An address of offset(SP) may point to the callee args area, and
may be used to move things into/out of the args/results. If an
address like that is spilled and picked up by the GC, it may hold
an arg/result live in the callee, which may not actually be live
(e.g. a result not initialized at function entry). Make sure
they are rematerializeable, so they are always short-lived and
never picked up by the GC.
This CL changes 386, PPC64, and Wasm. On AMD64 we already have
the rule (line 2159). On other architectures, we already have
similar rules like
(OffPtr [off] ptr:(SP)) => (MOVDaddr [int32(off)] ptr)
to avoid this problem. (Probably me in the past had run into
this...)
Fixes#42944.
Change-Id: Id2ec73ac08f8df1829a9a7ceb8f749d67fe86d1e
Reviewed-on: https://go-review.googlesource.com/c/go/+/275174
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
timer.when must always be positive. addtimer and modtimer already check
that it is non-negative; we expand it to include zero. Also upgrade from
pinning bad values to throwing, as these values shouldn't be possible to
pass (except as below).
timeSleep may overflow timer.nextwhen. This would previously have been
pinned by resetForSleep, now we fix it manually.
runOneTimer may overflow timer.when when adding timer.period. Detect
this and pin to maxWhen.
addtimer is now too strict to allow TestOverflowRuntimeTimer to test an
overflowed timer. Such a timer should not be possible; to help guard
against accidental inclusion siftup / siftdown will check timers as it
goes. This has been replaced with tests for period and sleep overflows.
Change-Id: I17f9739e27ebcb20d87945c635050316fb8e9226
Reviewed-on: https://go-review.googlesource.com/c/go/+/274853
Trust: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
timer when == 0, in the context of timer0When and timerModifiedEarliest,
is a sentinel value meaning there are no timers on the heap.
TestCheckRuntimeTimerOverflow reaching into the runtime to set a timer
to when = 0 when it is otherwise not possible breaks this invariant.
After golang.org/cl/258303, we will no longer detect and run this timer,
thus blocking any other timers lower on the heap from running. This
manifests as random timers failing to fire in other tests.
The need to set this overflowed timer to when = 0 is gone with the old
timer proc implementation, so we can simply remove it.
Fixes#42424
Change-Id: Iea32100136ad8ec1bedfa77b1e7d9ed868812838
Reviewed-on: https://go-review.googlesource.com/c/go/+/274632
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Michael Pratt <mpratt@google.com>
This resolves all TODOs for the runtime and compiler and mentions
several other changes.
For #40700.
Fixes#42892.
Fixes#42894.
Change-Id: I18d14cfe572baf679ecf8b0a4e82c4b866da5a04
Reviewed-on: https://go-review.googlesource.com/c/go/+/275176
Trust: Austin Clements <austin@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
It's useful to have quick access to the types.Field that a given
selector or method value expression refer to. Previously we abused Opt
for this, but couldn't do that for OCALLPART because escape analysis
uses Opt.
Now that we have more flexibility, we can simply add additional
pointer fields for this. This also allows getting rid of an unneeded
ONAME node for OCALLPART.
Passes buildall w/ toolstash -cmp.
Change-Id: I980d7bdb19abfd0b6f58a232876861b88dee1e47
Reviewed-on: https://go-review.googlesource.com/c/go/+/275034
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
The s390x assembly for shlVU does a forward copy when the shift amount s
is 0. This causes corruption of the result z when z is aliased to the
input x.
This fix removes the s390x assembly for both shlVU and shrVU so the pure
go implementations will be used.
Test cases have been added to the existing TestShiftOverlap test to
cover shift values of 0, 1 and (_W - 1).
Fixes#42838
Change-Id: I75ca0e98f3acfaa6366a26355dcd9dd82499a48b
Reviewed-on: https://go-review.googlesource.com/c/go/+/274442
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
The next CL adds ConstExpr, which is a more memory efficient
representation for constant expressions than Name. However, currently
a bunch of Val helper methods are defined on Name. This CL changes
them into standalone functions that work with any Node.Val
implementation.
There's also an existing standalone function named Int64Val, which
takes a Type argument to specify what type of integer is expected. So
to avoid collisions, this CL renames it to IntVal.
Passes buildall w/ toolstash -cmp.
