Change-Id: I6396bb7ddbef171a96876bdeaf7a1c585a6d725b
Reviewed-on: https://go-review.googlesource.com/c/go/+/167389
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Matthew Dempsky [Thu, 14 Mar 2019 20:54:05 +0000 (13:54 -0700)]
cmd/compile: move Strongly Connected Components code into new file
This logic is used by the current escape analysis pass, but otherwise
logically independent. Move (unchanged) into a separate file to make
that clearer, and to make it easier to replace esc.go later.
Updates #23109.
Change-Id: Iec8c0c47ea04c0008165791731c11d9104d5a474
Reviewed-on: https://go-review.googlesource.com/c/go/+/167715 Reviewed-by: Robert Griesemer <gri@golang.org>
Matthew Dempsky [Wed, 12 Dec 2018 19:15:37 +0000 (11:15 -0800)]
cmd/compile: rewrite f(g()) for multi-value g() during typecheck
This is a re-attempt at CL 153841, which caused two regressions:
1. crypto/ecdsa failed to build with -gcflags=-l=4. This was because
when "t1, t2, ... := g(); f(t1, t2, ...)" was exported, we were losing
the first assignment from the call's Ninit field.
2. net/http/pprof failed to run with -gcflags=-N. This is due to a
conflict with CL 159717: as of that CL, package-scope initialization
statements are executed within the "init.ializer" function, rather
than the "init" function, and the generated temp variables need to be
moved accordingly too.
[Rest of description is as before.]
This CL moves order.go's copyRet logic for rewriting f(g()) into t1,
t2, ... := g(); f(t1, t2, ...) earlier into typecheck. This allows the
rest of the compiler to stop worrying about multi-value functions
appearing outside of OAS2FUNC nodes.
This changes compiler behavior in a few observable ways:
1. Typechecking error messages for builtin functions now use general
case error messages rather than unnecessarily differing ones.
2. Because f(g()) is rewritten before inlining, saved inline bodies
now see the rewritten form too. This could be addressed, but doesn't
seem worthwhile.
3. Most notably, this simplifies escape analysis and fixes a memory
corruption issue in esc.go. See #29197 for details.
Fixes #15992.
Fixes #29197.
Change-Id: I930b10f7e27af68a0944d6c9bfc8707c3fab27a4
Reviewed-on: https://go-review.googlesource.com/c/go/+/166983
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Richard Musiol [Tue, 12 Mar 2019 22:40:03 +0000 (23:40 +0100)]
misc/wasm: add workaround for missed timeout events
TryBot is sometimes running into deadlocks on js/wasm. We haven't been
able to reproduce them yet. This workaround is an experiment to resolve
these deadlocks by retrying a missed timeout event.
A timeout event is scheduled by Go to be woken by JavaScript after a
certain amount of time. The checkTimeouts function then checks which
notes to wake by comparing their deadline to nanotime. If this
check fails erroneously then the note may stay asleep forever, causing
a deadlock. This may or may not be the reason of the observed
deadlocks.
Carlos Eduardo Seo [Thu, 31 Jan 2019 17:22:21 +0000 (11:22 -0600)]
cmd/internal/obj/ppc64: fix wrong register encoding in XX1-Form instructions
A bug in the encoding of XX1-Form is flipping bit 31 of such instructions.
This may result in register clobering when using VSX instructions.
This was not exposed before because we currently don't generate these
instructions in SSA, and the asm files in which they are present aren't
affected by register clobbering.
This change fixes the bug and adds a testcase for the problem.
Russ Cox [Thu, 14 Mar 2019 00:06:37 +0000 (20:06 -0400)]
crypto/x509: move debug prints to standard error
Standard output is reserved for actual program output.
Debug print should be limited in general (here they are
enabled by an environment variable) and always go to
standard error.
Russ Cox [Thu, 7 Mar 2019 03:24:54 +0000 (22:24 -0500)]
fmt: put back named results in ss.scanBasePrefix
CL 165619 removed these names when it removed
the use of the plain 'return'. But the names help for
documentation purposes even without being mentioned
directly in the function, so removing them makes the
code less readable. Put them back. I renamed found
to zeroFound to make the meaning clearer.
