Michael Anthony Knyszek [Mon, 2 Nov 2020 16:58:38 +0000 (16:58 +0000)]
runtime: make getMCache inlineable
This change moves the responsibility of throwing if an mcache is not
available to the caller, because the inlining cost of throw is set very
high in the compiler. Even if it was reduced down to the cost of a usual
function call, it would still be too expensive, so just move it out.
This choice also makes sense in the context of #42339 since we're going
to have to handle the case where we don't have an mcache to update stats
in a few contexts anyhow.
Also, add getMCache to the list of functions that should be inlined to
prevent future regressions.
getMCache is called on the allocation fast path and because its not
inlined actually causes a significant regression (~10%) in some
microbenchmarks.
Fixes #42305.
Change-Id: I64ac5e4f26b730bd4435ea1069a4a50f55411ced
Reviewed-on: https://go-review.googlesource.com/c/go/+/267157
Trust: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com> Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Michael Matloob [Mon, 2 Nov 2020 19:19:12 +0000 (19:19 +0000)]
cmd/go: support cgo files in overlays
This is a roll-forward of golang.org/cl/262618, which was reverted in
golang.org/cl/267037. The only differences between this CL and the
original are the three calls to fflush from the C files in
build_overlay.txt, to guarantee that the string we're expecting is
actually written out.
This requires rewriting the paths of the files passed to the cgo tool
toolchain to use the overlaid paths instead of the disk paths of
files. Because the directories of the overlaid paths don't exist in
general, the cgo tool have been updated to run in base.Cwd instead of
the package directory.
For #39958
Change-Id: If7e5e057c62c0c22ddb724f9fe650902fc5f4832
Reviewed-on: https://go-review.googlesource.com/c/go/+/267197 Reviewed-by: Bryan C. Mills <bcmills@google.com>
Trust: Michael Matloob <matloob@golang.org>
Austin Clements [Sun, 1 Nov 2020 22:48:42 +0000 (17:48 -0500)]
runtime: default to MADV_DONTNEED on Linux
In Go 1.12, we changed the runtime to use MADV_FREE when available on
Linux (falling back to MADV_DONTNEED) in CL 135395 to address issue
#23687. While MADV_FREE is somewhat faster than MADV_DONTNEED, it
doesn't affect many of the statistics that MADV_DONTNEED does until
the memory is actually reclaimed under OS memory pressure. This
generally leads to poor user experience, like confusing stats in top
and other monitoring tools; and bad integration with management
systems that respond to memory usage.
We've seen numerous issues about this user experience, including
#41818, #39295, #37585, #33376, and #30904, many questions on Go
mailing lists, and requests for mechanisms to change this behavior at
run-time, such as #40870. There are also issues that may be a result
of this, but root-causing it can be difficult, such as #41444 and
#39174. And there's some evidence it may even be incompatible with
Android's process management in #37569.
This CL changes the default to prefer MADV_DONTNEED over MADV_FREE, to
favor user-friendliness and minimal surprise over performance. I think
it's become clear that Linux's implementation of MADV_FREE ultimately
doesn't meet our needs. We've also made many improvements to the
scavenger since Go 1.12. In particular, it is now far more prompt and
it is self-paced, so it will simply trickle memory back to the system
a little more slowly with this change. This can still be overridden by
setting GODEBUG=madvdontneed=0.
Fixes #42330 (meta-issue).
Fixes #41818, #39295, #37585, #33376, #30904 (many of which were
already closed as "working as intended").
Bryan C. Mills [Mon, 2 Nov 2020 14:12:28 +0000 (09:12 -0500)]
all: update dependency on golang.org/x/sys and regenerate Windows syscalls
Steps run:
$ cd $(go env GOROOT)/src
$ go get -d golang.org/x/sys
$ go mod tidy
$ go mod vendor
$ go generate syscall/... internal/syscall/...
$ cd cmd
$ go get -d golang.org/x/sys
$ go mod tidy
$ go mod vendor
$ cd ..
$ git add .
This change subsumes CL 260860.
For #36905
Change-Id: I7c677c6aa1ad61b9cbd8cf9ed208ed5a30f29c87
Reviewed-on: https://go-review.googlesource.com/c/go/+/267103
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Bryan C. Mills [Mon, 12 Oct 2020 14:51:34 +0000 (10:51 -0400)]
syscall: switch go:generate directives back to mksyscall_windows.go
Adjust mksyscall_windows.go to activate module mode and set
-mod=readonly, and to suppress its own deprecation warning when run
from within GOROOT/src.
We can't vendor the mkwinsyscall tool in to the std module directly,
because std-vendored dependencies (unlike the dependencies of all
other modules) turn into actual, distinct packages in 'std' when
viewed from outside the 'std' module. We don't want to introduce a
binary in the 'std' meta-pattern, but we also don't particularly want
to add more special-cases to the 'go' command right now when we have
an existing wrapper program that can do the job.
I also regenerated the affected packages to ensure that they are
consistent with the current version of mksyscall, which produced some
declaration-order changes in
internal/syscall/windows/zsyscall_windows.go.
Fixes #41916
Updates #25922
Change-Id: If6e6f8ba3dd372a7ecd6820ee6c0ca38d55f0f35
Reviewed-on: https://go-review.googlesource.com/c/go/+/261499
Trust: Bryan C. Mills <bcmills@google.com>
Trust: Alex Brainman <alex.brainman@gmail.com> Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org> Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Jonathan Swinney [Fri, 30 Oct 2020 18:46:23 +0000 (18:46 +0000)]
runtime: improve memmove performance on arm64
Replace the memmove implementation for moves of 17 bytes or larger
with an implementation from ARM optimized software. The moves of 16
bytes or fewer are unchanged, but the registers used are updated to
match the rest of the implementation.