[git-generate]
cd src/cmd/compile/internal/ir
rf 'mv Int64Val IntVal'
sed -i -E -e 's/\(n \*Name\) (CanInt64|((I|Ui)nt64|Bool|String)Val)\(/\1(n Node/' name.go
cd ../gc
rf '
ex {
import "cmd/compile/internal/ir"
var n ir.Node
n.CanInt64() -> ir.CanInt64(n)
n.Int64Val() -> ir.Int64Val(n)
n.Uint64Val() -> ir.Uint64Val(n)
n.BoolVal() -> ir.BoolVal(n)
n.StringVal() -> ir.StringVal(n)
}
'
cd ../ir
rf '
mv CanInt64 Int64Val Uint64Val BoolVal StringVal val.go
rm Node.CanInt64 Node.Int64Val Node.Uint64Val Node.BoolVal Node.StringVal
'
Change-Id: I003140bda1690d770fd608bdd087e6d4ff00fb1f
Reviewed-on: https://go-review.googlesource.com/c/go/+/275032
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Recreated manually to push below some CLs it depended on.
Change-Id: I1b3316fcdce39cbb33e5cbb471f5cd1cd2efc1f5
Reviewed-on: https://go-review.googlesource.com/c/go/+/274599
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
All instructions in the FMA extension on x86 are VEX prefixed.
VEX prefixed instructions generally require OSXSAVE to be enabled.
The execution of FMA instructions emitted by the Go compiler on amd64
will generate an invalid opcode exception if OSXSAVE is not enabled.
Fixes#41022
Change-Id: I49881630e7195c804110a2bd81b5bec8cac31ba8
Reviewed-on: https://go-review.googlesource.com/c/go/+/274479
Trust: Martin Möhrmann <moehrmann@google.com>
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
assign.go:59:28: error: ‘x’ repeated on left side of :=
assign.go:65:20: error: ‘a’ repeated on left side of :=
method2.go:36:11: error: reference to method ‘val’ in type that is pointer to interface, not interface
method2.go:37:11: error: reference to method ‘val’ in type that is pointer to interface, not interface
Change-Id: I8f385c75a82fae4eacf4618df8f9f65932826494
Reviewed-on: https://go-review.googlesource.com/c/go/+/274447
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>
This CL improves handling of OCONVNOP nodes during ssa generation,
so it is not toolstash safe.
An OCONVNOP wrapper is necessary for the "for" condition of
certain compiled range loops, and the boolean evaluator was
not looking through them properly, generating unnecessary
temporaries. That change saved 8k of the (13 MB) go binary.
The other changes just streamline the handling of OCONVNOP
to be more like what OSTMTEXPR will be like. They have no
effect on output size but do tweak the ssa graph a little, which
causes different register decisions and therefore different output.
Change-Id: I9e1dcd413b60944e21554c3e3f2bdc9adcee7634
Reviewed-on: https://go-review.googlesource.com/c/go/+/274598
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Returning an error about integer overflow is needlessly pedantic.
The meaning of ReadForm(MaxInt64) is easily understood
(accept a lot of data) and can be implemented.
Fixes#40430.
Change-Id: I8a522033dd9a2f9ad31dd2ad82cf08d553736ab9
Reviewed-on: https://go-review.googlesource.com/c/go/+/275112
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Currently, for data moving, we generate an msanread of the source,
followed by an msanwrite of the destination. msanread checks
the source is initialized.
This has a problem: if the source is an aggregate type containing
alignment paddings, the padding bytes may not be thought as
initialized by MSAN. If we copy the aggregate type by value, if
it counts as a read, MSAN reports using uninitialized data. This
CL changes it to use __msan_memmove for data copying, which tells
MSAN to propagate initialized-ness but not check for it.
Caveat: technically __msan_memmove is not a public API of MSAN,
although the C compiler does generate direct calls to it.
Also, when instrumenting a load of a struct, split the
instrumentation to fields, instead of generating an msanread for
the whole struct. This skips padding bytes, which may not be
considered initialized in MSAN.
Fixes#42820.
Change-Id: Id861c8bbfd94cfcccefcc58eaf9e4eb43b4d85c6
Reviewed-on: https://go-review.googlesource.com/c/go/+/270859
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
The check for blank in okAs is redundant with what its callers already
done, so just inline the conversion in callers side instead.
Passes toolstash-check.
Change-Id: I606105e2d2cf8e80214722a13c3101c464d20d82
Reviewed-on: https://go-review.googlesource.com/c/go/+/274793
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>
Noticed the typ field was duplicated, since it is also in miniExpr inside Name.
Also clarified the comments for Func, now that it is actually the ODCLFUNC node.
Change-Id: Ia483a0ad34bb409cd92c43d4ae0a6852f9e4f644
Reviewed-on: https://go-review.googlesource.com/c/go/+/274619
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
For statements like goto that don't need an init, use an
explicit block statement instead of forcing them to have one.
There is also one call to addinit that is being replaced with
a block. That call is the source of much of my confusion
regarding init statements: walkstmt calls addinit on a statement,
whereas all the other uses of addinit are on expressions.