Change-Id: I1010931f08290af0b0ede7d21b1404c2eea196a0
Reviewed-on: https://go-review.googlesource.com/c/go/+/165899
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Daniel Martí [Wed, 13 Mar 2019 17:57:03 +0000 (17:57 +0000)]
builtin: make len's godoc less ambiguous
The len godoc uses a blockquote to list the rules for its semantics.
The item that describes channels is a bit long, so it's split in two
lines. However, the first line ends with a semicolon, and the second
line can be read as a sentence of its own, so it's easy to misinterpret
that the two lines are separate.
Making that easy mistake would lead to an incorrect understanding of
len:
if v is nil, len(v) is zero.
This could lead us to think that len(nil) is valid and should return
zero. When in fact, that statement only applies to nil channels.
To make this less ambiguous, add a bit of indentation to the follow-up
line, to align with the channel body. If lists are added to godoc in the
future via #7873, perhaps this text can be simplified.
Hana Kim [Mon, 14 Jan 2019 16:13:34 +0000 (11:13 -0500)]
cmd/go: fix the default build output name for versioned binaries
`go build` has chosen the last element of the package import path
as the default output name when -o option is given. That caused
the output of a package build when the module root is the major
version component such as 'v2'.
A similar issue involving `go install` was fixed in
https://golang.org/cl/128900. This CL refactors the logic added
with the change and makes it available as
internal/load.DefaultExecName.
This CL makes 'go test' to choose the right default test binary
name when the tested package is in the module root. (E.g.,
instead of v2.test, choose pkg.test for the test of 'path/pkg/v2')
Fixes #27283.
Change-Id: I6905754f0906db46e3ce069552715f45356913ae
Reviewed-on: https://go-review.googlesource.com/c/go/+/140863 Reviewed-by: Bryan C. Mills <bcmills@google.com>
Matthew Dempsky [Thu, 28 Feb 2019 02:18:47 +0000 (18:18 -0800)]
cmd/compile: simplify isGoConst
The only ways to construct an OLITERAL node are (1) a basic literal
from the source package, (2) constant folding within evconst (which
only folds Go language constants), (3) the universal "nil" constant,
and (4) implicit conversions of nil to some concrete type.
Passes toolstash-check.
Change-Id: I30fc6b07ebede7adbdfa4ed562436cbb7078a2ff
Reviewed-on: https://go-review.googlesource.com/c/go/+/166981
Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Robert Griesemer <gri@golang.org>
Carlos Eduardo Seo [Fri, 11 Jan 2019 19:16:28 +0000 (17:16 -0200)]
cmd/compile: add processor level selection support to ppc64{,le}
ppc64{,le} processor level selection allows the compiler to generate instructions
targeting newer processors and processor-specific optimizations without breaking
compatibility with our current baseline. This feature introduces a new environment
variable, GOPPC64.
GOPPC64 is a GOARCH=ppc64{,le} specific option, for a choice between different
processor levels (i.e. Instruction Set Architecture versions) for which the
compiler will target. The default is 'power8'.
Change-Id: Ic152e283ae1c47084ece4346fa002a3eabb3bb9e
Reviewed-on: https://go-review.googlesource.com/c/go/+/163758
Run-TryBot: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
David Chase [Mon, 28 Jan 2019 23:00:01 +0000 (18:00 -0500)]
cmd/compile: move statement marks from jumps to targets
When a jump at the end of a block is about to be marked as
a statement, if the first real instruction in the target
block is also a statement for the same line, remove the
mark from the jump.
This is a first effort at a minimal-harm heuristic.
A better heuristic might skip over any "not-statement"
values preceding a definitely marked value.
Daniel Martí [Sun, 3 Mar 2019 17:24:57 +0000 (17:24 +0000)]
cmd/go/internal/base: remove MergeEnvLists
This internally exported function allowed merging environment variable
lists, and was mostly a convenience for the rest of cmd/go/internal.
It seems to date all the way back to 2013.
However, since CL 37586 in early 2017, os/exec has already taken care of
deduplicating environment variable lists. Thus, it's unnecessary for
cmd/go to take care of that before calling exec.Cmd.Start.
Moreover, because os/exec will deduplicate the list in any case, we're
adding extra work in all these scenarios.
Finally, remove an unnecessary addition of GOROOT= in internal/tool.
cfg.OrigEnv may not have the correct GOROOT set up, but os.Environ does;
cmd/go's main function makes sure of that.
