Logical instructions can have RSP as its destination. Support it.
Note that the two-operand form, like "AND $1, RSP", which is
equivalent to the three-operand form "AND $1, RSP, RSP", is
invalid, because the source register is not allowed to be RSP.
Also note that instructions that set the conditional flags, like
ANDS, cannot target RSP. Because of this, we split out the optab
entries of AND et al. and ANDS et al.
Merge the optab entries of BIC et al. to AND et al., because they
are same.
Fixes#24332.
Change-Id: I3584d6f2e7cea98a659a1ed9fdf67c353e090637
Reviewed-on: https://go-review.googlesource.com/100217
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
The compiler now emits TBZ like instructions, but the assembler's
too-far-branch patch code didn't include that case. Add it.
Fixes#23889.
Change-Id: Ib75f9250c660b9fb652835fbc83263a5d5073dc5
Reviewed-on: https://go-review.googlesource.com/94902
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This is the "3rd bug" that caused compilations to sometimes
produce different results when dwarf location lists were
enabled.
A loop had not been properly rewritten in an earlier
optimization CL, and it accessed uninitialized data,
which was deterministically perhaps wrong when single
threaded, but variably wrong when multithreaded.
Change-Id: Ib3da538762fdf7d5e4407106f2434f3b14a1d7ea
Reviewed-on: https://go-review.googlesource.com/99935
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
There are few uses for the majority of the API in go/token for the
average user. The exception to this is getting the filename, line, and
column information from a token.Pos (reported and absolute. This is
straightforward but figuring out how to do it requires combing through
a lot of documentation. This example makes it more easily discoverable.
Updates #24352.
Change-Id: I0a45da6173b3dabebf42484bbbed30d9e5e20e01
Reviewed-on: https://go-review.googlesource.com/100058
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Every time I poke at #14921, the g.waitreason string
pointer writes show up.
They're not particularly important performance-wise,
but it'd be nice to clear the noise away.
And it does open up a few extra bytes in the g struct
for some future use.
Change-Id: I7ffbd52fbc2a286931a2218038fda52ed6473cc9
Reviewed-on: https://go-review.googlesource.com/99078
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Generated by running gofmt -s on the files in question.
Change-Id: If6578b150e1bfced8657196d2af01f5d36879f93
Reviewed-on: https://go-review.googlesource.com/100135
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Currently, order desugars map assignment operations like
m[k] op= r
into
m[k] = m[k] op r
which in turn is transformed during walk into:
tmp := *mapaccess(m, k)
tmp = tmp op r
*mapassign(m, k) = tmp
However, this is suboptimal, as we could instead produce just:
*mapassign(m, k) op= r
One complication though is if "r == 0", then "m[k] /= r" and "m[k] %=
r" will panic, and they need to do so *before* calling mapassign,
otherwise we may insert a new zero-value element into the map.
It would be spec compliant to just emit the "r != 0" check before
calling mapassign (see #23735), but currently these checks aren't
generated until SSA construction. For now, it's simpler to continue
desugaring /= and %= into two map indexing operations.
Fixes#23661.
Change-Id: I46e3739d9adef10e92b46fdd78b88d5aabe68952
Reviewed-on: https://go-review.googlesource.com/91557
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
cmd/asm now supports three-operand form of IMUL,
so instead of using IMUL with resultInArg0, emit IMUL3 instruction.
This results in less redundant MOVs where SSA assigns
different registers to input[0] and dst arguments.
Note: these have exactly the same encoding when reg0=reg1:
IMUL3x $const, reg0, reg1
IMULx $const, reg
Two-operand IMULx is like a crippled IMUL3x, with dst fixed to input[0].
This is why we don't bother to generate IMULx for the case where
dst is the same as input[0].
Change-Id: I4becda475b3dffdd07b6fdf1c75bacc82af654e4
Reviewed-on: https://go-review.googlesource.com/99656
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Keith Randall <khr@golang.org>
The current code assigns vector register arrangement a wrong value
when the arrangement specifier is S2, which causes the incorrect
assembly.