This implementation makes use of new optimizations:
- software pipelined loop for large (>128 byte) moves
- medium size moves (17..128 bytes) have a new implementation
- address realignment when src or dst is unaligned
- preference for aligned src (loads) or dst (stores) depending on CPU
To support preference for aligned loads or aligned stores, a new CPU
flag is added. This flag indicates that the detected micro
architecture performs better with aligned loads. Some tested CPUs did
not exhibit a significant difference and are left with the default
behavior of realigning based on the destination address (stores).
Cherry Zhang [Sun, 1 Nov 2020 22:05:32 +0000 (17:05 -0500)]
cmd/compile: remove racefuncenterfp when it is not needed
We already remove racefuncenter and racefuncexit if they are not
needed (i.e. the function doesn't have any other race calls).
racefuncenterfp is like racefuncenter but used on LR machines.
Remove unnecessary racefuncenterfp as well.
Change-Id: I65edb00e19c6d9ab55a204cbbb93e9fb710559f1
Reviewed-on: https://go-review.googlesource.com/c/go/+/267099
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
Change-Id: Iec1e236ba793f24394442d04eb846f8a73ab2e68
Reviewed-on: https://go-review.googlesource.com/c/go/+/267037
Trust: Dmitri Shuralyov <dmitshur@golang.org>
Trust: Alberto Donizetti <alb.donizetti@gmail.com>
Run-TryBot: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Alberto Donizetti <alb.donizetti@gmail.com>
Emmanuel T Odeke [Fri, 29 May 2020 09:17:38 +0000 (02:17 -0700)]
cmd/vet: bring in pass to catch invalid uses of testing.T in goroutines
Add "go/analysis/passes/testinggoroutine" from x/tools and vendor its source in.
This pass will catch misuses of:
* testing.T.Fail*
* testing.T.Fatal*
* testing.T.Skip*
inside goroutines explicitly started by the go keyword.
The pass was implemented in CL 212920.
While here, found 2 misuses in:
* database/sql/sql_test.go
* runtime/syscall_windows_test.go
and fixed them in CL 235527.
Fixes #5746
Change-Id: I1740ad3f1d677bb5d78dc5d8d66bac6ec287a2b1
Reviewed-on: https://go-review.googlesource.com/c/go/+/235677
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Matthew Dempsky [Fri, 30 Oct 2020 19:58:28 +0000 (12:58 -0700)]
cmd/compile: fix recognition of unnamed return variables
In golang.org/cl/266199, I reused the existing code in inlining that
recognizes anonymous variables. However, it turns out that code
mistakenly recognizes anonymous return parameters as named when
inlining a function from the same package.
The issue is funcargs (which is only used for functions parsed from
source) synthesizes ~r names for anonymous return parameters, but
funcargs2 (which is only used for functions imported from export data)
does not.
This CL fixes the behavior so that anonymous return parameters are
handled identically whether a function is inlined within the same
package or across packages. It also adds a proper cross-package test
case demonstrating #33160 is fixed in both cases.
Change-Id: Iaa39a23f5666979a1f5ca6d09fc8c398e55b784c
Reviewed-on: https://go-review.googlesource.com/c/go/+/266719
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com> Reviewed-by: David Chase <drchase@google.com>
Trust: Matthew Dempsky <mdempsky@google.com>
Colin Arnott [Thu, 16 Jul 2020 17:42:47 +0000 (17:42 +0000)]
os: export errFinished as ErrProcessDone
(*Process).Signal returns an error sentinel, previously errFinished,
when (*Process).done or syscall.ESRCH. Callers would like the ability to
test for this state, so the value has been exported as ErrProcessDone.
Fixes #39444
Change-Id: I510e7647cc032af290180de5149f35ab7b09a526
Reviewed-on: https://go-review.googlesource.com/c/go/+/242998
Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Joel Sing [Mon, 26 Oct 2020 07:35:40 +0000 (18:35 +1100)]
cmd/compile: avoid unnecessary sign/zero extension for consts on riscv64
Sign extension for consts is unnecessary and zero extension for consts can be avoided
via casts. This removes over 16,000 instructions from the Go binary, in part because it
allows for better zero const absorbtion in blocks - for example,
`(BEQ (MOVBU (MOVBconst [0])) cond yes no)` now becomes `(BEQZ cond yes no)` when
this change is combined with existing rules.
Change-Id: I27e791bfa84869639db653af6119f6e10369ba3d
Reviewed-on: https://go-review.googlesource.com/c/go/+/265041
Trust: Joel Sing <joel@sing.id.au> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Michael Matloob [Thu, 15 Oct 2020 15:50:31 +0000 (11:50 -0400)]
cmd/go: support cgo files in overlays
This requires rewriting the paths of the files passed to the cgo tool
toolchain to use the overlaid paths instead of the disk paths of
files. Because the directories of the overlaid paths don't exist in
general, the cgo tool have been updated to run in base.Cwd instead of
the package directory.
For #39958
Change-Id: I8986de889f56ecc2e64fa69f5f6f29fa907408f9
Reviewed-on: https://go-review.googlesource.com/c/go/+/262618
Trust: Michael Matloob <matloob@golang.org>
Run-TryBot: Michael Matloob <matloob@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com> Reviewed-by: Bryan C. Mills <bcmills@google.com>
Michael Matloob [Tue, 27 Oct 2020 21:51:58 +0000 (17:51 -0400)]
cmd/cgo: add -trimpath flag allowing paths to be rewritten in outputs
cmd/cgo now has a -trimpath flag that behaves the same as the
-trimpath flag to cmd/compile. This will be used to correct paths
to cgo files that are overlaid.
The code that processes trimpath in internal/objapi has been slightly
refactored because it's currently only accessible via AbsFile, which
does some additional processing to the path names. Now an
ApplyRewrites function is exported that just applies the trimpath
rewrites.
Also remove unused srcfile argument to cmd/cgo.(*Package).godefs.