Daniel Martí [Sun, 25 Nov 2018 17:30:36 +0000 (17:30 +0000)]
encoding/base64: speed up the decoder
Most of the decoding time is spent in the first Decode loop, since the
rest of the function only deals with the few remaining bytes. Any
unnecessary work done in that loop body matters tremendously.
One such unnecessary bottleneck was the use of the enc.decodeMap table.
Since enc is a pointer receiver, and the field is used within the
non-inlineable function decode64, the decoder must perform a nil check
at every iteration.
To fix that, move the enc.decodeMap uses to the parent function, where
we can lift the nil check outside the loop. That gives roughly a 15%
speed-up. The function no longer performs decoding per se, so rename it.
While at it, remove the now unnecessary receivers.
An unfortunate side effect of this change is that the loop now contains
eight bounds checks on src instead of just one. However, not having to
slice src plus the nil check removal well outweigh the added cost.
The other piece that made decode64 slow was that it wasn't inlined, and
had multiple branches. Use a simple bitwise-or trick suggested by Roger
Peppe, and collapse the rest of the bitwise logic into a single
expression. Inlinability and the reduced branching give a further 10%
speed-up.
Finally, add these two functions to TestIntendedInlining, since we want
them to stay inlinable.
Apply the same refactor to decode32 for consistency, and to let 32-bit
architectures see a similar performance gain for large inputs.
Mikio Hara [Sun, 10 Mar 2019 05:30:06 +0000 (14:30 +0900)]
runtime, internal/poll, net: report event scanning error on read event
This change makes it possible the runtime-integrated network poller and
APIs in the package internal/poll to report an event scanning error on a
read event.
The latest Go releases open up the way of the manipulation of the poller
for users. On the other hand, it starts misleading users into believing
that the poller accepts any user-configured file or socket perfectly
because of not reporting any error on event scanning, as mentioned in
issue 30426. The initial implementation of the poller was designed for
just well-configured, validated sockets produced by the package net.
However, the assumption is now obsolete.
Clément Chigot [Wed, 20 Feb 2019 15:42:11 +0000 (16:42 +0100)]
cmd: always allow bigtoc generation with gcc on aix/ppc64
-mcmodel=large and -Wl,-bbigtoc must always be passed to gcc in order to
prevent TOC overflow error. However, a warning is still issued by ld. It
is removed as it doesn't give any useful information.
Change-Id: I95a78e8993cc7b5c0f329654d507409785f7eea6
Reviewed-on: https://go-review.googlesource.com/c/go/+/164008
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Clément Chigot [Wed, 20 Feb 2019 15:39:09 +0000 (16:39 +0100)]
cmd: disable DWARF with old ld on aix/ppc64
DWARF relocations isn't working with some older ld, because of
-Wl,-bnoobjreorder which is needed on Go.
This commit checks ld's version and disable DWARF generation in cmd/link
if it's too old. Some tests must therefore be skipped.
Change-Id: I2e794c263eb0dfe0b42e7062fb80c26f086b44d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/164007
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Fri, 8 Mar 2019 01:44:50 +0000 (17:44 -0800)]
math/big: add support for underscores '_' in numbers
The primary change is in nat.scan which now accepts underscores for base 0.
While at it, streamlined error handling in that function as well.
Also, improved the corresponding test significantly by checking the
expected result values also in case of scan errors.
The second major change is in scanExponent which now accepts underscores when
the new sepOk argument is set. While at it, essentially rewrote that
function to match error and underscore handling of nat.scan more closely.
Added a new test for scanExponent which until now was only tested
indirectly.
Finally, updated the documentation for several functions and added many
new test cases to clients of nat.scan.
A major portion of this CL is due to much better test coverage.
Daniel Martí [Mon, 11 Mar 2019 17:04:48 +0000 (17:04 +0000)]
text/tabwriter: use a single defer per Write call
Lines with single cells prompt a flush. Unfortunately, a call to
Writer.Flush also means two defers, which is an expensive operation to
do if many lines consist of single cells.
This is common when formatting code with aligned comments. Most lines
aren't going to have any comments at all, so the performance hit is
going to be noticeable.
The Write method already has a "defer handlePanic" of its own, so we
don't need to worry about panics leaking out. The error will now mention
"Write" instead of "Flush" if a panic is encountered during that nested
flush, but arguably that's a good thing; the user called Write, not
Flush.
