sparse conditional constant propagation can discover optimization opportunities that cannot be found by just combining constant folding and constant propagation and dead code elimination separately.
Updates #59399
Change-Id: Ia954e906480654a6f0cc065d75b5912f96f36b2e
GitHub-Last-Rev: 90fc02db99
GitHub-Pull-Request: golang/go#59575
Reviewed-on: https://go-review.googlesource.com/c/go/+/483875
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
Before this CL, we use GetFrom3&SetFrom3 to get or set a source operand
which not fit into Prog.Reg. Those APIs operate the first element in
Prog.RestArgs without checking the type so they're fragile to break if
we have more than one different type of operands in the slice, which
will be a common case in Arm64.
This CL deprecates & renames some APIs related to Prog.RestArgs to make
those APIs more reasonable and robust than before.
Change-Id: I70d56edc1f23ccfffbcd6df34844e2cef2288432
Reviewed-on: https://go-review.googlesource.com/c/go/+/493355
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Eric Fang <eric.fang@arm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Eric Fang <eric.fang@arm.com>
With the Garbage collector (GC), we observed a false-sharing between
work.full and work.empty. (referenced most from runtime.gcDrain and
runtime.getempty)
This false-sharing becomes worse and impact performance on multi core
system. On Intel Xeon 8480+ and default GC setting(GC=100), we can
observed top HITM>4% (by perf c2c)caused by it.
After resolveed this false-sharing issue, we can get performance 8%~9.7%
improved. Verify workloads:
DeathStarBench/hotelReservation: 9.7% of RPS improved
https://github.com/delimitrou/DeathStarBench/tree/master/hotelReservation
gRPC-go/benchmark: 8% of RPS improved
https://github.com/grpc/grpc-go/tree/master/benchmark
gRPC-go/benchmark 9 iterations' data with master branch:
master w/ fs opt.
208862.4 246390.9
221680.0 266019.3
223886.9 248789.7
212169.3 257837.8
219922.4 234331.8
197401.7 261627.7
214562.4 255429.7
214328.5 237087.8
229443.2 230591.3
max 229443.2 266019.3 116%
med 214562.4 248789.7 116%
avg 215806.3 248678.5 115%
Change-Id: Ib386de021cd2dbb802a107f487556d848ba9212d
Reviewed-on: https://go-review.googlesource.com/c/go/+/496915
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
This test is meant to detect the effect of Chdir not being
observed in other concurrent goroutines, possible in Plan 9
because each M runs in a separate OS process with its own
working directory. The test depends on Getwd to report the
correct working directory, but if Chdir fails then Getwd
may fail for the same reasons. We add a consistency check
that Stat(Getwd()) and Stat(".") refer to the same file.
Also change channel usage and add a sync.WaitGroup to
ensure test goroutines are not left blocked or running
when the main test function exits.
For #58802
Change-Id: I80d554fcf3617427c28bbe16e5e396367dcfe673
Reviewed-on: https://go-review.googlesource.com/c/go/+/472555
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: David du Colombier <0intro@gmail.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Ian Lance Taylor <iant@golang.org>
In the type checkers, add Config.ErrorURL (or Config._ErrorURL for
go/types) to configure whether and how an error message should report
a URL for errors that have an error code.
In the compiler, configure types2 to report an error URL of the form
" [go.dev/e/XXX]", where XXX stands for the error code, with the URL
appended to the first line of an error.
Rename the compiler flag -url to -errorurl. At the moment this flag
is disabled by default.
Example for a one-line error message:
<pos>: undefined: f [go.dev/e/UndeclaredName]
Example for a multi-line error message:
<pos>: not enough arguments in call to min [go.dev/e/WrongArgCount]
have ()
want (P, P)
Change-Id: I26651ce2c92ad32fddd641f003db37fe12fdb1cd
Reviewed-on: https://go-review.googlesource.com/c/go/+/497715
Auto-Submit: Robert Griesemer <gri@google.com>
Reviewed-by: Robert Findley <rfindley@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Robert Griesemer <gri@google.com>
Reviewed-by: Robert Griesemer <gri@google.com>
When unifying two types A and B where one or both of them are
interfaces, consider the shared method signatures in unification.
