Matthew Dempsky [Thu, 18 Feb 2021 01:49:40 +0000 (17:49 -0800)]
cmd/compile: declare inlined result params early for empty returns
The code for delayed declaration of inlined result parameters only
handles non-empty return statements. This is generally okay, because
we already early declare if there are any (non-blank) named result
parameters.
But if a user writes a function with only blank result parameters and
with exactly one return statement, which is empty, then they could end
up hitting the dreaded "Value live at entry" ICE.
This CL fixes the issue by ensuring we always early declare inlined
result parameters if there are any empty return statements.
Fixes #44355.
Change-Id: I315f3853be436452883b1ce31da1bdffdf24d506
Reviewed-on: https://go-review.googlesource.com/c/go/+/293293
TryBot-Result: Go Bot <gobot@golang.org>
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>
Ivan Trubach [Thu, 25 Feb 2021 19:15:04 +0000 (19:15 +0000)]
database/sql: close driver.Connector if it implements io.Closer
This change allows driver implementations to manage resources in
driver.Connector, e.g. to share the same underlying database handle
between multiple connections. That is, it allows embedded databases
with in-memory backends like SQLite and Genji to safely release the
resources once the sql.DB is closed.
This makes it possible to address oddities with in-memory stores in
SQLite and Genji drivers without introducing too much complexity in
the driver implementations.
See also:
- https://github.com/mattn/go-sqlite3/issues/204
- https://github.com/mattn/go-sqlite3/issues/511
- https://github.com/genjidb/genji/issues/210
Cuong Manh Le [Sun, 21 Feb 2021 15:09:03 +0000 (22:09 +0700)]
cmd/compile: fix mishandling of unsafe-uintptr arguments with call method in go/defer
In CL 253457, we did the same fix for direct function calls. But for
method calls, the receiver argument also need to be passed through the
wrapper function, which we are not doing so the compiler crashes with
the code in #44415.
As we already rewrite t.M(...) into T.M(t, ...) during walkCall1, to fix
this, we can do the same trick in wrapCall, so the receiver argument
will be treated as others.
Fixes #44415
Change-Id: I396182983c85d9c5e4494657da79d25636e8a079
Reviewed-on: https://go-review.googlesource.com/c/go/+/294849
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Andy Pan [Mon, 23 Nov 2020 07:42:48 +0000 (15:42 +0800)]
runtime: batch moving gFree list between local p and global schedt
Change-Id: I0ca1fcee6d3f08bdfcfa51f0dc774118d7355636
Reviewed-on: https://go-review.googlesource.com/c/go/+/271914
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Ian Lance Taylor [Mon, 15 Feb 2021 01:59:48 +0000 (17:59 -0800)]
test: add test case that failed with gccgo
bug511.dir/b.go:10:14: error: reference to undefined field or method 'M'
Change-Id: I9f96dc5c7254b310bc3e15b0bc588d62718cb4b2
Reviewed-on: https://go-review.googlesource.com/c/go/+/292009
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Ian Lance Taylor [Sun, 21 Feb 2021 23:28:41 +0000 (15:28 -0800)]
testing/fstest: treat dash specially when building glob
"[-]" is not a valid path.Match pattern.
Fixes #44474
Change-Id: I0932bbf08ffb8ad0c5337d69d0893f53c1ba89ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/294869
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
Jason A. Donenfeld [Thu, 25 Feb 2021 01:58:04 +0000 (02:58 +0100)]
syscall: return error if GetQueuedCompletionStatus truncates key
This function has the wrong signature, so return an error when that
actually might lead to unexpected results. Users should switch to
x/sys/windows for the real version of this function.
Updates #44538.
Change-Id: I4d1f3d1e380815733ecfea683f939b1d25dcc32a
Reviewed-on: https://go-review.googlesource.com/c/go/+/296154
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Bryan C. Mills <bcmills@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Tobias Klauser [Tue, 23 Feb 2021 13:58:32 +0000 (14:58 +0100)]
syscall, os: use pipe2 syscall on DragonflyBSD instead of pipe
Follow the implementation used by the other BSDs and account for the
intricacy of having to pass the fds array even though the file
descriptors are returned.
Re-submit of CL 130996 with corrected pipe2 wrapper.