After this CL, they're all expressions.
Passes buildall w/ toolstash -cmp.
Change-Id: Ifdef9d318c236dc1a7567f9e9ef4a6bedd3fe81f
Reviewed-on: https://go-review.googlesource.com/c/go/+/274597
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Not toolstash -cmp safe, so split into its own CL.
Change-Id: I523227514a95e19c9afe17556d754a03164bce76
Reviewed-on: https://go-review.googlesource.com/c/go/+/274595
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
OEMPTY is an empty *statement*, but it confusingly
gets handled as an expression in a few places.
More confusingly, OEMPTY often has an init list,
making it not empty at all. Replace uses and analysis
of OEMPTY with OBLOCK instead.
Passes buildall w/ toolstash -cmp.
Change-Id: I8d4fcef151e4f441fa19b1b96da5272d778131d6
Reviewed-on: https://go-review.googlesource.com/c/go/+/274594
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
It turned out that "go get" was using the network to look up
https://github.com?go-get=1 while resolving github.com/google/go-cmp,
and that is not the fastest page to load.
Stop that lookup by adjusting the path prefixes in the vcs table.
It also turned out that "go get" was using the network to look up
https://rsc.io?go-get=1 while resolving https://rsc.io/nonexist.svn.
That's a bit more defensible maybe, since rsc.io is not a known VCS host.
But for tests we really want to avoid the network entirely, so this CL
adds a special case in repoRootFromVCSPaths that returns a hard error
for plain "rsc.io" instead of doing the web fetch.
To keep us honest in the future, I added two automatically-set env
variables TESTGONETWORK=panic and TESTGOVCS=panic.
These cause the go command to panic rather than make a network request
or invoke a VCS command.
go test -short cmd/go now passes with these checks.
This reduced the time spent in go test -short cmd/go on my
Google workstation from 154s to 30s. (Yay network firewalls.)
Change-Id: I49207fca7f901fa011765fb984dc9cec8b691f11
Reviewed-on: https://go-review.googlesource.com/c/go/+/274441
Trust: Russ Cox <rsc@golang.org>
Trust: Jay Conrod <jayconrod@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
io/ioutil was a poorly defined collection of helpers.
Proposal #40025 moved out the generic I/O helpers to io.
This CL for proposal #42026 moves the OS-specific helpers to os,
making the entire io/ioutil package deprecated.
For #42026.
Change-Id: I018bcb2115ef2ff1bc7ca36a9247eda429af21ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/266364
Trust: Russ Cox <rsc@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
This if follow up of CL 274332.
Updates #9399.
Change-Id: Ic6dd534dc18227a799cbb9577979f2285596b825
Reviewed-on: https://go-review.googlesource.com/c/go/+/274393
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: Ian Lance Taylor <iant@golang.org>
Unbuffered channels passed into signal.Notify can be lost
as the docs for signal.Notify caution with:
Package signal will not block sending to c: the caller must ensure
that c has sufficient buffer space to keep up with the expected signal
rate. For a channel used for notification of just one signal value,
a buffer of size 1 is sufficient.
Found by a static analyzer from Orijtech, Inc. called "sigchanyzer", but
it'll be donated to the Go project soon.
Updates #9399.
Change-Id: Ia0690e447582da028694ed65ace7b97961997b84
Reviewed-on: https://go-review.googlesource.com/c/go/+/274332
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>
Now that filepath.WalkDir is available, it is more efficient
and should be used in place of filepath.Walk.
Update the tree to reflect best practices.
As usual, the code compiled with Go 1.4 during bootstrap is excluded.
(In this CL, that's only cmd/dist.)
For #42027.
Change-Id: Ib0f7b1e43e50b789052f9835a63ced701d8c411c
Reviewed-on: https://go-review.googlesource.com/c/go/+/267719
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Pseudo branch instructions BGT, BGTU, BLE, and BLEU implemented In
CL 226397 were translated inconsistently compared to other ones due
to the inversion of registers. For instance, while "BLT a, b" generates
"jump if a < b", "BLE a, b" generates "jump if b <= a."
This CL fixes the translation in the assembler and the tests.
Change-Id: Ia757be73e848734ca5b3a790e081f7c4f98c30f2
Reviewed-on: https://go-review.googlesource.com/c/go/+/271911
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Joel Sing <joel@sing.id.au>
Run-TryBot: Joel Sing <joel@sing.id.au>
This allows directly creating an ONONAME, which is a primordial Name
before having its Op initialized. Then after an Op is assigned, we
never allow it to be reassigned.
Passes buildall w/ toolstash -cmp.