1) If a defined interface (an interface with a type name) is unified
with another (defined) interface, currently they must originate
in the same type declaration (same origin) for unification to
succeed. This is more restrictive than necessary for assignments:
when interfaces are assigned to each other, corresponding methods
must match, but the interfaces don't have to be identical.
In unification, we don't know which direction the assignment is
happening (or if we have an assignment in the first place), but
in any case one interface must implement the other. Thus, we
check that one interface has a subset of the methods of the other
and that corresponding method signatures unify.
The assignment or instantiation may still not be possible but that
will be checked when instantiation and parameter passing is checked.
If two interfaces are compared as part of another type during
unification, the types must be equal. If they are not, unifying
a method subset may still succeed (and possibly produce more type
arguments), but that is ok: again, subsequent instantiation and
assignment will fail if the types are indeed not identical.
2) In a non-interface type is unified with an interface, currently
unification fails. If this unification is a consequence of an
assignment (parameter passing), this is again too restrictive:
the non-interface type must only implement the interface (possibly
among other type set requirements). In any case, all methods of the
interface type must be present in the non-interface type and unify
with the corresponding interface methods. If they don't, unification
will fail either way. If they do, we may infer additional type
arguments. Again, the resulting types may still not be correct but
that will be determined by the instantiation and parameter passing
or assignment checks. If the non-interface type and the interface
type appear as component of another type, unification may now
produce additional type arguments. But that is again ok because the
respective types won't pass instantiation or assignment checks since
they are different types.
This CL introduces a new unifier flag, enableInterfaceInference, to
enable this new behavior. It is currently disabled.
For #60353.
For #41176.
For #57192.
Change-Id: I983d0ad5f043c7fe9d377dbb95f6b9342f36f45f
Reviewed-on: https://go-review.googlesource.com/c/go/+/497656
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@google.com>
Reviewed-by: Robert Findley <rfindley@google.com>
Run-TryBot: Robert Griesemer <gri@google.com>
Auto-Submit: Robert Griesemer <gri@google.com>
This relocation is not (yet?) defined in ELFv2, but has been added to
gnu gas a couple years ago. It is the same reloc as
R_PPC64_REL24_NOTOC, but hints power10 instructions should not be
emitted.
See binutils commit 7aba54da426b9999085d8f84e7896b8afdbb9ca6.
Fixes#60348
Change-Id: Ie953cd7bf1ffc621b498d4dbebb5de1231833c8f
Reviewed-on: https://go-review.googlesource.com/c/go/+/496918
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Currently the pinner is created outside of the benchmarking loop.
However, this means that we get to reuse the same pinner for each loop;
in general, users are expected to create a pinner for a e.g. a cgo
call and then that variable will expire with the frame it lives in. (If
they can reuse the variable, great! However, I don't expect that to be
common.)
In essence, this benchmarks a harder case. It's not more right or wrong
than the previous version, but the fact that it's a slightly harder case
(that still mostly captures what the original version was capturing) is
useful.
Change-Id: I94987127f54d7bfecd7b8e6a5e632631ea57ad24
Reviewed-on: https://go-review.googlesource.com/c/go/+/497616
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
When unifying two types A and B where one or both of them are
interfaces, consider the shared method signatures in unification.
1) If a defined interface (an interface with a type name) is unified
with another (defined) interface, currently they must originate
in the same type declaration (same origin) for unification to
succeed. This is more restrictive than necessary for assignments:
when interfaces are assigned to each other, corresponding methods
must match, but the interfaces don't have to be identical.
In unification, we don't know which direction the assignment is
happening (or if we have an assignment in the first place), but
in any case one interface must implement the other. Thus, we
check that one interface has a subset of the methods of the other
and that corresponding method signatures unify.
The assignment or instantiation may still not be possible but that
will be checked when instantiation and parameter passing is checked.
If two interfaces are compared as part of another type during
unification, the types must be equal. If they are not, unifying
a method subset may still succeed (and possibly produce more type
arguments), but that is ok: again, subsequent instantiation and
assignment will fail if the types are indeed not identical.