Change-Id: Ie36d8214cba60c4fdb579f18bfc1c1ab3ead3ddc
Reviewed-on: https://go-review.googlesource.com/c/go/+/295372
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Jason A. Donenfeld [Thu, 25 Feb 2021 00:29:58 +0000 (01:29 +0100)]
runtime: subtract one from ip when determining abort
On windows/arm, the abort is given from one byte off of the function
address, perhaps because Windows wants to simulate x86/amd64 modes, or
because it's jumping from thumb mode. This is not the case with
windows/arm64, though.
This prevents a failure in the builders with the TestAbort test:
Change-Id: I8939c60611863cc0c325e179a772601acea9fd4a
Reviewed-on: https://go-review.googlesource.com/c/go/+/296153
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Jason A. Donenfeld [Wed, 24 Feb 2021 21:59:59 +0000 (22:59 +0100)]
syscall: restore broken GetQueuedCompletionStatus signature but make it not crash
This reverts commit dc4698f52b5ad3f0251e0cc25bc7ffbd10e23f2c, and then
fixes the memory corruption issue. It also suggests users switch to
x/sys/windows for the proper function.
This requires CL 295174 to be submitted first.
Updates #44538.
Change-Id: I0ee602f1c1d6a89cc585aff426833a4cb3f2be50
Reviewed-on: https://go-review.googlesource.com/c/go/+/296149
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
David Chase [Mon, 1 Feb 2021 18:26:47 +0000 (13:26 -0500)]
cmd/compile: plumb abi info into expandCalls
Work in progress.
TODO:
- insert debugging output for all the steps listed below
- emit modified call instructions w/ multiple register inputs
and Result-typed outputs (next CL)
- initially just change output from "mem" to "Result{mem}"
= most places this hits will be future work.
- change OpArg to use registerized variants
- (done) match abi paramresultinfo with particular arg, use Name
- (this CL) push register offsets for "loads" and "stores" into
recursive decomposition.
- hand registerized Result to exit block
For #40724.
Change-Id: Ie5de9d71f8fd4e092f5ee9260b54de35abf91016
Reviewed-on: https://go-review.googlesource.com/c/go/+/293390
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org>
Dan Scales [Wed, 24 Feb 2021 21:03:17 +0000 (13:03 -0800)]
cmd/compile: disable inlining functions with closures for now
Added a flag '-d=inlfuncswithclosures=1' to allow inlining functions with
closures, and change the default to off for now, until #44370 is fixed.
Updates #44370.
Change-Id: Ic17723aa5c091d91f5f5004d8b63ec7125257acf
Reviewed-on: https://go-review.googlesource.com/c/go/+/296049
Run-TryBot: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Austin Clements [Wed, 24 Feb 2021 01:48:22 +0000 (20:48 -0500)]
cmd/compile/internal-abi: fix ABI0-equivalence for zero-sized values
This fixes a bug in the internal ABI specification that made it not
equivalent to ABI0 even with zero architectural argument registers in
the case of a zero-sized argument with alignment > 1.
In ABI0, even zero-sized arguments cause alignment padding in the
stack frame.
Currently, in the internal ABI, zero-sized arguments get
register-assigned even if there are no registers because they don't
consume any registers. Hence, they don't create alignment padding in
the stack frame.
Austin Clements [Wed, 24 Feb 2021 03:03:29 +0000 (22:03 -0500)]
cmd/compile/internal-abi: use x87 mode, not MMX mode
Florian Weimer pointed out that my justification for using MMX mode
was nonsense and that staying in x87 mode simplifies transitions to
and from C. Hence, switch the spec to say we're always in x87 mode.
Alejandro García Montoro [Wed, 30 Dec 2020 17:41:36 +0000 (18:41 +0100)]
cmd/compile: add rule to coalesce writes
The code generated when storing eight bytes loaded from memory created a
series of small writes instead of a single, large one. The specific
pattern of instructions generated stored 1 byte, then 2 bytes, then 4
bytes, and finally 1 byte.
The new rules match this specific pattern both for amd64 and for s390x,
and convert it into a single instruction to store the 8 bytes. arm64 and
ppc64le already generated the right code, but the new codegen test
covers also those architectures.