The patch fixes the issue and adds the test cases.
Fixes#24249
Change-Id: I9736df1279494003d0b178da1af9cee9cd85ce21
Reviewed-on: https://go-review.googlesource.com/98555
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Remove old tests from asm_test.
Change-Id: Ib408ec7faa60068bddecf709b93ce308e0ef665a
Reviewed-on: https://go-review.googlesource.com/100075
Reviewed-by: Alberto Donizetti <alb.donizetti@gmail.com>
The tool has gotten better over time, so re-generating the files brings
some advantages like fewer objects, dropping the use of fmt, and
dropping unnecessary bounds checks.
While at it, add the missing go:generate line for obj.AddrType.
Change-Id: I120c9795ee8faddf5961ff0384b9dcaf58d831ff
Reviewed-on: https://go-review.googlesource.com/100015
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
I found files to change with this command:
git grep 'DO NOT EDIT' | grep -v 'Code generated .* DO NOT'
There are more files that match that grep, but I do not intend on fixing
them.
Change-Id: I4b474f1c29ca3135560d414785b0dbe0d1a4e52c
GitHub-Last-Rev: 65804b0263
GitHub-Pull-Request: golang/go#24334
Reviewed-on: https://go-review.googlesource.com/99955
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The "updates" lines, such as RUN, do not contain a colon. However,
test2json looked for one anyway, meaning that it would be thrown off if
it encountered a line like:
=== RUN TestWithColons/[::1]
In that case, it must not use the first colon it encounters to separate
the action from the test name.
Fixes#23920.
Change-Id: I82eff23e24b83dae183c0cf9f85fc5f409f51c25
Reviewed-on: https://go-review.googlesource.com/98445
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Added examples for use of Mode, Whitespace, and IsIdentRune properties.
Fixes#23768
Change-Id: I2528e14fde63a4476f3c25510bf0c5b73f38ba5d
Reviewed-on: https://go-review.googlesource.com/93199
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Before DWARF location lists can be turned on, 3 bugs need
fixing.
This CL addresses two -- lack of register definitions for
various architectures, and bugs on 32-bit platforms.
The third bug comes later.
Passes
GO_GCFLAGS=-dwarflocationlists ./run.bash -no-rebuild
(-no-rebuild because the map dependence causes trouble)
Change-Id: I4223b48ade84763e4b048e4aeb81149f082c7bc7
Reviewed-on: https://go-review.googlesource.com/99255
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This change updates go/scanner to recognize the extended line
directives that are now also handled by cmd/compile:
//line filename:line
//line filename:line:column
/*line filename:line*/
/*line filename:line:column*/
As before, //-style line directives must start in column 1.
/*-style line directives may be placed anywhere in the code.
In both cases, the specified position applies to the character
immediately following the comment; for line comments that is
the first character on the next line (after the newline of the
comment).
The go/token API is extended by a new method
File.AddLineColumnInfo(offset int, filename string, line, column int)
which extends the existing
File.AddLineInfo(offset int, filename string, line int)
by adding a column parameter.
Adjusted token.Position computation is changed to take into account
column information if provided via a line directive: A (line-directive)
relative position will have a non-zero column iff the line directive
specified a column; if the position is on the same line as the line
directive, the column is relative to the specified column (otherwise
it is relative to the line beginning). See also #24183.
Finally, Position.String() has been adjusted to not print a column
value if the column is unknown (== 0).
Fixes#24143.
Change-Id: I5518c825ad94443365c049a95677407b46ba55a1
Reviewed-on: https://go-review.googlesource.com/97795
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
There are two less optimized SSA rules in my previous CL
https://go-review.googlesource.com/c/go/+/95075 .