2) In a non-interface type is unified with an interface, currently
unification fails. If this unification is a consequence of an
assignment (parameter passing), this is again too restrictive:
the non-interface type must only implement the interface (possibly
among other type set requirements). In any case, all methods of the
interface type must be present in the non-interface type and unify
with the corresponding interface methods. If they don't, unification
will fail either way. If they do, we may infer additional type
arguments. Again, the resulting types may still not be correct but
that will be determined by the instantiation and parameter passing
or assignment checks. If the non-interface type and the interface
type appear as component of another type, unification may now
produce additional type arguments. But that is again ok because the
respective types won't pass instantiation or assignment checks since
they are different types.
This CL introduces a new Config flag, EnableInterfaceInference, to
enable this new behavior. If not set, unification remains unchanged.
To be able to test the flag durign unification, a *Checker is passed
and stored with the unifier.
For #60353.
Fixes#41176.
Fixes#57192.
Change-Id: I6b167a9afa378d0682e9b101d9d86f5777308af7
Reviewed-on: https://go-review.googlesource.com/c/go/+/497015
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@google.com>
Run-TryBot: Robert Griesemer <gri@google.com>
Reviewed-by: Robert Findley <rfindley@google.com>
Auto-Submit: Robert Griesemer <gri@google.com>
b.ResetTimer used to also stop the timer, however it does not anymore.
These benchmarks hadn't been fixed and as a result ended up measuring
some additional things.
Also, make some for loops more conventional.
Change-Id: I76ca68456d85eec51722a80587e5b2c9f5d836a1
Reviewed-on: https://go-review.googlesource.com/c/go/+/496996
Run-TryBot: Damien Neil <dneil@google.com>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Auto-Submit: Damien Neil <dneil@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Run-TryBot: Ian Lance Taylor <iant@google.com>
CL 494915 introduced an additional fcntl(F_GETFL) syscall to determine
whether the file is in append-only mode. The existing unix.IsNonblock
call also issues an fcntl(F_GETFL) syscall. The two can be combined and
both the append-only mode and the non-blocking flags can be determined
from that syscall's result.
Change-Id: I915589ed94e079f6abaa2fd0032ef01f78698f7f
Reviewed-on: https://go-review.googlesource.com/c/go/+/497075
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Auto-Submit: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
In order to have some test coverage of concurrent use of the go/types
APIs, update the Stdlib test to type-check concurrently. In combination
with non-deterministic ordering, this should hopefully provide moderate
test coverage of concurrent use.
Also, remove the arbitrary 10ms timeout in short mode, in favor of
simply not running.
After this change, TestStdlib went from taking 16s on my laptop to 2s,
in part because of the parallelism and in part because we are no longer
type-checking twice (once for the import er, once for the test).
Fixesgolang/go#47729
Change-Id: Ie49743947ab2d5aec051c3d09ce045acf5b94ad4
Reviewed-on: https://go-review.googlesource.com/c/go/+/484540
Reviewed-by: Robert Griesemer <gri@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Robert Findley <rfindley@google.com>
Currently the CPU stats are only updated once every mark termination,
but for writing robust tests, it's often useful to force this update.
Refactor the CPU stats accumulation out of gcMarkTermination and into
its own function. This is also a step toward real-time CPU stats.
While we're here, fix some incorrect documentation about dedicated GC
CPU time.
For #59749.
For #60276.
Change-Id: I8c1a9aca45fcce6ce7999702ae4e082853a69711
Reviewed-on: https://go-review.googlesource.com/c/go/+/487215
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Currently pidleget gets passed "now" from before the M goes into
netpoll, resulting in incorrect accounting of idle CPU time.
lastpoll is also stored with a stale "now": the mistake was added in the
same CL it was added for pidleget.
Recompute "now" after returning from netpoll.
Also, start tracking idle time on js/wasm at all.
Credit to Rhys Hiltner for the test case.
Fixes#60276.