Fixes #41663
Change-Id: Ifb9b464be2d59c2ed5034acf7b9c3e473f344030
Reviewed-on: https://go-review.googlesource.com/c/go/+/280456 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Jason A. Donenfeld [Tue, 23 Feb 2021 12:29:40 +0000 (13:29 +0100)]
syscall: do not overflow key memory in GetQueuedCompletionStatus
The third argument to GetQueuedCompletionStatus is a pointer to a
uintptr, not a uint32. Users of this functions have therefore been
corrupting their memory every time they used it. Either that memory
corruption was silent (dangerous), or their programs didn't work so they
chose a different API to use.
Fixes #44538.
RELNOTES=yes
Change-Id: Idf48d4c712d13da29791e9a460159255f963105b
Reviewed-on: https://go-review.googlesource.com/c/go/+/295371
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Egon Elbre [Sat, 2 Jan 2021 14:28:11 +0000 (16:28 +0200)]
cmd/compile: reduce inline cost of OCONVOP
OCONVOP doesn't have effect in the compiled code so, it can be safely
excluded from inline cost calculation.
Also make sequence ODEREF OCONVNOP* OADDR cost 1. This is rather common
conversion, such as *(*uint32)(unsafe.Pointer(&x)).
Fixes #42788
Change-Id: I5001f7e89d985c198b6405694cdd5b819cf3f47a
Reviewed-on: https://go-review.googlesource.com/c/go/+/281232 Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Elias Naur <mail@eliasnaur.com>
YunQiang Su [Tue, 9 Jun 2020 04:09:58 +0000 (04:09 +0000)]
cmd/go, cmd/cgo: pass -mfp32 and -mhard/soft-float to MIPS GCC
For mips32 currently, we are using FP32, while the gcc may be FPXX,
which may generate .MIPS.abiflags and .gnu.attributes section with
value as FPXX. So the kernel will treat the exe as FPXX, and may
choose to use FR=1 FPU mode for it.
Currently, in Go, we use 2 lwc1 to load both half of a double value
to a pair of even-odd FPR. This behavior can only work with FR=0 mode.
In FR=1 mode, all of 32 FPR are 64bit. If we lwc1 the high-half of a double
value to an odd FPR, and try to use the previous even FPR to compute, the
real high-half of even FPR will be unpredicatable.
We set -mfp32 to force the gcc generate FP32 code and section value.
More details about FP32/FPXX/FP64 are explained in:
https://web.archive.org/web/20180828210612/https://dmz-portal.mips.com/wiki/MIPS_O32_ABI_-_FR0_and_FR1_Interlinking
When GOMIPS/GOMIPS64 is set as softfloat, we should also pass
-msoft-float to gcc.
Here we also add -mno-odd-spreg option, since Loongson's CPU cannot use
odd-number FR in FR=0 mode.
Ian Lance Taylor [Fri, 22 Jan 2021 00:59:29 +0000 (16:59 -0800)]
io/ioutil: forward TempFile and TempDir to os package
For #42026
Fixes #44311
Change-Id: I3dabcf902d155f95800b4adf1d7578906a194ce6
Reviewed-on: https://go-review.googlesource.com/c/go/+/285378
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
encoding/json: reduce allocated space in Unmarshal
The decodeState type is a large part of the allocated space during Unmarshal.
The errorContext field is infrequently used, and only on error.
Extract it into a pointer and allocate it separate when necessary.
This test contained a data race.
On line 437, db.BeginTx starts a goroutine that runs tx.awaitDone,
which reads tx.keepConnOnRollback.
On line 445, the test writes to tx.keepConnOnRollback.
tx.awaitDone waits on ctx, but because ctx is timeout-based,
there's no ordering guarantee between the write and the read.
The race detector never caught this before
because the context package implementation of Done
contained enough synchronization to make it safe.
That synchronization is not package of the context API or guarantees,
and the first several releases it was not present.
Another commit soon will remove that synchronization,
exposing the latent data race.
To fix the race, emulate a time-based context
using an explicit cancellation-based context.
This gives us enough control to avoid the race.
Rob Findley [Tue, 23 Feb 2021 17:01:26 +0000 (12:01 -0500)]
go/types: review of call.go
The changes from the (reviewed) dev.regabi copy of call.go can be seen
by comparing patchset 1 and 4. The actual changes are removing the
"// REVIEW INCOMPLETE" marker, deleting some leftover handling of type
instantiation in Checker.call, and adding a comment that exprOrTypeList
should be refactored.