For #39958
Change-Id: I497d48d0bc2fe1f6ab2b5835cbe79f15b839ee59
Reviewed-on: https://go-review.googlesource.com/c/go/+/266358
Trust: Michael Matloob <matloob@golang.org>
Run-TryBot: Michael Matloob <matloob@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com> Reviewed-by: Bryan C. Mills <bcmills@google.com>
Dan Scales [Thu, 29 Oct 2020 22:31:16 +0000 (15:31 -0700)]
cmd/compile: fmt improvements for AST nodes and some comments on AST nodes
Changed fmt.go to print out some extra information for various kinds of
Nodes. This includes some extra (small) info in the %j (jconv) output,
and some missing sections (such as Dcls and the body of a closure) in
nodedump().
Also, added some extra doc comments for a few Node types in syntax.go
Change-Id: I2ec7184e2abe0d5fbe3fb5a2506da7c7b06f2fb1
Reviewed-on: https://go-review.googlesource.com/c/go/+/266437
Run-TryBot: Dan Scales <danscales@google.com>
Trust: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Michael Anthony Knyszek [Mon, 20 Jul 2020 18:19:56 +0000 (18:19 +0000)]
runtime: release worldsema with a direct G handoff
Currently worldsema is not released with direct handoff, so the
semaphore is an unfair synchronization mechanism. If, for example,
ReadMemStats is called in a loop, it can continuously stomp on attempts
by the GC to stop the world.
Note that it's specifically possible for ReadMemStats to delay a STW to
end GC since ReadMemStats is able to STW during a GC since #19112 was
fixed.
While this particular case is unlikely and the right answer in most
applications is to simply not call such an expensive operation in a
loop, this pattern is used often in tests.
Fixes #40459.
Change-Id: Ia4a54f0fd956ea145a319f9f06c4cd37dd52fd8a
Reviewed-on: https://go-review.googlesource.com/c/go/+/243977
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Michael Knyszek <mknyszek@google.com> Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Austin Clements <austin@google.com> Reviewed-by: David Chase <drchase@google.com>
Michael Pratt [Fri, 30 Oct 2020 19:22:52 +0000 (15:22 -0400)]
runtime: tighten systemstack in lock assertions
We use systemstack on the locking path to avoid stack splits which could
cause locks to be recorded out of order (see comment on lockWithRank).
This concern is irrelevant on lock assertions, where we simply need to
see if a lock is held and don't care if another is taken in the
meantime. Thus we can simply drop these unless we actually need to
crash.
Updates #40677
Change-Id: I85d730913a59867753ee1ed0386f8c5efda5c432
Reviewed-on: https://go-review.googlesource.com/c/go/+/266718
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Trust: Michael Pratt <mpratt@google.com>
Michael Pratt [Thu, 29 Oct 2020 20:03:57 +0000 (16:03 -0400)]
runtime: elide timer re-check if P has no timers
In golang.org/cl/264477, I missed this new block after rebasing past
golang.org/cl/232298. These fields must be zero if there are no timers.
Updates #28808
Updates #18237
Change-Id: I2d9e1cbf326497c833daa26b11aed9a1e12c2270
Reviewed-on: https://go-review.googlesource.com/c/go/+/266367
Run-TryBot: Michael Pratt <mpratt@google.com> Reviewed-by: Austin Clements <austin@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Michael Pratt <mpratt@google.com>
Michael Pratt [Fri, 21 Aug 2020 15:59:55 +0000 (11:59 -0400)]
runtime: add heap lock assertions
Some functions that required holding the heap lock _or_ world stop have
been simplified to simply requiring the heap lock. This is conceptually
simpler and taking the heap lock during world stop is guaranteed to not
contend. This was only done on functions already called on the
systemstack to avoid too many extra systemstack calls in GC.
Updates #40677
Change-Id: I15aa1dadcdd1a81aac3d2a9ecad6e7d0377befdc
Reviewed-on: https://go-review.googlesource.com/c/go/+/250262
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Trust: Michael Pratt <mpratt@google.com>
Michele Di Pede [Fri, 30 Oct 2020 10:55:18 +0000 (11:55 +0100)]
cmd/compile: code cleanup
Change-Id: Ibf68e663f29a5cb3b64a7d923c005c16da647769
Reviewed-on: https://go-review.googlesource.com/c/go/+/266537 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Matthew Dempsky [Fri, 30 Oct 2020 17:36:31 +0000 (10:36 -0700)]
cmd/compile: fix reassignVisitor
reassignVisitor was short-circuiting on assignment statements after
checking the LHS, but there might be further assignment statements
nested within the RHS expressions.
Fixes #42284.
Change-Id: I175eef87513b973ed5ebe6a6527adb9766dde6cf
Reviewed-on: https://go-review.googlesource.com/c/go/+/266618
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com> Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Cherry Zhang [Thu, 29 Oct 2020 15:27:17 +0000 (11:27 -0400)]
runtime, cmd: support race detector on darwin/arm64
https://reviews.llvm.org/D90435 is the counterpart in LLVM TSAN.
race_linux_arm64.syso is built with LLVM commit 00da38ce2d36c07f12c287dc515d37bb7bc410e9 on a macOS/ARM64 machine.
(It is not built on a builder with golang.org/x/build/cmd/racebuild
as we don't have darwin/arm64 builder for now.)
Updates #38485.
Change-Id: I391efdacd9480197e308370bfccd05777deb4aee
Reviewed-on: https://go-review.googlesource.com/c/go/+/266373
Trust: Cherry Zhang <cherryyz@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Bryan C. Mills [Thu, 29 Oct 2020 20:36:50 +0000 (16:36 -0400)]
cmd/go/internal/modload: handle NotExist errors in (*mvsReqs).Previous
Previous is used during downgrading. If the module proxy does not
advertise any versions (for example, because it contains only
pseudo-versions), then Previous should return "none" instead of a
non-nil error.
For #37438
Change-Id: I4edfec19cfeb3ffe50df4979f99a01321c442509
Reviewed-on: https://go-review.googlesource.com/c/go/+/266370
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Jay Conrod <jayconrod@google.com> Reviewed-by: Michael Matloob <matloob@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Bryan C. Mills [Thu, 29 Oct 2020 20:06:19 +0000 (16:06 -0400)]
cmd/go: make TestScript/mod_get_patchmod self-contained
I find it pretty difficult to reason about test-dependency modules
when they aren't in the same file as the rest of the test.