For the reset call, add a non-deferred call as part of flushNoDefers, as
that's still necessary. Otherwise, the exported Flush method still does
a "defer b.reset".
The current tabwriter benchmarks are unaffected, since they don't
contain many single-cell lines, and because lines are written one at a
time. For that reason, we add a benchmark which has both of these
characteristics.
name old time/op new time/op delta
Code-8 2.72µs ± 0% 1.77µs ± 0% -34.88% (p=0.000 n=6+5)
name old alloc/op new alloc/op delta
Code-8 648B ± 0% 648B ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
Code-8 13.0 ± 0% 13.0 ± 0% ~ (all equal)
Perhaps unsurprisingly, go/printer also gets a bit faster, as it too
buffers its output before writing it to tabwriter.
name old time/op new time/op delta
Print-8 6.53ms ± 0% 6.39ms ± 0% -2.22% (p=0.008 n=5+5)
Change-Id: Ie01fea5ced43886a9eb796cb1e6c810f7a810853
Reviewed-on: https://go-review.googlesource.com/c/go/+/166797
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Bryan C. Mills [Tue, 12 Mar 2019 20:43:45 +0000 (16:43 -0400)]
cmd/go: fix typo in GoGetInsecure to actually set GOPROXY
I typo'd this variable in CL 165745, and neither I, the reviewer, nor the TryBots noticed.
But the longtest builder noticed, and it's not happy about it.
cmd/compile: make rulegen magic variable prediction more precise
The sheer length of the generated rules files makes my
editor and git client unhappy.
This change is a small step towards shortening them.
We recognize a few magic variables during rulegen: b, config, fe, typ.
Of these, only b appears prone to false positives.
By tightening the heuristic and fixing one case in MIPS.rules,
we can make the heuristic enough that it has no failures.
That allows us to remove the hedge assignments to _,
removing 3000 pointless lines of code.
Shubham Sharma [Mon, 11 Mar 2019 06:09:25 +0000 (11:39 +0530)]
cmd/go: document GOGCCFLAGS
Fixes #30123
Change-Id: I310b89a344a262bb758d39dfdd485ed2940cc6aa
Reviewed-on: https://go-review.googlesource.com/c/go/+/166577
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Bryan C. Mills [Tue, 12 Mar 2019 18:41:09 +0000 (14:41 -0400)]
cmd/internal/obj/x86: avoid os.Chdir in issue19518_test.go
Chdir leaves the test in the wrong working directory if objdumpOutput
calls t.Fatalf (or panics), and it isn't necessary here anyway. Set
the Dir field on the commands instead.
Change-Id: I9f0eb0d4f8d15043f1e13472126ca1a1ce4b7cb3
Reviewed-on: https://go-review.googlesource.com/c/go/+/167081
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Jay Conrod <jayconrod@google.com>
Robert Griesemer [Mon, 4 Feb 2019 23:33:19 +0000 (15:33 -0800)]
spec: document new Go2 number literals
This CL documents the new binary and octal integer literals,
hexadecimal floats, generalized imaginary literals and digit
separators for all number literals in the spec.
Added empty lines between abutting paragraphs in some places
(a more thorough cleanup can be done in a separate CL).
A minor detail: A single 0 was considered an octal zero per the
syntax (decimal integer literals always started with a non-zero
digit). The new octal literal syntax allows 0o and 0O prefixes
and when keeping the respective octal_lit syntax symmetric with
all the others (binary_lit, hex_lit), a single 0 is not automatically
part of it anymore. Rather than complicating the new octal_lit syntax
to include 0 as before, it is simpler (and more natural) to accept
a single 0 as part of a decimal_lit. This is purely a notational
change.
Change-Id: Ib9fdc6e781f6031cceeed37aaed9d05c7141adec
Reviewed-on: https://go-review.googlesource.com/c/go/+/161098 Reviewed-by: Rob Pike <r@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Ben Hoyt [Thu, 6 Dec 2018 13:53:29 +0000 (08:53 -0500)]
bytes, strings: speed up TrimSpace 4-5x for common ASCII cases
This change adds a fast path for ASCII strings to both
strings.TrimSpace and bytes.TrimSpace. It doesn't slow down the
non-ASCII path much, if at all.