I started to refactor exprOrTypeList, but thought it best to mark this
code as reviewed before diverging from types2.
Change-Id: Icf7fbff5a8def49c5f1781472fd7ba7b73dd9a9c
Reviewed-on: https://go-review.googlesource.com/c/go/+/295531
Trust: Robert Findley <rfindley@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Findley <rfindley@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
DQNEO [Wed, 10 Feb 2021 13:34:09 +0000 (22:34 +0900)]
spec: improve sentence structure for passing a slice
Change-Id: I453d06da2f596eb0b99905aec46a05547d73c62c
Reviewed-on: https://go-review.googlesource.com/c/go/+/290872
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Trust: Robert Griesemer <gri@golang.org>
Trust: Ian Lance Taylor <iant@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org> Reviewed-by: Rob Pike <r@golang.org>
David Benjamin [Fri, 25 Dec 2020 17:02:55 +0000 (12:02 -0500)]
unicode: correctly handle negative runes
Is and isExcludingLatin did not handle negative runes when dispatching
to is16. TestNegativeRune covers this along with the existing uint32
casts in IsGraphic, etc. (For tests, I picked the smallest non-Latin-1
code point in each range.)
Updates #43254
Change-Id: I17261b91f0d2b5b5125d19219411b45c480df74f
Reviewed-on: https://go-review.googlesource.com/c/go/+/280493
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Rob Pike <r@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Andy Pan [Tue, 10 Nov 2020 09:56:33 +0000 (17:56 +0800)]
image: resolve the TODO of doc comment style
Change-Id: Ic7701a9e4635fe1a331c9a1df776ed580759eb9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/268758 Reviewed-by: Nigel Tao <nigeltao@golang.org> Reviewed-by: Rob Pike <r@golang.org>
Trust: Nigel Tao <nigeltao@golang.org>
Trust: Alberto Donizetti <alb.donizetti@gmail.com>
Trust: Rob Pike <r@golang.org>
Rob Findley [Tue, 23 Feb 2021 16:48:15 +0000 (11:48 -0500)]
go/types: minor updates to comments to align with types2
Change-Id: Ic4fcd67cd9222eae6b72d9e91e37f3b0293b0b8d
Reviewed-on: https://go-review.googlesource.com/c/go/+/295530 Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Robert Findley <rfindley@google.com>
Run-TryBot: Robert Findley <rfindley@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Daniel S. Fava [Fri, 12 Feb 2021 08:54:50 +0000 (09:54 +0100)]
testing/race: fixing intermittent test failure
Test NoRaceMutexPureHappensBefore in runtime/race/testdata/mutex_test.go
expects the second spawned goroutine to run after the first. The test
attempts to force this scheduling with a 10 millisecond wait. Following
a suggestion by Bryan Mills, we force this scheduling using a shared
variable whose access take place within the existing mutex.
Fixes #35745.
Change-Id: Ib23ec51492ecfeed4752e020401dd25755a669ed
Reviewed-on: https://go-review.googlesource.com/c/go/+/291292 Reviewed-by: Bryan C. Mills <bcmills@google.com> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Robert Griesemer [Tue, 23 Feb 2021 20:47:08 +0000 (12:47 -0800)]
go/types, types2: constraints may be parenthesized and that includes "any"
Change-Id: I9a234cc1f04ca762375b51ec8ef009fb264c7ed1
Reviewed-on: https://go-review.googlesource.com/c/go/+/295689
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Dmitri Shuralyov [Mon, 22 Feb 2021 22:21:10 +0000 (17:21 -0500)]
doc: start draft go1.17 release notes, move go1.16 to x/website
This template is based on CL 248198 and previous ones like it.
Continue to eagerly include often-used sections, and clarify that
the TODO is about completing the section, or removing if it turns
out not to be needed.
Move the Go 1.16 release notes to x/website, since that's the new
home for past Go release notes as of CL 291711. They're added to
x/website in CL 295249.
'relnote -html' does not report any CLs with RELNOTE annotations
since 2021/02/01.
For #44513.
Updates #40700.