Now that 'go get' supports replacements (CL 258220 and CL 266018),
we can localize tests that need 'go get' but don't specifically depend
on module proxy semantics.
For #36460
For #37438
Change-Id: Ib37a6c170f251435399dfc23e60d96681a81eadc
Reviewed-on: https://go-review.googlesource.com/c/go/+/266369
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Jay Conrod <jayconrod@google.com> Reviewed-by: Michael Matloob <matloob@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Than McIntosh [Fri, 16 Oct 2020 13:22:31 +0000 (09:22 -0400)]
cmd/link: emit include directories in DWARF line table prologue
This patch changes the way the linker emits the DWARF line table
prologue, specifically the file table. Previously files were left
unmodified, and the directory table was empty. For each compilation
unit we now scan the unit file table and build up a common set of
directories, emit them into the directory table, and then emit file
entries that refer to the dirs. This provides a modest binary size
savings.
[where the value following the section name above is the section size
in hex, so roughly a 30% decrease in this case.]
The actual savings will depend on the length of the pathnames
involved, so it's hard to really pin down how much savings we'll see
here. In addition, emitting the files this way reduces the
"compressibility" of the line table, so there could even be cases
where we don't win at all.
Updates #6853, #19784, #36495.
Change-Id: I298d8561da5ed3ebc9d38aa772874851baa2f4f4
Reviewed-on: https://go-review.googlesource.com/c/go/+/263017
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Trust: Than McIntosh <thanm@google.com>
Ian Lance Taylor [Thu, 29 Oct 2020 23:23:27 +0000 (16:23 -0700)]
runtime: simplify nobarrierWakeTime
Also use the simplified nobarrierWakeTime in findrunnable, as it no
longer needs the current time.
Change-Id: I77b125d6a184dde0aeb517fc068164c274f0a046
Reviewed-on: https://go-review.googlesource.com/c/go/+/266304
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Michael Pratt <mpratt@google.com>
Than McIntosh [Wed, 14 Oct 2020 12:36:11 +0000 (08:36 -0400)]
reflect,runtime: use internal ABI for selected ASM routines, attempt 2
[This is a roll-forward of CL 262319, with a fix for some Darwin test
failures].
Change the definitions of selected runtime assembly routines
from ABI0 (the default) to ABIInternal. The ABIInternal def is
intended to indicate that these functions don't follow the existing Go
runtime ABI. In addition, convert the assembly reference to
runtime.main (from runtime.mainPC) to ABIInternal. Finally, for
functions such as "runtime.duffzero" that are called directly from
generated code, make sure that the compiler looks up the correct
ABI version.
This is intended to support the register abi work, however these
changes should not have any issues even when GOEXPERIMENT=regabi is
not in effect.
Updates #27539, #40724.
Change-Id: Idf507f1c06176073563845239e1a54dad51a9ea9
Reviewed-on: https://go-review.googlesource.com/c/go/+/266638
Trust: Than McIntosh <thanm@google.com>
Run-TryBot: Than McIntosh <thanm@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Alberto Donizetti [Thu, 29 Oct 2020 11:03:18 +0000 (12:03 +0100)]
cmd/compile: replace int32(a.Off()) calls with a.Off32()
Change-Id: Ib833954e8400a63624359a81f5196ca9080199a3
Reviewed-on: https://go-review.googlesource.com/c/go/+/266081
Trust: Alberto Donizetti <alb.donizetti@gmail.com>
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Rob Findley [Fri, 30 Oct 2020 15:35:48 +0000 (11:35 -0400)]
go/types: reorganize error codes into categories
In CL 264179, some reorganization of error codes was deferred in order
to minimize diffs between patch-sets.
This CL reorganizes the error codes as discussed. It is a pure
reordering, with no material changes other than the changing of internal
const values.
For #42290
Change-Id: I0e9b421a92e96b19e53039652f8de898c5255290
Reviewed-on: https://go-review.googlesource.com/c/go/+/266637
Run-TryBot: Robert Findley <rfindley@google.com>
Trust: Robert Findley <rfindley@google.com>
Trust: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Robert Griesemer [Wed, 7 Oct 2020 18:33:11 +0000 (11:33 -0700)]
spec: split shift examples into groups for 32- and 64-bit ints
In the current (pre-CL) version of the spec, the 2nd last shift
example appears to be using the array declared in the last example.
On a 32-bit platform, that array would have length 0, which would
lead to a panic in the 2nd last example. Also, if this code were
inside a function, it wouldn't compile (array declared after use).
Use an explicitly declared array for that specific shift example.
Also, split out all cases that produce different results for 32-
vs 64-bit ints.
Fixes #41835.
Change-Id: Ie45114224509e4999197226f91f7f6f934449abb
Reviewed-on: https://go-review.googlesource.com/c/go/+/260398
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Russ Cox <rsc@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Michael Pratt [Tue, 13 Oct 2020 22:11:26 +0000 (18:11 -0400)]
runtime: simplify gcBgMarkWorker preemption
gcBgMarkWorker G's are primarily scheduled by findRunnableGCWorker, but
that no longer needs to be strictly enforced. Temporary preemption to a
runq is fine when the P is not in use.
We still releasem in gopark in the normal case for efficiency: if
gcDrain stops because gp.preempt is set, then gopark would always
preempt. That is fine, but inefficient, since it will reschedule simply
to park again. Thus, we keep releasem in unlockf to skip this extra
cycle.
Change-Id: I6d1a42e3ca41b76227142a6b5bfb376c9213e3c9
Reviewed-on: https://go-review.googlesource.com/c/go/+/262349
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
Michael Pratt [Tue, 13 Oct 2020 16:39:13 +0000 (12:39 -0400)]
runtime: manage gcBgMarkWorkers with a global pool
Background mark workers perform per-P marking work. Currently each
worker is assigned a P at creation time. The worker "attaches" to the P
via p.gcBgMarkWorker, making itself (usually) available to
findRunnableGCWorker for scheduling GC work.