Change-Id: I5dd677471f74c915dfcf3d01621430876c3ff307
Reviewed-on: https://go-review.googlesource.com/c/go/+/496183
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
CL 466099 rewrote stack symbolization in race reports. Prior to this
CL, physical frames consisting entirely of wrapper logical frame would
print the wrapper, even though in other cases we try to avoid
printing wrappers. CL 466099 unintentionally changed this behavior and
now physical frames consisting entirely of wrapper frames instead fail
to symbolize and print "??()".
Fix this by taking the outermost wrapper frame if the entire logical
frame expansion consists of wrappers.
Fixes#60245.
Change-Id: I13de8857e508b757ea10d1fc7a47258d7fddbfdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/497235
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
Auto-Submit: Austin Clements <austin@google.com>
As part of the work for #57179 we moved configurable defaults
to GOROOT/go.env, so that packagers don't have to modify
source code to change those defaults. Since packagers may want
to modify GOTOOLCHAIN's default, move it to go.env too.
This CL modifies 'go env' to print GOTOOLCHAIN by default.
It also refines CL 496957 from yesterday to recognize any env
var in either go.env or the user go/env, not just the user go/env.
When I put GOTOOLCHAIN in go.env, but before I added it to
the default printing list, 'go env GOTOOLCHAIN' was printing
an empty string, and it was incredibly confusing.
For #57001.
Fixes#60361 while we're here.
Also includes a fix for a review comment on CL 497079 that I forgot to mail.
Change-Id: I7b904d9202f05af789aaa33aed93f903b515aa28
Reviewed-on: https://go-review.googlesource.com/c/go/+/497437
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
CL 493275 started using gcBits for pinner bits. This means gcBits can be
allocated while holding the mspanSpecial lock. This is safe because
these were just parallel in the partial order, but now they need an
explicit edge between them.
For #58277.
Change-Id: I37917730e12d59cf0580f198d732198413a56424
Reviewed-on: https://go-review.googlesource.com/c/go/+/497475
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Reapples CL 495596, which was reverted at CL 496185. The x/tools
failure, #60263, has been resolved. The ppc64 failures, #60368, have
_not_ been resolved, but are believed to be specific to that port. This
CL will make ppc64 flaky while the issue is investigated, but give more
soak time on primary ports.
Build the compiler with PGO. As go build -pgo=auto is enabled by
default, we just need to store a profile in the compiler's
directory.
The profile is collected from building all std and cmd packages on
Linux/AMD64 machine, using profile.sh.
This improves the compiler speed. On Linux/AMD64,
name old time/op new time/op delta
Template 138ms ± 5% 136ms ± 4% -1.44% (p=0.005 n=36+39)
Unicode 147ms ± 4% 140ms ± 4% -4.99% (p=0.000 n=40+39)
GoTypes 780ms ± 3% 778ms ± 4% ~ (p=0.172 n=39+39)
Compiler 105ms ± 5% 99ms ± 7% -5.64% (p=0.000 n=40+40)
SSA 5.83s ± 6% 5.80s ± 6% ~ (p=0.556 n=40+40)
Flate 89.0ms ± 5% 87.0ms ± 6% -2.18% (p=0.000 n=40+40)
GoParser 172ms ± 4% 167ms ± 4% -2.72% (p=0.000 n=39+40)
Reflect 333ms ± 4% 333ms ± 3% ~ (p=0.426 n=40+39)
Tar 128ms ± 4% 126ms ± 4% -1.82% (p=0.000 n=39+39)
XML 173ms ± 4% 170ms ± 4% -1.39% (p=0.000 n=39+40)
[Geo mean] 253ms 248ms -2.13%
The profile is pretty transferable. Using the same profile, we
see a bigger win on Darwin/ARM64,
name old time/op new time/op delta
Template 71.0ms ± 2% 68.3ms ± 2% -3.90% (p=0.000 n=20+20)
Unicode 71.8ms ± 2% 66.8ms ± 2% -6.90% (p=0.000 n=20+20)
GoTypes 444ms ± 1% 428ms ± 1% -3.53% (p=0.000 n=19+20)
Compiler 48.9ms ± 3% 45.6ms ± 3% -6.81% (p=0.000 n=20+20)
SSA 3.25s ± 2% 3.09s ± 1% -5.03% (p=0.000 n=19+20)
Flate 44.0ms ± 2% 42.3ms ± 2% -3.72% (p=0.000 n=19+20)
GoParser 76.7ms ± 1% 73.5ms ± 1% -4.15% (p=0.000 n=18+19)
Reflect 172ms ± 1% 165ms ± 1% -4.13% (p=0.000 n=20+19)
Tar 63.1ms ± 1% 60.4ms ± 2% -4.24% (p=0.000 n=19+20)
XML 83.2ms ± 2% 79.2ms ± 2% -4.79% (p=0.000 n=20+20)
[Geo mean] 127ms 121ms -4.73%
For #60368.