Change-Id: Idd389335500ec4dec2764cbbaa385918cc8a79ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/295209
Trust: Dmitri Shuralyov <dmitshur@golang.org>
Run-TryBot: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Alexander Rakoczy <alex@golang.org> Reviewed-by: Carlos Amedee <carlos@golang.org>
Keith Randall [Wed, 3 Feb 2021 20:25:46 +0000 (12:25 -0800)]
runtime: use BX instead of R15 in race detector
If the race detector were runnable in dynamic linking mode,
then R15 would get clobbered. I don't think it is, so maybe
not a problem, but can't hurt to clean it up.
It also lets CL 283474 pass cleanly when checking the whole stdlib
(together with CL 288452).
Change-Id: I5a5021ecc7e7b8bed1cd3a7067c39b24c09e0783
Reviewed-on: https://go-review.googlesource.com/c/go/+/289270
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Keith Randall [Mon, 1 Feb 2021 08:45:41 +0000 (00:45 -0800)]
runtime: save R15 before checking AVX state
When in dynlink mode, reading a global can clobber R15.
Just to be safe, save R15 before checking the AVX state to see
if we need to VZEROUPPER or not.
This could cause a problem in buildmodes that aren't supported yet.
Keith Randall [Wed, 9 Dec 2020 22:59:40 +0000 (14:59 -0800)]
cmd/compile: fold MOV*nop and MOV*const
MOV*nop and MOV*reg seem superfluous. They are there to keep type
information around that would otherwise get thrown away. Not sure
what we need it for. I think our compiler needs a normalization of
how types are represented in SSA, especially after lowering.
MOV*nop gets in the way of some optimization rules firing, like for
load combining.
For now, just fold MOV*nop and MOV*const. It's certainly safe to
do that, as the type info on the MOV*const isn't ever useful.
R=go1.17
Change-Id: I3630a80afc2455a8e9cd9fde10c7abe05ddc3767
Reviewed-on: https://go-review.googlesource.com/c/go/+/276792
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
zhengjianxun [Tue, 23 Feb 2021 03:12:56 +0000 (03:12 +0000)]
runtime: clarify GC fractional mode description
nowdays, in runtime/mgc.go,we can see the comment descrition : The fractional worker is necessary when GOMAXPROCS*gcBackgroundUtilization is not an integer.
but it not true such as GOMAXPROCS=5.
in the implemet of startCycle() , Fractional Mode happend only when
GOMAXPROCS<=3 or GOMAXPROCS=6. so utilization can closest to 25%.
Fixes #44380
Joel Sing [Mon, 30 Nov 2020 12:47:27 +0000 (23:47 +1100)]
cmd/compile: change riscv64 Eq32/Neq32 to zero extend before subtraction
As done with other equality tests, zero extend before subtraction rather than
after (or in this case, at the same time). While at face value this appears to
require more instructions, in reality it allows for most sign extensions to
be completely eliminated due to correctly typed loads. Existing optimisations
(such as subtraction of zero) then become more effective.
This removes more than 10,000 instructions from the Go binary and in particular,
a writeBarrier check only requires three instructions (AUIPC, LWU, BNEZ) instead
of the current four (AUIPC, LWU, NEGW, BNEZ).
Change-Id: I7afdc1921c4916ddbd414c3b3f5c2089107ec016
Reviewed-on: https://go-review.googlesource.com/c/go/+/274066
Trust: Joel Sing <joel@sing.id.au> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Russ Cox [Tue, 23 Feb 2021 16:38:24 +0000 (11:38 -0500)]
runtime: fix usleep on windows/arm
Changed calling convention to pre-multiply the argument by -100,
and then deleted the * 100 but not the negation in the windows/arm assembly.
Delete the negation.
Fixes the current all.bash breakage on windows/arm builder.
(Maybe that will uncover more.)
This is partial plumbing recycled from the original register abi test work;
these are the parts that translate easily. Some other bits are deferred till
later when they are ready to be used.
Keith Randall [Fri, 12 Feb 2021 23:55:25 +0000 (15:55 -0800)]
cmd/compile: improve bit test code
Some bit test instruction generation stopped triggering after
the change to addressing modes. I suspect this was just because
ANDQload was being generated before the rewrite rules could discover
the BTQ. Fix that by decomposing the ANDQload when it is surrounded
by a TESTQ (thus re-enabling the BTQ rules).