I added benchmarks for strings.TrimSpace as it didn't have any, and
I fleshed out the benchmarks for bytes.TrimSpace as it just had one
case (for ASCII). The benchmarks (and the code!) are now the same
between the two versions. Below are the benchmark results:
Elias Naur [Tue, 12 Mar 2019 15:21:43 +0000 (16:21 +0100)]
misc/android: fix detection of GOROOT tests
strings.HasPrefix is not good enough to determine whether a path
is a subdirectory of another because it does not respect path
boundaries. filepath.Rel is good eonugh as long as we filter out results
that use parent directories, "..".
Hopefully fix the android emulator builders on the subrepositories.
fanzha02 [Mon, 11 Mar 2019 03:51:06 +0000 (03:51 +0000)]
cmd/compile: add handling for new floating-point comparisons flags
The CL 164718 adds new condition flags for floating-point comparisons
in arm64 backend, but dose not add the handling in rewrite.go for
corresponding Ops, which causes issue 30679. And this CL fixes this
issue.
Michael Stapelberg [Thu, 17 Jan 2019 15:53:41 +0000 (16:53 +0100)]
syscall: correctly set up uid/gid mappings in user namespaces
Before this CL, uid/gid mapping was always set up from the parent
process, which is a privileged operation.
When using unprivileged user namespaces, a process can modify its
uid/gid mapping after the unshare(2) call (but setting the uid/gid
mapping from another process is NOT possible).
Fixes #29789
Change-Id: I8c96a03f5da23fe80bbb83ef051ad89cf185d750
Reviewed-on: https://go-review.googlesource.com/c/go/+/158298
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Than McIntosh [Mon, 11 Mar 2019 19:05:10 +0000 (15:05 -0400)]
test: add new test for gccgo compilation problem
New test for issue 30659 (compilation error due to bad
export data).
Updates #30659.
Change-Id: I2541ee3c379e5b22033fea66bb4ebaf720cc5e1f
Reviewed-on: https://go-review.googlesource.com/c/go/+/166917
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
The deadcode pass runs a lot.
I'd like it to run even more.
This change adds dedicated storage for deadcode to ssa.Cache.
In addition to being a nice win now, it makes
deadcode easier to add other places in the future.
Elias Naur [Sat, 9 Mar 2019 01:30:35 +0000 (02:30 +0100)]
misc/android: copy less from GOROOT to the device
The android emulator builders is running out of space after CL 165797
copied most of GOROOT to the device.
The pkg directory is by far the largest, so only include what seems
necessary to build the x/ repositories: pkg/android_$GOARCH and
pkg/tool/android_$GOARCH.
While here, rename the device root directory to match the exec
wrapper name and make sure the deferred cleanups actually run before
os.Exit.
Bryan C. Mills [Fri, 1 Mar 2019 15:12:30 +0000 (10:12 -0500)]
all: move internal/x to vendor/golang.org/x and revendor using 'go mod vendor'
This also updates the vendored-in versions of several packages: 'go
mod vendor' selects a consistent version of each module, but we had
previously vendored an ad-hoc selection of packages.
Notably, x/crypto/hkdf was previously vendored in at a much newer
commit than the rest of x/crypto. Bringing the rest of x/crypto up to
that commit introduced an import of golang.org/x/sys/cpu, which broke
the js/wasm build, requiring an upgrade of x/sys to pick up CL 165749.
Bryan C. Mills [Thu, 28 Feb 2019 21:21:48 +0000 (16:21 -0500)]
cmd: refresh cmd/vendor to match 'go mod vendor'
This change preserves the maximum versions from cmd/vendor/vendor.json
where feasible, but bumps the versions of x/sys (for CL 162987) and
x/tools (for CL 162989 and CL 160837) so that 'go test all' passes in
module mode when run from a working directory in src/cmd.
A small change to cmd/vet (not vendored) was necessary to preserve its
flag behavior given a pristine copy of x/tools; see CL 162989 for more
detail.
This change was generated by running 'go mod vendor' at CL 164622.
(Welcoooome to the fuuuuuture!)
Updates #30228
Updates #30241
Change-Id: I889590318dc857d4a6e20c3023d09a27128d8255
Reviewed-on: https://go-review.googlesource.com/c/go/+/164618
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com>
Both are fairly cryptic and hard to review, but the former
is at least compact, which helps to not obscure the structure
of the rest of the rule.