This CL fixes that issue and a test case gets about 10%
performance improvement.
name old time/op new time/op delta
MNEG-4 263µs ± 3% 235µs ± 3% -10.53% (p=0.000 n=20+20)
(https://github.com/benshi001/ugo1/blob/master/mneg_7_test.go)
Change-Id: I30087097e281dd9d9d1c870d32e13b4ef4a96ad3
Reviewed-on: https://go-review.googlesource.com/99495
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The implementation of runtime.abort on arm64 currently branches to
address 0, which results in a signal from PC 0, rather than from
runtime.abort, so the runtime fails to recognize it as an abort.
Fix runtime.abort on arm64 to read from address 0 like what other
architectures do and recognize this in the signal handler.
Should fix the linux/arm64 build.
Change-Id: I960ab630daaeadc9190287604d4d8337b1ea3853
Reviewed-on: https://go-review.googlesource.com/99895
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Add helper methods that validate n.Op and convert to/from the
appropriate type.
Notably, there was a lot of code in walk.go that thought setting
Etype=1 on an OADDR node affected escape analysis.
Passes toolstash-check.
TBR=marvin
Change-Id: Ieae7c67225c1459c9719f9e6a748a25b975cf758
Reviewed-on: https://go-review.googlesource.com/99535
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
bytes.IndexByte is heavily optimized. Use it in findnull.
This is second attempt, similar to CL97523.
In this version we never call IndexByte on region of memory,
that crosses page boundary. A bit slower than CL97523,
but still fast:
name old time/op new time/op delta
GoString-6 164ns ± 2% 118ns ± 0% -28.00% (p=0.000 n=10+6)
findnull is also used in gostringnocopy,
which is used in many hot spots in the runtime.
Fixes#23830
Change-Id: Id843dd4f65a34309d92bdd8df229e484d26b0cb2
Reviewed-on: https://go-review.googlesource.com/98015
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This lets SIGPROF signals get a useful traceback.
Without it we just see sysvicallN calling asmcgocall.
Updates #24142
Change-Id: I5dfe3add51f0c3a4cb1c98acb7738be6396214bc
Reviewed-on: https://go-review.googlesource.com/99617
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
This change adds a README file inside the test/codegen directory,
explaining how to run the codegen tests and the syntax of the regexps
comments used to match assembly instructions.
Change-Id: Ica4eb3ffa9c6975371538cc8ae0ac3c1a3a03baf
Reviewed-on: https://go-review.googlesource.com/99156
Reviewed-by: Keith Randall <khr@golang.org>
The index 248 results in the decoder calling reflect.MakeMapWithSize
with a size of 14754407682 - just under 15GB - which ends up in a
runtime out of memory panic after some recent runtime changes on
machines with 8GB of memory.
Until that is fixed in either runtime or gob, skip the troublesome
index.
Updates #24308.
Change-Id: Ia450217271c983e7386ba2f3f88c9ba50aa346f4
Reviewed-on: https://go-review.googlesource.com/99655
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Borrowed from cmd/compile, TestSizeof ensures
that the size of important types doesn't change unexpectedly.
It also helps reviewers see the impact of intended changes.
Change-Id: If57955f0c3e66054de3f40c6bba585b88694c7be
Reviewed-on: https://go-review.googlesource.com/99837
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This is optimization is only for IPv4. It allocates a result buffer and
writes the IPv4 octets as dotted decimal into it before converting
it to a string just once, reducing allocations.
Benchmark shows performance improvement:
name old time/op new time/op delta
IPString/IPv4-8 284ns ± 4% 144ns ± 6% -49.35% (p=0.000 n=19+17)
IPString/IPv6-8 1.34µs ± 5% 1.14µs ± 5% -14.37% (p=0.000 n=19+20)
name old alloc/op new alloc/op delta
IPString/IPv4-8 24.0B ± 0% 16.0B ± 0% -33.33% (p=0.000 n=20+20)
IPString/IPv6-8 232B ± 0% 224B ± 0% -3.45% (p=0.000 n=20+20)
name old allocs/op new allocs/op delta
IPString/IPv4-8 3.00 ± 0% 2.00 ± 0% -33.33% (p=0.000 n=20+20)
IPString/IPv6-8 12.0 ± 0% 11.0 ± 0% -8.33% (p=0.000 n=20+20)
Fixes#24306
Change-Id: I4e2d30d364e78183d55a42907d277744494b6df3
Reviewed-on: https://go-review.googlesource.com/99395
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
And delete them from asm_go.