Jason A. Donenfeld [Mon, 15 Feb 2021 19:20:15 +0000 (20:20 +0100)]
cmd/go: recognize DLL magic from llvm binaries
When using LLD with c-shared, the magic in the output DLL is slightly
different than for EXEs.
Change-Id: Icc5f34f7bb61f11a9d75494236b7797cc1988b40
Reviewed-on: https://go-review.googlesource.com/c/go/+/291638
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Cherry Zhang [Tue, 23 Feb 2021 00:45:21 +0000 (19:45 -0500)]
cmd/compile: guard special register usage with GOEXPERIMENT=regabi
Previously, some special register uses are only guarded with ABI
wrapper generation (-abiwrap). This CL makes it also guarded with
the GOEXPERIMENT. This way we can enable only the wrapper
generation without fully the new ABI, for benchmarking purposes.
Change-Id: I90fc34afa1dc17c9c73e7b06e940e79e4c4bf7f6
Reviewed-on: https://go-review.googlesource.com/c/go/+/295289
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Andy Pan [Tue, 10 Nov 2020 13:02:18 +0000 (21:02 +0800)]
runtime: optimize the memory padding in p struct
Since allocation for p struct will be rounded up to the next size class,
the two relevant adjacent classes for this case are 9728 bytes and 10240 bytes.
A p is currently 10072 bytes, so it gets rounded up to 10240 bytes when we allocate one,
So the pad in p struct is unnecessary, eliminate it and add comments for
warning the false sharing.
Change-Id: Iae8b32931d1beddbfff1f58044d8401703da6407
Reviewed-on: https://go-review.googlesource.com/c/go/+/268759 Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Ian Lance Taylor <iant@golang.org>
Jason A. Donenfeld [Sun, 14 Feb 2021 18:43:31 +0000 (19:43 +0100)]
runtime/cgo: use correct lean and mean macro
WIN64_LEAN_AND_MEAN is not the correct macro to use and doesn't ever
exist.
Change-Id: I32a5523cc0f7cc3f3a4d022071cf81f88db39aa9
Reviewed-on: https://go-review.googlesource.com/c/go/+/291634
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Trust: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Change-Id: I8766104508f7acd832088a590ee7d68afa0d6065
Reviewed-on: https://go-review.googlesource.com/c/go/+/291633
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Change-Id: Iffd2345d757f23dd737e63bd464cd412527077c4
Reviewed-on: https://go-review.googlesource.com/c/go/+/291632
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Jason A. Donenfeld [Sat, 13 Feb 2021 23:04:48 +0000 (00:04 +0100)]
cmd/link: set SizeOfRawData rather than VirtualSize in COFF files for .bss section
GCC and Clang both set the SizeOfRawData field rather than the
VirtualSize field for communicating the size of the .bss section. As a
consequence, LLD does not look at VirtualSize and collapses the .bss
section into whatever is around it, resulting in runtime crashes. This
commit changes the logic so that if the requested "file size" is 0, then
the SizeOfRawData field is set rather than the VirtualSize field as the
sole length marker.
Change-Id: Ied89ddaa0a717fed840238244c6e4848845aeeb6
Reviewed-on: https://go-review.googlesource.com/c/go/+/291630
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Than McIntosh [Wed, 6 Jan 2021 18:27:17 +0000 (13:27 -0500)]
cmd/internal/dwarf: minor cleanups
Remove a stale comment, demote PutInlinedFunc from public to private,
and remove an unused interface originally used for sorting vars.
No change in functionality.
Change-Id: I5ee1ad2b10b78b158e2223c6979bab830202db95
Reviewed-on: https://go-review.googlesource.com/c/go/+/295009
Trust: Than McIntosh <thanm@google.com>
Run-TryBot: Than McIntosh <thanm@google.com> Reviewed-by: Jeremy Faller <jeremy@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Joel Sing [Mon, 7 Dec 2020 17:14:34 +0000 (04:14 +1100)]
runtime: enable race detector on openbsd/amd64
Now that this commit[1] has landed in LLVM the .syso file can be generated for
OpenBSD.
With the changes to src/runtime running the sample race[2] detects the data
race as expected.
Based on golang/go#39464 (https://go-review.googlesource.com/c/go/+/237057) from
Aaron Bieber <deftly@gmail.com>, however the race_openbsd_amd64.syso file has
been built on OpenBSD 6.4 and necessary changes added to race.bash.