Support that by adjusting rulegen's expansion.
Instead of looking for an op that begins with "(", ends with " ",
and has exactly one set of parens in it, look for everything of the
form "(...|...)".
That has false positives: Go code in the && conditions and AuxInt expressions.
Those are easily checked for syntactically: && conditions are between && and ->,
and AuxInt expressions are inside square brackets.
After ruling out those false positives, we can keep everything else,
regardless of where it is.
No change to the generated code for existing rules.
cmd/compile: normalize more whitespace in rewrite rules
If you write a rewrite rule:
(something) && noteRule("X")-> (something)
then rulegen will panic with an error message about commutativity.
The real problem is the lack of a space between the ) and the ->.
Normalize that bit of whitespace too.
Alberto Donizetti [Sat, 9 Mar 2019 17:01:26 +0000 (18:01 +0100)]
syscall: skip non-root user namespace test if kernel forbids
The unprivileged_userns_clone sysctl prevents unpriviledged users from
creating namespaces, which the AmbientCaps test does. It's set to 0 by
default in a few Linux distributions (Debian and Arch, possibly
others), so we need to check it before running the test.
Jason A. Donenfeld [Wed, 6 Mar 2019 18:26:29 +0000 (19:26 +0100)]
runtime: safely load DLLs
While many other call sites have been moved to using the proper
higher-level system loading, these areas were left out. This prevents
DLL directory injection attacks. This includes both the runtime load
calls (using LoadLibrary prior) and the implicitly linked ones via
cgo_import_dynamic, which we move to our LoadLibraryEx. The goal is to
only loosely load kernel32.dll and strictly load all others.
Meanwhile we make sure that we never fallback to insecure loading on
older or unpatched systems.
math/big: add fast path for pure Go addVW for large z
In the normal case, only a few words have to be updated when adding a word to a vector.
When that happens, we can simply copy the rest of the words, which is much faster.
However, the overhead of that makes it prohibitive for small vectors,
so we check the size at the beginning.
The implementation is a bit weird to allow addVW to continued to be inlined; see #30548.
The AddVW benchmarks are surprising, but fully repeatable.
The SubVW benchmarks are more or less as expected.
I expect that removing the indirect function call will
help both and make them a bit more normal.
math/big: remove bounds checks in pure Go implementations
These routines are quite sensitive to BCE.
This change eliminates bounds checks from loops.
It does so at the cost of a bit of safety:
malformed input will now return incorrect answers
instead of panicking.
This isn't as bad as it sounds: math/big has very good
test coverage, and the alternative implementations are in
assembly, which could do much worse things with malformed input.
If the compiler's BCE improves, so could these routines.
Notable BCE improvements for these routines would be:
* Allowing and propagating more cross-slice length hints.
Then hints like _ = y[:len(z)] would eliminate bounds checks for y[i].
* Propagating enough information so that we could do
n := len(x)
if len(z) < n {
n = len(z)
}
and then have i < n eliminate the same bounds checks as
i < len(x) && i < len(z) currently does.
* Providing some way to do BCE for unrolled loops.
Now that we have math/bits implementations,
it is possible to write things like ADC chains in
pure Go, if you can reasonably unroll loops.
Benchmarks below are for amd64, using -tags=math_big_pure_go.
Daniel Martí [Sat, 9 Mar 2019 18:09:10 +0000 (18:09 +0000)]
reflect: make all flag.mustBe* methods inlinable
mustBe was barely over budget, so manually inlining the first flag.kind
call is enough. Add a TODO to reverse that in the future, once the
compiler gets better.
mustBeExported and mustBeAssignable were over budget by a larger amount,
so add slow path functions instead. This is the same strategy used in
the sync package for common methods like Once.Do, for example.
Lots of exported reflect.Value methods call these assert-like unexported
methods, so avoiding the function call overhead in the common case does
shave off a percent from most exported APIs.
Finally, add the methods to TestIntendedInlining.
While at it, replace a couple of uses of the 0 Kind with its descriptive
name, Invalid.
name old time/op new time/op delta
Call-8 68.0ns ± 1% 66.8ns ± 1% -1.81% (p=0.000 n=10+9)
PtrTo-8 8.00ns ± 2% 7.83ns ± 0% -2.19% (p=0.000 n=10+9)