While running gcMarkDone, the worker "detaches" from the P (by clearing
p.gcBgMarkWorker), since it may park for other reasons and should not be
scheduled by findRunnableGCWorker.
Unfortunately, this design is complex and difficult to reason about. We
simplify things by changing the design to eliminate the hard P
attachment. Rather than workers always performing work from the same P,
workers perform work for whichever P they find themselves on. On park,
the workers are placed in a pool of free workers, which each P's
findRunnableGCWorker can use to run a worker for its P.
Now if a worker parks in gcMarkDone, a P may simply use another worker
from the pool to complete its own work.
The P's GC worker mode is used to communicate the mode to run to the
selected worker. It is also used to emit the appropriate worker
EvGoStart tracepoint. This is a slight change, as this G may be
preempted (e.g., in gcMarkDone). When it is rescheduled, the trace
viewer will show it as a normal goroutine again. It is currently a bit
difficult to connect to the original worker tracepoint, as the viewer
does not display the goid for the original worker (though the data is in
the trace file).
Change-Id: Id7bd3a364dc18a4d2b1c99c4dc4810fae1293c1b
Reviewed-on: https://go-review.googlesource.com/c/go/+/262348
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
Rob Findley [Mon, 19 Oct 2020 15:51:17 +0000 (11:51 -0400)]
go/types: add internal error codes
Tools using go/types sometimes need to implement special handling for
certain errors produced by the type-checker. They can offer suggested
fixes, expand the error position to surrounding syntax, highlight
related syntax (for example in the case of a declaration cycle), link to
additional documentation, group errors by category, or correlate errors
with signals produced by other static analysis tools.
All these require a notion of error identity. Tools need to be able to
reliably determine the nature of an error without re-implementing type
checking logic or parsing error messages. This CL is a first-pass at
adding such an identifier to types.Error: a (for the moment unexported)
field containing one of many declared errorCode constants.
A wide variety of error code constants are defined, and assigned to type
checker errors according to their 'functional equivalence', meaning that
they should be ideally be stable under refactoring.
With few exceptions, each error code is documented with an example that
produces it. This is enforced by tests.
When error codes are exported they will represent quite a large API
surface. For this reason, as well as the likelihood that error codes
will change at the start, both the code field and the codes themselves
are initially unexported. gopls will read these fields using reflection
during this experimental phase. Others can obviously do the same,
provided they accept the lack of forward compatibility.
For #42290
Change-Id: I15e3c2bffd2046c20297b1857057d421f633098a
Reviewed-on: https://go-review.googlesource.com/c/go/+/264179
Run-TryBot: Robert Findley <rfindley@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Robert Griesemer <gri@golang.org>
Trust: Robert Findley <rfindley@google.com>
Cherry Zhang [Thu, 29 Oct 2020 19:50:53 +0000 (15:50 -0400)]
runtime: allocate at desired address when race detector is on
Currently, on all supported platforms, the race detector (LLVM
TSAN) expects the Go heap is at 0xc000000000 - 0xe000000000.
Move the raceenabled condition first, so we always allocate
there.
This means on Linux/ARM64 when race detector is on we will
allocate to 0xc000000000 - 0xe000000000, instead of 0x4000000000.
The old address is meant for 39-bit VMA. But the race detector
only supports 48-bit VMA anyway. So this is fine.
Change-Id: I51ac8eff68297b37c8c651a93145cc94f83a939d
Reviewed-on: https://go-review.googlesource.com/c/go/+/266372
Trust: Cherry Zhang <cherryyz@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Matthew Dempsky [Thu, 29 Oct 2020 20:30:54 +0000 (13:30 -0700)]
cmd/compile: fix devirtualization of promoted interface methods
A method selector expression can pick out a method or promoted method
(represented by ODOTMETH), but it can also pick out an interface
method from an embedded interface-typed field (represented by
ODOTINTER).
In the case that we're picking out an interface method, we're not able
to fully devirtualize the method call. However, we're still able to
improve escape analysis somewhat. E.g., the included test case
demonstrates that we can optimize "i.M()" to "i.(T).I.M()", which
means the T literal can be stack allocated instead of heap allocated.
Fixes #42279.
Change-Id: Ifa21d19011e2f008d84f9624b7055b4676b6d188
Reviewed-on: https://go-review.googlesource.com/c/go/+/266300
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
Robert Griesemer [Fri, 30 Oct 2020 00:09:16 +0000 (17:09 -0700)]
strings: complete Reader doc string
Follow-up on https://golang.org/cl/247523.
Change-Id: I9e91a6d77271e640d84851f2e2a4c6d2150a2b43
Reviewed-on: https://go-review.googlesource.com/c/go/+/266438
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Ben Hoyt [Fri, 14 Aug 2020 10:58:49 +0000 (22:58 +1200)]
strconv: fix incorrect bit size in ParseComplex; add tests
In ParseComplex, the "size" passed to parseFloatPrefix should be 64 for
complex128, not 128. It still works because of how parseFloatPrefix
is forgiving about the size if it's not 32, but worth fixing anyway.
Make ParseComplex and ParseFloat return a bit size error for anything
other than 128 or 64 (for ParseComplex), or 64 or 32 (for ParseFloat).
Add "InvalidBitSize" tests for these cases.
Add tests for ParseComplex with bitSize==64: this is done in a similar
way to how the ParseFloat 32-bit tests work, re-using the tests for the
larger bit size.
Add tests for FormatComplex -- there were none before.
Fixes #40706
Change-Id: I16ddd546e5237207cc3b8c2181dd708eca42b04f
Reviewed-on: https://go-review.googlesource.com/c/go/+/248219
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Minux Ma <minux@golang.org>
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Norman B. Lancaster [Thu, 1 Oct 2020 19:14:04 +0000 (19:14 +0000)]
strings: complete documentation of strings.Reader
There is no documentation on a number of methods of the strings.Reader
struct, so this change adds documentation referring to the relevant
io.* interfaces implemented. This is consistent with pre-existing
documentation in this struct.