Change-Id: Ic2479ccfa91d6b2cb4585346a11d813d96450f68
Reviewed-on: https://go-review.googlesource.com/c/go/+/275892
Trust: Joel Sing <joel@sing.id.au>
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Joel Sing [Wed, 17 Feb 2021 10:03:59 +0000 (21:03 +1100)]
cmd/internal/obj/riscv: prevent constant loads that do not target registers
Check that the target of a constant load is a register and add test coverage
for this error condition. While here, rename the RISC-V testdata and tests
to be consistent with other platforms.
Change-Id: I7fd0bfcee8cf9df0597d72e65cd74a2d0bfd349a
Reviewed-on: https://go-review.googlesource.com/c/go/+/292895
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>
Cuong Manh Le [Wed, 17 Feb 2021 04:38:14 +0000 (11:38 +0700)]
cmd/compile: remove selectnbrecv2
Previous CL did remove selectnbrecv2 in runtime, the compiler now only
call selectnbrecv, so remove this. Make this as separated CL because it
adds much of noise to git stat.
Change-Id: I06e89c823c0403e9bd66f2633409c455a46d6e79
Reviewed-on: https://go-review.googlesource.com/c/go/+/292891
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Cuong Manh Le [Wed, 17 Feb 2021 04:36:58 +0000 (11:36 +0700)]
cmd/compile,runtime: make selectnbrecv return two values
The only different between selectnbrecv and selectnbrecv2 is the later
set the input pointer value by second return value from chanrecv.
So by making selectnbrecv return two values from chanrecv, we can get
rid of selectnbrecv2, the compiler can now call only selectnbrecv and
generate simpler code.
Change-Id: Ifaf6cf1314c4f47b06ed9606b1578319be808507
Reviewed-on: https://go-review.googlesource.com/c/go/+/292890
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Cuong Manh Le [Fri, 27 Nov 2020 19:46:00 +0000 (02:46 +0700)]
cmd/compile: use transitive relations for slice len/cap in poset
Currently, we keep track of slice len by mapping from slice ID to
len/cap SSA value. However, slice len/cap can have multiple SSA values,
so when updating fact table for slice len/cap, we only update in one
place.
Instead, we can take advantage of the transitive relations provided by
poset. So all duplicated slice lens are set as equal to one another.
When updating fact table for one, that fact will be reflected to all
others. The same mechanism is applied for slice cap.
Removes 15 bounds checks from std/cmd.
Fixes #42603
Change-Id: I32c07968824cc33765b1e441b3ae2c4b5f5997c3
Reviewed-on: https://go-review.googlesource.com/c/go/+/273670
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Robert Griesemer [Sat, 20 Feb 2021 04:06:24 +0000 (20:06 -0800)]
cmd/compile/internal/types2: review of sanitize.go
Remove the "// UNREVIEWED" marker and add guards (as in go/types)
to prevent data races. To see the added guards, see compare patch
sets 3 and 4. The equivalent changes for go/types were done in
https://golang.org/cl/294411.
Change-Id: Ibef07eaae400bd32bff32b102cc743580290d135
Reviewed-on: https://go-review.googlesource.com/c/go/+/294510
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Robert Griesemer [Tue, 23 Feb 2021 01:02:46 +0000 (17:02 -0800)]
go/types, cmd/compile/internal/types2: use regular type printing for unsafe.Pointer
Type string printing special-cased printing of unsafe.Pointer because
it's a built-in type; yet it's declared in a package like any other
imported or used-defined type (unlike built-in types such as int).
Use the same mechanism for printing unsafe.Pointer like any other
(non-basic) type. This will make it possible to use the package
Qualifier if so desired.
Fixes #44515.
Change-Id: I0dd1026f850737ecfc4bb99135cfb8e3c18be9e7
Reviewed-on: https://go-review.googlesource.com/c/go/+/295271
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Robert Griesemer [Sat, 20 Feb 2021 04:00:33 +0000 (20:00 -0800)]
cmd/compile/internal/types2: review of conversions.go
The changes between (equivalent, and reviewed) go/types/conversions.go
and conversions.go can be seen by comparing patchset 1 and 2. The actual
change is just removing the "// UNREVIEWED" marker.