Change-Id: I0057cbd90ca55fa51c596e32406e190f3866f93e
Reviewed-on: https://go-review.googlesource.com/99815
Reviewed-by: Keith Randall <khr@golang.org>
hwcap is set in archauxv, setup_auxv no longer exists.
Change-Id: I0fc9393e0c1c45192e0eff4715e9bdd69fab2653
Reviewed-on: https://go-review.googlesource.com/99779
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Directly use rc.Close instead of wrapping it with a closure.
Change-Id: I3dc1c21ccbfe031c230b035126d5ea3bc62055c3
Reviewed-on: https://go-review.googlesource.com/99716
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Fixes#23971
Change-Id: I073f278cc058aa15a23c0ea06292c02d50a3df21
Reviewed-on: https://go-review.googlesource.com/95582
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Only RotateLeft{64,32} were tested, and just for ppc64. This CL adds
tests for RotateLeft{64,32,16,8} on arm64 and amd64/386, for the cases
where the calls are actually instrinsified.
RotateLeft tests (the last ones for math/bits functions) are deleted
from asm_test.
This CL also adds a space between the "//" and the arch name in the
comments, to uniform this file to the style used in all the other
files.
Change-Id: Ifc2a27261d70bcc294b4ec64490d8367f62d2b89
Reviewed-on: https://go-review.googlesource.com/99596
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Also grey out instants that represent events occurred outside the
task's span. Furthermore, if the unrelated instants represent user
annotation events but not for the task of the interest, skip rendering
completely.
This helps users to focus on the task-related events better.
UI screen shot:
https://gist.github.com/hyangah/1df5d2c8f429fd933c481e9636b89b55#file-golang-org_cl_99035
Change-Id: I2b5aef41584c827f8c1e915d0d8e5c95fe2b4b65
Reviewed-on: https://go-review.googlesource.com/99035
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
Strangely enough, the existing implementation used adjusted (by line
directives) source positions to determine layout and thus required
position corrections when printing a line directive.
Instead, just use the unadjusted, absolute source positions and then
printing a line directive doesn't require any adjustments, only some
care to make sure it remains in column 1 as before.
The new code doesn't need to parse line directives anymore and simply
ensures that comments with the //line prefix and starting in column 1
remain in that position. That is a slight change from the old behavior
(which ignored incorrect line directives, e.g. because they had an
invalid line number) but unlikely to show up in real code.
This is prep work for handling of line directives that also specify
columns (which now won't require much special handling anymore).
For #24143.
Change-Id: I07eb2e1b35b37337e632e3dbf5b70c783c615f8a
Reviewed-on: https://go-review.googlesource.com/99621
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
'"' has special semantic meaning that conflicts with using it as Comma.
Change-Id: Ife25ba43ca25dba2ea184c1bb7579a230d376059
Reviewed-on: https://go-review.googlesource.com/99696
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
It's a bit mysterious that _defer.sp is a uintptr that gets
stack-adjusted explicitly while _panic.argp is an unsafe.Pointer that
doesn't, but turns out to be critically important when a deferred
function grows the stack before doing a recover.
Add a comment explaining that this works because _panic values live on
the stack. Enforce this by marking _panic go:notinheap.
Change-Id: I9ca49e84ee1f86d881552c55dccd0662b530836b
Reviewed-on: https://go-review.googlesource.com/99735
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
On all non-x86 arches, runtime.abort simply reads from nil.