Fixes #40381
Change-Id: I3dec65ecafca5b79d85d30a676d297e5ee9ab47e
GitHub-Last-Rev: f42429946a2b90b9fbfbd1ea5943f0c50e0439b5
GitHub-Pull-Request: golang/go#40654
Reviewed-on: https://go-review.googlesource.com/c/go/+/247523
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Ian Lance Taylor <iant@golang.org>
surechen [Fri, 15 May 2020 14:59:38 +0000 (22:59 +0800)]
math/big: reduce allocations for building decimal strings
Append operations in the decimal String function may cause several allocations.
Use make to pre allocate slices in String that have enough capacity to avoid additional allocations in append operations.
name old time/op new time/op delta
DecimalConversion-8 139µs ± 7% 109µs ± 2% -21.06% (p=0.000 n=10+10)
Change-Id: Id0284d204918a179a0421c51c35d86a3408e1bd9
Reviewed-on: https://go-review.googlesource.com/c/go/+/233980
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Giovanni Bajo <rasky@develer.com> Reviewed-by: Martin Möhrmann <moehrmann@google.com> Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Giovanni Bajo <rasky@develer.com>
Trust: Martin Möhrmann <moehrmann@google.com>
Rémy Oudompheng [Fri, 23 Oct 2020 20:23:21 +0000 (22:23 +0200)]
strconv: make Eisel-Lemire handle long mantissas
In many cases, it is not necessary to parse long
decimal mantissas entirely to produce the correctly
rounded floating-point number. It is enough to parse
the short, rounded lower and upper bounds and in most cases
they round to the same floating point number because uint64
can hold 19 digits.
Previously this case was handled by the extFloat code path
(Grisu3 algorithm).
Cherry Zhang [Thu, 29 Oct 2020 02:46:41 +0000 (22:46 -0400)]
cmd/link: put C static symbols in the symbol table, attempt 2
We don't put Go static symbols in the symbol table, as they are
always compiler-generated (there is no way to define a static
symbol in user code in Go). We retain static symbols in assembly
code, as it may be user-defined. Also retain static symbols in C.
This is the second attempt of CL 263259, which was reverted
because it broke AIX tests in that it brought TOC.stmp symbols
in the symbol table. This time we use SymPkg(s) == "" to identify
non-Go symbols, instead of IsExternal(s), as the latter also
includes linker-modified Go symbols.
Change-Id: I5c752c54f0fc6ac4cde6a0e8161dac5b72a47d56
Reviewed-on: https://go-review.googlesource.com/c/go/+/266237
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Russ Cox [Thu, 29 Oct 2020 17:57:12 +0000 (13:57 -0400)]
cmd/internal/obj: use panic instead of log.Fatalf for two messages
These messages can happen if there are
duplicate body-less function declarations.
Using panic gives the panic handler
a chance to handle the panic by printing the
queued error messages instead of an internal error.
And if there are no queued error messages,
using panic pinpoints the stack trace leading
to the incorrect use of NewFuncInfo/NewFileInfo.
Matthew Dempsky [Thu, 29 Oct 2020 19:47:15 +0000 (12:47 -0700)]
cmd/compile: gracefully fail when devirtualization fails
We should still be able to devirtualize here, but I need to understand
the AST better. While I'm doing that, at least switch to a graceful
failure case (i.e., skip the optimization and print a warning message)
to fix the x/text builders.
Updates #42279.
Change-Id: Ie2b0b701fccf590d0cabfead703fc2fa999072cf
Reviewed-on: https://go-review.googlesource.com/c/go/+/266359
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Matthew Dempsky [Thu, 29 Oct 2020 01:49:10 +0000 (18:49 -0700)]
cmd/compile: early devirtualization of interface method calls
After inlining, add a pass that looks for interface calls where we can
statically determine the interface value's concrete type. If such a
case is found, insert an explicit type assertion to the concrete type
so that escape analysis can see it.
Fixes #33160.
Change-Id: I36932c691693f0069e34384086d63133e249b06b
Reviewed-on: https://go-review.googlesource.com/c/go/+/264837
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
Matthew Dempsky [Sat, 24 Oct 2020 09:08:06 +0000 (02:08 -0700)]
cmd/compile: improve inlining and static analysis
When inlining a function call "f()", if "f" contains exactly 1
"return" statement and doesn't name its result parameters, it's
inlined to declare+initialize the result value using the AST
representation that's compatible with staticValue.
Also, extend staticValue to skip over OCONVNOP nodes (often introduced
by inlining), and fix various bits of code related to handling method
expressions.
Updates #33160.
Change-Id: If8652e319f0a5700cf9d40a7a62e369a2a359229
Reviewed-on: https://go-review.googlesource.com/c/go/+/266199
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
Joel Sing [Tue, 20 Oct 2020 11:04:55 +0000 (22:04 +1100)]
cmd/link: add loadelf support for riscv64
Update #36641
Change-Id: I8618da30d8940a56d6cc86a37a2f54b31ee029e5
Reviewed-on: https://go-review.googlesource.com/c/go/+/263601
Trust: Joel Sing <joel@sing.id.au> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Go Bot <gobot@golang.org>
Joel Sing [Thu, 29 Oct 2020 17:32:18 +0000 (04:32 +1100)]
Revert "cmd/compile,cmd/internal/sys: enable additional build modes on linux/riscv64"
This reverts CL 263457.
It turns out that this still missed changes to cmd/link/internal/ld/config.go
and some of these build modes also fail once cgo is enabled. Disable again for
now.