Change-Id: I86d20d8100ec29fe3be996b975c9b4aff01be85e
Reviewed-on: https://go-review.googlesource.com/c/go/+/294509
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Robert Griesemer [Sat, 20 Feb 2021 00:58:24 +0000 (16:58 -0800)]
cmd/compile/internal/types2: review of issues_test.go
The changes between (equivalent, and reviewed) go/types/issues_test.go
and issues_test.go can be seen by comparing patchset 1 and 3. The actual
change is just removing the "// UNREVIEWED" marker and making making
some minor code adjustments to match go/types's version more closely.
Change-Id: I26f3f700d12db69fc68161a6b0dc081a0e9cd0d4
Reviewed-on: https://go-review.googlesource.com/c/go/+/294473
Trust: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Dan Scales [Fri, 19 Feb 2021 17:40:35 +0000 (09:40 -0800)]
cmd/compile: fix import of functions of multiple nested closure
For import of functions with closures, the connections among closure
variables are constructed on-the-fly via CaptureName(). For multiple
nested closures, we need to temporarily set r.curfn to each closure we
construct, so that the processing of closure variables will be correct
for any nested closure inside that closure.
Fixes #44335
Change-Id: I34f99e2822250542528ff6b2232bf36756140868
Reviewed-on: https://go-review.googlesource.com/c/go/+/294212
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
yangwenmai [Fri, 25 Dec 2020 01:36:41 +0000 (09:36 +0800)]
all: faster midpoint computation in binary search
On my machine (3.1 GHz Quad-Core Intel Core i7, macOS 10.15.7 10.15.7), go 1.15.6
benchstat:
name old time/op new time/op delta
SearchInts-8 20.3ns ± 1% 16.6ns ± 6% -18.37% (p=0.000 n=9+10)
Change-Id: I346e5955fd6df6ce10254b22267dbc8d5a2b16c0
Reviewed-on: https://go-review.googlesource.com/c/go/+/279439 Reviewed-by: Ben Shi <powerman1st@163.com> Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Robert Griesemer <gri@golang.org>
Rob Findley [Mon, 22 Feb 2021 20:39:00 +0000 (15:39 -0500)]
go/parser: return ast.BadExpr for missing index operands
The parser was returning the indexed operand when a slice or index or
instance expression was missing any index arguments (as in the
expression `a[]`). This can result in returning an *ast.Ident for the
LHS of the (invalid) assignment `a[] = ...` -- in this case parsing the
LHS as just `a`. Unfortunately, as the indexed operand `a` has already
been resolved, this results in a panic for duplicate resolution.
Fix this by instead returning an ast.BadExpr. This can suppress some
subsequent errors from the typechecker, but those errors may or may not
be correct anyway. Other interpretations, such as an *ast.IndexExpr with
bad or missing X, run into potential misinterpretations downstream (both
caused errors in go/types and/or gopls).
Fixes #44504
Change-Id: I5ca8bed4a1861bcc7db8898770b08937110981d4
Reviewed-on: https://go-review.googlesource.com/c/go/+/295151
Trust: Robert Findley <rfindley@google.com>
Run-TryBot: Robert Findley <rfindley@google.com>
TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Cherry Zhang [Mon, 22 Feb 2021 22:43:08 +0000 (17:43 -0500)]
cmd/compile: correctly use X15 to zero frame
In CL 288093 we reserve X15 as the zero register and use that to
zero values. It only covered zeroing generated in SSA but missed
zeroing the frame generated late in the compilation. The latter
still uses X0, but now DUFFZERO expects X15, so it doesn't
actually zero the frame. Change it to use X15.
Should fix #44333.
Change-Id: I239d2b78a5f6468bc86b70aecdd294045311759f
Reviewed-on: https://go-review.googlesource.com/c/go/+/295210
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>
cmd/compile: reject some rare looping CFGs in shortcircuit
One CFGs that shortcircuit looks for is:
p q
\ /
b
/ \
t u
The test case creates a CFG like that in which p == t.
That caused the compiler to generate a (short-lived) invalid phi value.
Fix this with a relatively big hammer: Disallow single-length loops entirely.
This is probably overkill, but it such loops are very rare.
This doesn't change the generated code for anything in std.
It generates worse code for the test case:
It no longer compiles the entire function away.