Unfortunately, if this happens on a user stack, the signal handler
will dutifully turn this into a panicmem, which lets user defers run
and which user code can even recover from.
To fix this, add an explicit check to the signal handler that turns
faults in abort into hard crashes directly in the signal handler. This
has the added benefit of giving a register dump at the abort point.
Change-Id: If26a7f13790745ee3867db7f53b72d8281176d70
Reviewed-on: https://go-review.googlesource.com/93661
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Everything except for amd64, amd64p32, and 386 currently defines and
uses an abort function. This CL makes these match. The next CL will
recognize the abort function to make this more useful.
Change-Id: I7c155871ea48919a9220417df0630005b444f488
Reviewed-on: https://go-review.googlesource.com/93660
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Currently, throw may grow the stack, which means whenever we call it
from a context where it's not safe to grow the stack, we first have to
switch to the system stack. This is pretty easy to get wrong.
Fix this by making throw switch to the system stack so it doesn't grow
the stack and is hence safe to call without a system stack switch at
the call site.
The only thing this complicates is badsystemstack itself, which would
now go into an infinite loop before printing anything (previously it
would also go into an infinite loop, but would at least print the
error first). Fix this by making badsystemstack do a direct write and
then crash hard.
Change-Id: Ic5b4a610df265e47962dcfa341cabac03c31c049
Reviewed-on: https://go-review.googlesource.com/93659
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Currently parts of unrecoverable panic handling (notably, printing
panic messages) can happen on the user stack. This may grow the stack,
which is generally fine, but if we're handling a runtime panic, it's
better to do as little as possible in case the runtime is in an
inconsistent state.
Hence, this commit rearranges the handling of unrecoverable panics so
that it's done entirely on the system stack.
This is mostly a matter of shuffling code a bit so everything can move
into a systemstack block. The one slight subtlety is in the "panic
during panic" case, where we now depend on startpanic_m's caller to
print the stack rather than startpanic_m itself. To make this work,
startpanic_m now returns a boolean indicating that the caller should
avoid trying to print any panic messages and get right to the stack
trace. Since the caller is already in a position to do this, this
actually simplifies things a little.
Change-Id: Id72febe8c0a9fb31d9369b600a1816d65a49bfed
Reviewed-on: https://go-review.googlesource.com/93658
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
The previous CL introduced isConstDelta. Use it to simplify the
OpSlicemask optimization in the prove pass. This passes toolstash
-cmp.
Change-Id: If2aa762db4cdc0cd1c581a536340530a9831081b
Reviewed-on: https://go-review.googlesource.com/87481
Reviewed-by: Keith Randall <khr@golang.org>
This adds four new deductions to the prove pass, all related to adding
or subtracting one from a value. This is the first hint of actual
arithmetic relations in the prove pass.
The most effective of these is
x-1 >= w && x > min ⇒ x > w
This helps eliminate bounds checks in code like
if x > 0 {
// do something with s[x-1]
}
Altogether, these deductions prove an additional 260 branches in std
and cmd. Furthermore, they will let us eliminate some tricky
compiler-inserted panics in the runtime that are interfering with
static analysis.
Fixes#23354.
Change-Id: I7088223e0e0cd6ff062a75c127eb4bb60e6dce02
Reviewed-on: https://go-review.googlesource.com/87480
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>
This adds a few simple deductions to the prove pass' fact table to
derive unsigned concrete limits from signed concrete limits where
possible.
This tweak lets the pass prove 70 additional branch conditions in std
and cmd.
This is based on a comment from the recently-deleted factsTable.get:
"// TODO: also use signed data if lim.min >= 0".