Change-Id: Iaf40d44e1551afd5b040d357f04af134f55a64a9
Reviewed-on: https://go-review.googlesource.com/c/go/+/266317 Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Joel Sing <joel@sing.id.au>
Andrew G. Morgan [Wed, 28 Oct 2020 20:35:57 +0000 (13:35 -0700)]
syscall: handle undefined r2 value on linux-ppc64x
This change fixes two failng tests on linux-ppc64x:
- TestAllThreadsSyscall() exposed a real bug in the ppc64x support:
- It turns out that the r2 syscall return value is not defined
on all architectures. Notably linux-ppc64x so address that by
introducing a private architectural constant in the syscall
package, archHonorsR2: true if r2 has a determanistic value.
- TestSetuidEtc() was sensitive to /proc/<PID>/status content:
- The amount of padding space has changed with kernel vintage.
- Stress testing revealed a race with /proc files disappearing.
Fixes #42178
Change-Id: Ie6fc0b8f2f94a409ac0e5756e73bfce113274709
Reviewed-on: https://go-review.googlesource.com/c/go/+/266202
Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Branden J Brown [Sun, 25 Oct 2020 18:26:19 +0000 (14:26 -0400)]
cmd/compile: inline functions evaluated in go and defer statements
The inlining pass previously bailed upon encountering a go or defer statement, so it would not inline functions e.g. used to provide arguments to the deferred function. This change preserves the behavior of not inlining the
deferred function itself, but it allows the inlining walk to proceed into its arguments.
Fixes #42194
Change-Id: I4e82029d8dcbe69019cc83ae63a4b29af45ec777
Reviewed-on: https://go-review.googlesource.com/c/go/+/264997
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
The final piece of //go:embed support: have the go command stitch
together parsing in go/build, low-level data initialization in cmd/compile,
and the new data structures in package embed, to make the //go:embed
feature actually function.
And test, now that all the pieces are available to work together.
For #41191.
(Issue not fixed: still need to add a tool for use by Bazel.)
Change-Id: Ib1d198345c3b4d557d340f292eda13b984b65d65
Reviewed-on: https://go-review.googlesource.com/c/go/+/243945
Trust: Russ Cox <rsc@golang.org>
Trust: Jay Conrod <jayconrod@google.com>
Trust: Johan Brandhorst <johan.brandhorst@gmail.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com> Reviewed-by: Johan Brandhorst <johan.brandhorst@gmail.com>
Katie Hockman [Mon, 26 Oct 2020 13:39:56 +0000 (13:39 +0000)]
Revert "crypto/x509: fix duplicate import"
This reverts CL 250497. It also moves all blank identifier imports below the rest of the imports for clarity.
Reason for revert: The blank identifier import was intentional to show that it's needed for its registration side effect. The duplicate import should stay since it communicates that the side-effect is important to tools and to future developers updating this file.
Than McIntosh [Wed, 14 Oct 2020 12:36:11 +0000 (08:36 -0400)]
reflect,runtime: use internal ABI for selected ASM routines
Change the definitions of selected runtime assembly routines
from ABI0 (the default) to ABIInternal. The ABIInternal def is
intended to indicate that these functions don't follow the existing Go
runtime ABI. In addition, convert the assembly reference to
runtime.main (from runtime.mainPC) to ABIInternal. Finally, for
functions such as "runtime.duffzero" that are called directly from
generated code, make sure that the compiler looks up the correct
ABI version.
This is intended to support the register abi work, however these
changes should not have any issues even when GOEXPERIMENT=regabi is
not in effect.
Updates #27539, #40724.
Change-Id: I9846f8dcaccc95718cf2e61a18b7e924a0677e4c
Reviewed-on: https://go-review.googlesource.com/c/go/+/262319
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Than McIntosh <thanm@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Joel Sing [Tue, 25 Aug 2020 08:04:47 +0000 (18:04 +1000)]
runtime: add support for openbsd/mips64
Update #40995
Change-Id: Ie028dfd87ef8731804567a0501f1f7758e8dd203
Reviewed-on: https://go-review.googlesource.com/c/go/+/250580
Trust: Joel Sing <joel@sing.id.au> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Joel Sing <joel@sing.id.au>
This patch enables VSLI, VUADDW(2), VUSRA and FMOVQ SIMD instructions
required by the issue #40725. And the GNU syntax of 'FMOVQ' is 128-bit
ldr/str(immediate, simd&fp).
David Chase [Tue, 13 Oct 2020 23:24:04 +0000 (19:24 -0400)]
cmd/compile: delay expansion of OpArg until expand_calls
As it says, delay expanpsion of OpArg to the expand_calls phase,
to enable (eventually) interprocedural SSA optimizations, and
(sooner) change to a register ABI.
Includes a round of cleanup to function names and comments,
largely to match the expanded scope of the functions.
This CL removes the per-function dependence on GOSSAHASH,
but the go116lateCallExpansion kill switch remains (and was
tested locally to ensure it worked).
Two functions in expand_calls.go that performed overlapping
things were combined into a single function that is called
twice.
Fixes #42236.
For #40724.
Change-Id: Icbb78947eaa39f17f2c1210d5c2caef20abd6571
Reviewed-on: https://go-review.googlesource.com/c/go/+/262117
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Nigel Tao [Fri, 23 Oct 2020 01:11:35 +0000 (12:11 +1100)]
strconv: remove extfloat.go atof code path
Prior to this commit, strconv.ParseFloat (known in C as atof) takes the
first of four algorithms to succeed: atof64exact, eiselLemire64,
extFloat, fallback. The Eisel-Lemire implementation is a recent addition
but, now that it exists, the extFloat implementation (based on the
algorithm used by https://github.com/google/double-conversion) is
largely redundant. This Go program:
func parseOneMillionFloats(bitSize int, normallyDistributed bool) {
rng := rand.New(rand.NewSource(1))
for i := 0; i < 1_000_000; {
x := 0.0
if normallyDistributed {
x = rng.NormFloat64()
} else if bitSize == 32 {
x = float64(math.Float32frombits(rng.Uint32()))
} else {
x = math.Float64frombits(
uint64(rng.Uint32())<<32 | uint64(rng.Uint32()))
}
if math.IsInf(x, 0) {
continue
}
s := strconv.FormatFloat(x, 'g', -1, bitSize)
strconv.ParseFloat(s, bitSize)
i++
}
}
triggers the four algorithms by these percentages:
This commit removes the extfloat.go atof code (but keeps the extfloat.go
ftoa code for now), reducing the number of atof algorithms from 4 to 3.