Change-Id: Ibc2f413dc68c0af6a96abfe653c25ce31b184287
Reviewed-on: https://go-review.googlesource.com/c/go/+/274620
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Allows getting rid of the SetPkg method and also addresses a
long-standing TODO in the exporter. Suggested by rsc@.
Passes buildall w/ toolstash -cmp.
Change-Id: Ib294f75f1350572efb2e0d993d49efef884de3d4
Reviewed-on: https://go-review.googlesource.com/c/go/+/274440
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
It's noisy and not doing any harm, and we still have an entire release
cycle to revisit and address the issue properly.
Updates #42938
Change-Id: I1de5cfb495a8148c9c08b215deba38f2617fb467
Reviewed-on: https://go-review.googlesource.com/c/go/+/274732
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
This CL adds the remaining constructors needed to abstract away
construction of Types, and updates the compiler to use them
throughout. There's now just a couple uses within test cases to
remove.
While at it, I also replace the Func.Outnamed field with a simple
helper function, which reduces the size of function types somewhat.
Passes toolstash/buildall.
Change-Id: If1aa1095c98ae34b00380d0b3531bd63c10ce885
Reviewed-on: https://go-review.googlesource.com/c/go/+/274713
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
There's no need for the bucket type to be precise. The compiler
doesn't actually generate code that references these fields; it just
needs it for size and GC bitmap calculations.
However, changing the type field does alter the runtime type
descriptor and relocations emitted by the compiler, so this change
isn't safe for toolstash.
Change-Id: Icf79d6c4326515889b13435a575d618e3bbfbcd7
Reviewed-on: https://go-review.googlesource.com/c/go/+/274712
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
The gofrontend code doesn't distinguish semicolon and newline,
and it doesn't have special treatment for EOF.
syntax/semi6.go:9:47: error: unexpected semicolon or newline in type declaration
syntax/semi6.go:11:62: error: unexpected semicolon or newline in type declaration
Change-Id: I9996b59a4fc78ad1935e779f354ddf75c0fb44e0
Reviewed-on: https://go-review.googlesource.com/c/go/+/274692
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Apparently, the darwin kernel may cache the code signature at
mmap. When we mmap the output buffer, it doesn't have a code
signature (as we haven't generated one). Invalidate the kernel
cache after writing the file.
See https://github.com/golang/go/issues/42684#issuecomment-731704900
for more information.
Updates #38485.
Fixes#42684.
Change-Id: Iac2ef756ca1454c856944423e5040b8e17a6b420
Reviewed-on: https://go-review.googlesource.com/c/go/+/272258
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
As the code signature contains hashes of the entire file (except
the signature itself), rewriting buildid will invalidate the
signature. This CL makes it regenerate the signature when
rewriting the buildid. It only does it when the file already has
a code signature, with proper size (darwin/arm64 binaries
generated by the Go linker should have).
Updates #38485, #42684.
Change-Id: I082d9e5808b0ee6a35f9c362d7262aadd9113c81
Reviewed-on: https://go-review.googlesource.com/c/go/+/272257
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
This CL lets the linker code-sign output binaries on
darwin/arm64, as the kernel requires binaries must be signed in
order to run.
This signature will likely be invalidated when we stamp the
buildid after linking. We still do it in the linker, for
- plain "go tool link" works.
- the linker generates the LC_CODE_SIGNATURE load command with
the right size and offset, so we don't need to update it when
stamping the buildid.
Updates #38485, #42684.
Change-Id: Ia306328906d73217221ba31093fe61a935a46122
Reviewed-on: https://go-review.googlesource.com/c/go/+/272256
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
These replacement rules assume that TST and TEQ set V. But TST and
TEQ do not set V. This is a problem because instructions like LT are
actually checking for N!=V. But with TST and TEQ not setting V, LT
doesn't do anything meaningful. It's possible to construct trivial
miscompilations from this, such as:
package main
var x = [4]int32{-0x7fffffff, 0x7fffffff, 2, 4}
func main() {
if x[0] > x[1] {
panic("fail 1")
}
if x[2]&x[3] < 0 {
panic("fail 2") // Fails here
}
}
That first comparison sets V, via the CMP that subtracts the values
causing the overflow. Then the second comparison operation thinks that
it uses the result of TST, when it actually uses the V from CMP.
Before this fix:
TST R0, R1
BLT loc_6C164
After this fix:
TST R0, R1
BMI loc_6C164
The BMI instruction checks the N flag, which TST sets. This commit
fixes the issue by using [LG][TE]noov instead of vanilla [LG][TE], and
also adds a test case for the direct issue.
Fixes#42876.
Change-Id: I13c62c88d18574247ad002b671b38d2d0b0fc6fa
Reviewed-on: https://go-review.googlesource.com/c/go/+/274026
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