Change-Id: Ib4340249e7733070f004a0aa31254adf5df8a392
Reviewed-on: https://go-review.googlesource.com/87479
Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>
Reviewed-by: Keith Randall <khr@golang.org>
Currently the prove pass uses implication queries. For each block, it
collects the set of branch conditions leading to that block, and
queries this fact table for whether any of these facts imply the
block's own branch condition (or its inverse). This works remarkably
well considering it doesn't do any deduction on these facts, but it
has various downsides:
1. It requires an implementation both of adding facts to the table and
determining implications. These are very nearly duals of each
other, but require separate implementations. Likewise, the process
of asserting facts of dominating branch conditions is very nearly
the dual of the process of querying implied branch conditions.
2. It leads to less effective use of derived facts. For example, the
prove pass currently derives facts about the relations between len
and cap, but can't make use of these unless a branch condition is
in the exact form of a derived fact. If one of these derived facts
contradicts another fact, it won't notice or make use of this.
This CL changes the approach of the prove pass to instead use
*contradiction* instead of implication. Rather than ever querying a
branch condition, it simply adds branch conditions to the fact table.
If this leads to a contradiction (specifically, it makes the fact set
unsatisfiable), that branch is impossible and can be cut. As a result,
1. We can eliminate the code for determining implications
(factsTable.get disappears entirely). Also, there is now a single
implementation of visiting and asserting branch conditions, since
we don't have to flip them around to treat them as facts in one
place and queries in another.
2. Derived facts can be used effectively. It doesn't matter *why* the
fact table is unsatisfiable; a contradiction in any of the facts is
enough.
3. As an added benefit, it's now quite easy to avoid traversing beyond
provably-unreachable blocks. In contrast, the current
implementation always visits all blocks.
The prove pass already has nearly all of the mechanism necessary to
compute unsatisfiability, which means this both simplifies the code
and makes it more powerful.
The only complication is that the current implication procedure has a
hack for dealing with the 0 <= Args[0] condition of OpIsInBounds and
OpIsSliceInBounds. We replace this with asserting the appropriate fact
when we process one of these conditions. This seems much cleaner
anyway, and works because we can now take advantage of derived facts.
This has no measurable effect on compiler performance.
Effectiveness:
There is exactly one condition in all of std and cmd that this fails
to prove that the old implementation could: (int64(^uint(0)>>1) < x)
in encoding/gob. This can never be true because x is an int, and it's
basically coincidence that the old code gets this. (For example, it
fails to prove the similar (x < ^int64(^uint(0)>>1)) condition that
immediately precedes it, and even though the conditions are logically
unrelated, it wouldn't get the second one if it hadn't first processed
the first!)
It does, however, prove a few dozen additional branches. These come
from facts that are added to the fact table about the relations
between len and cap. These were almost never queried directly before,
but could lead to contradictions, which the unsat-based approach is
able to use.
There are exactly two branches in std and cmd that this implementation
proves in the *other* direction. This sounds scary, but is okay
because both occur in already-unreachable blocks, so it doesn't matter
what we chose. Because the fact table logic is sound but incomplete,
it fails to prove that the block isn't reachable, even though it is
able to prove that both outgoing branches are impossible. We could
turn these blocks into BlockExit blocks, but it doesn't seem worth the
trouble of the extra proof effort for something that happens twice in
all of std and cmd.
Tests:
This CL updates test/prove.go to change the expected messages because
it can no longer give a "reason" why it proved or disproved a
condition. It also adds a new test of a branch it couldn't prove
before.
It mostly guts test/sliceopt.go, removing everything related to slice
bounds optimizations and moving a few relevant tests to test/prove.go.
Much of this test is actually unreachable. The new prove pass figures
this out and doesn't try to prove anything about the unreachable
parts. The output on the unreachable parts is already suspect because
anything can be proved at that point, so it's really just a regression
test for an algorithm the compiler no longer uses.
This is a step toward fixing #23354. That issue is quite easy to fix
once we can use derived facts effectively.
Change-Id: Ia48a1b9ee081310579fe474e4a61857424ff8ce8
Reviewed-on: https://go-review.googlesource.com/87478
Reviewed-by: Keith Randall <khr@golang.org>