Change-Id: I2cec0fc85e21c38d57ba6f0e5e90cde5d443ebd2
Reviewed-on: https://go-review.googlesource.com/c/go/+/497455
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Updates the DecryptPKCS1v15SessionKey function comment to be less cut
and dry about its protections against Bleichenbacher attacks. In
particular note that the protocol using this method must be explicitly
designed with these mitigations in mind, and call out usages which
may cause the migiations to be useless.
Change-Id: I06fd25157f12a3afb401bb08dff4faef7fb0a9b0
Reviewed-on: https://go-review.googlesource.com/c/go/+/469235
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Roland Shoemaker <roland@golang.org>
Auto-Submit: Roland Shoemaker <roland@golang.org>
Reviewed-by: David Chase <drchase@google.com>
We already refuse to build code in modules are too new (CL 476279).
This is a more comprehensive check: refuse to do anything at all with
modules or workspaces that are too new.
Since the module or workspace is new, it may have semantics we don't
understand and misinterpret well before we get to the actual building of code.
For example when we switched from // +build to //go:build that changed
the decision about which files go into a package, which affects the way
the overall load phase runs and which errors it reports. Waiting until the
building of code would miss earlier changes like that one.
Leaving the test from CL 476279 alone, but it's not load-bearing anymore.
For #57001.
Change-Id: I8c39943db1d7ddbcb9b5cae68d80459fddd68151
Reviewed-on: https://go-review.googlesource.com/c/go/+/497435
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Auto-Submit: Russ Cox <rsc@golang.org>
If a hash match is "disabled" (!enabled) indicate that in the
output with DISABLED. This is helpful in ensuring that multiple
package-directed command-line flags have the intended behavior,
e.g.
```
go build -a \
-gcflags=all=-d=gossahash=vn \
-gcflags=runtime=-d=gossahash=vy \
std
```
Output looks like
[DISABLED] [bisect-match 0x11d0ee166d9d61b4]
or (w/ "v"-prefixed hashcode )
sort/slice.go:23:29 note [DISABLED] [bisect-match 0xa5252e1c1b85f2ec]
gossahash triggered sort/slice.go:23:29 note [DISABLED] 100001011111001011101100
Change-Id: I797e02b3132f9781d97bacd0dcd2e80af0035cd8
Reviewed-on: https://go-review.googlesource.com/c/go/+/497216
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
To make the new go lines work with 'go get' as minimum requirements,
this CL creates a synthetic 'go' module that has as its versions the valid
versions that can be listed on the 'go' line.
In preparation for allowing 'toolchain' changes as well, an equivalent
synthetic module is introduced for 'toolchain'.
For #57001.
Change-Id: Id0ebbd283f0f991859d516d21dffe59a834db540
Reviewed-on: https://go-review.googlesource.com/c/go/+/497080
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
Auto-Submit: Russ Cox <rsc@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
go install m@v and go run m@v are the only commands
that ignore the local go.mod. As such they need to use a
different signal to find the Go version, namely the m@v go.mod.
Because there is no way to predict that Go version (no equivalent
of "go version" for interrogating the local go.mod), if we do switch
toolchains we always print about it.
For #57001.
Change-Id: I981a0b8fa61992b353589355ba72a3b9d55914e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/497079
Auto-Submit: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