The benchmarks (below) show some regressions but these are arguably
largely artificial situations.
Atof*RandomBits generates uniformly distributed uint32/uint64 values and
reinterprets the bits as float32/float64 values. The change in headline
numbers (arithmetic means) are primarily due to relatively large changes
for relatively rare cases.
fanzha02 [Fri, 16 Oct 2020 10:35:30 +0000 (18:35 +0800)]
cmd/asm: refactor some encoding functions for load/store with immediate offset on arm64
Some of the current functions for encoding load/store with
immediate offset instructions, like opstr12(), opstr9(),
opldr12(), opldr9() and opldrpp(), etc., they have the same
code, so this patch refactors them and merges them into two
functions opstr() and opldr().
Rémy Oudompheng [Sun, 25 Oct 2020 10:52:29 +0000 (11:52 +0100)]
cmd/compile: use magic multiply for unsigned values less than 1<<16 on 32-bit architectures
This is done by decomposing the number to be divided in 32-bit
components and using the 32-bit magic multiply. For the lowering to be
effective the constant must fit in 16 bits.
On ARM the expression n / 5 compiles to 25 instructions.
Bryan C. Mills [Wed, 28 Oct 2020 14:44:24 +0000 (10:44 -0400)]
cmd/go: use internal/testenv instead of computing canRun and skipExternal ad-hoc
Fixes #42223
Change-Id: Icf9bb61d48f0a6c7fd6f74e80e333a4837aa52ab
Reviewed-on: https://go-review.googlesource.com/c/go/+/265781
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alberto Donizetti [Wed, 28 Oct 2020 09:10:55 +0000 (10:10 +0100)]
cmd/compile: rename mergeSymTyped to mergeSym
Also make canMergeSym take Syms instead of interface{}
Change-Id: I4926a1fc586aa90e198249d67e5b520404b40869
Reviewed-on: https://go-review.googlesource.com/c/go/+/265817
Trust: Alberto Donizetti <alb.donizetti@gmail.com>
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Alberto Donizetti [Tue, 27 Oct 2020 10:30:14 +0000 (11:30 +0100)]
cmd/compile: delete log2, switch to log64
rewrite.go has two identical functions log2 and log64; the former has
been there for a while, while the latter was added together with
log{8,16,32} for use in typed rules.
This change deletes log2 and switches to using log64 everywhere.
Change-Id: I759b878814e4c115a5fa470274f22477738d69ef
Reviewed-on: https://go-review.googlesource.com/c/go/+/265457
Trust: Alberto Donizetti <alb.donizetti@gmail.com>
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com> Reviewed-by: Keith Randall <khr@golang.org>
Joel Sing [Tue, 25 Aug 2020 07:53:08 +0000 (17:53 +1000)]
runtime: add defs for openbsd/mips64
Update #40995
Change-Id: I6963ead1a7c4520092361cce80edb17010e7f436
Reviewed-on: https://go-review.googlesource.com/c/go/+/250579
Trust: Joel Sing <joel@sing.id.au>
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
Michael Pratt [Mon, 5 Oct 2020 22:12:35 +0000 (18:12 -0400)]
runtime: try to elide timer stealing if P has no timers
Following golang.org/cl/259578, findrunnable still must touch every
other P in checkTimers in order to look for timers to steal. This scales
poorly with GOMAXPROCS and potentially performs poorly by pulling remote
Ps into cache.
Add timerpMask, a bitmask that tracks whether each P may have any timers
on its timer heap.
Ideally we would update this field on any timer add / remove to always
keep it up to date. Unfortunately, updating a shared global structure is
antithetical to sharding timers by P, and doing so approximately doubles
the cost of addtimer / deltimer in microbenchmarks.
Instead we only (potentially) clear the mask when the P goes idle. This
covers the best case of avoiding looking at a P _at all_ when it is idle
and has no timers. See the comment on updateTimerPMask for more details
on the trade-off. Future CLs may be able to expand cases we can avoid
looking at the timers.
Note that the addition of idlepMask to p.init is a no-op. The zero value
of the mask is the correct init value so it is not necessary, but it is
included for clarity.
Benchmark results from WakeupParallel/syscall/pair/race/1ms (see
golang.org/cl/228577). Note that these are on top of golang.org/cl/259578:
Change-Id: I4239cd89f21ad16dfbbef58d81981da48acd0605
Reviewed-on: https://go-review.googlesource.com/c/go/+/264477
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
Antonio Huete Jimenez [Wed, 28 Oct 2020 11:44:26 +0000 (11:44 +0000)]
os: do not use procfs for os.Executable in dragonfly
procfs(5) is not always mounted in DragonFly BSD, for example during
the binary package build with synth. os.Executable() consumers
will then fail, we've spotted this when trying to build tinygo:
[...]
copying source files
./build/tinygo build-builtins -target=armv6m-none-eabi [...]
panic: could not get executable path: readlink /proc/curproc/file:
no such file or directory
Tzu-Chiao Yeh [Mon, 24 Aug 2020 14:04:17 +0000 (22:04 +0800)]
database/sql: fix tx stmt deadlock when rollback
Tx acquires tx.closemu W-lock and then acquires stmt.closemu.W-lock
to fully close the transaction and associated prepared statement.
Stmt query and execution run in reverse ways - acquires
stmt.closemu.R-lock and then acquires tx.closemu.R-lock to grab tx
connection, which may cause deadlock.
Prevent the lock is held around tx.closePrepared to ensure no
deadlock happens.
Fixes #40985
Change-Id: If53909822b87bce11861a6e3035ecb9476d2cd17
Reviewed-on: https://go-review.googlesource.com/c/go/+/250178
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Daniel Theophanes <kardianos@gmail.com>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>