David Chase [Sun, 28 Feb 2016 16:15:22 +0000 (11:15 -0500)]
[dev.ssa] cmd/compile: use 32-bit load to read writebarrier
Avoid targeting a partial register with load;
ensure source of load (writebarrier) is aligned.
Better yet would be "CMPB $1,writebarrier" but that requires
wrestling with flagalloc (mem operand complicates moving
instruction around).
Didn't see a change in time for
benchcmd -n 10 Build go build net/http
Verified that we clean the code up properly:
0x20a8 <main.main+104>: mov 0xc30a2(%rip),%eax
# 0xc5150 <runtime.writeBarrier>
0x20ae <main.main+110>: test %al,%al
Change-Id: Id5fb8c260eaec27bd727cb0ae1476c60343b0986
Reviewed-on: https://go-review.googlesource.com/19998 Reviewed-by: Keith Randall <khr@golang.org>
* It does very simple bounds checking elimination. E.g.
removes the second check in for i := range a { a[i]++; a[i++]; }
* Improves on the following redundant expression:
return a6 || (a6 || (a6 || a4)) || (a6 || (a4 || a6 || (false || a6)))
* Linear in the number of block edges.
I patched in CL 12960 that does bounds, nil and constant propagation
to make sure this CL is not just redundant. Size of pkg/tool/linux_amd64/*
(excluding compile which is affected by this change):
Todd Neal [Sat, 27 Feb 2016 14:04:48 +0000 (08:04 -0600)]
[dev.ssa] cmd/compile: add max arg length to opcodes
Add the max arg length to opcodes and use it in zcse. Doesn't affect
speed, but allows better checking in checkFunc and removes the need
to keep a list of zero arg opcodes up to date.
Keith Randall [Fri, 26 Feb 2016 19:01:14 +0000 (11:01 -0800)]
[dev.ssa] test: remove extra tests from non-SSA builds
non-SSA backends are all over the map as to whether nil checks
get removed or not. amd64, 386, 386/387, arm are all subtly different.
Remove these extra checks for now, they are in nilptr3_ssa.go so they
won't get lost.
Keith Randall [Thu, 25 Feb 2016 19:40:51 +0000 (11:40 -0800)]
[dev.ssa] cmd/compile: shrink stack guard
Our stack frame sizes look pretty good now. Lower the stack
guard from 1024 to 720.
Tip is currently using 720.
We could go lower (to 640 at least) except PPC doesn't like that.
Change-Id: Ie5f96c0e822435638223f1e8a2bd1a1eed68e6aa
Reviewed-on: https://go-review.googlesource.com/19922
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
David Chase [Thu, 25 Feb 2016 18:10:51 +0000 (13:10 -0500)]
[dev.ssa] cmd/compile: enhance command line option processing for SSA
The -d compiler flag can also specify ssa phase and flag,
for example -d=ssa/generic_cse/time,ssa/generic_cse/stats
Spaces in the phase names can be specified with an
underscore. Flags currently parsed (not necessarily
recognized by the phases yet) are:
on, off, mem, time, debug, stats, and test
On, off and time are handled in the harness,
debug, stats, and test are interpreted by the phase itself.
The pass is now attached to the Func being compiled, and a
new method logStats(key, ...value) on *Func to encourage a
semi-standardized format for that output. Output fields
are separated by tabs to ease digestion by awk and
spreadsheets. For example,
if f.pass.stats > 0 {
f.logStat("CSE REWRITES", rewrites)
}
Keith Randall [Wed, 24 Feb 2016 18:29:27 +0000 (10:29 -0800)]
[dev.ssa] cmd/compile: fix @ rewrite rules
The @ directive used to read the target block after some value
structure had already changed. I don't think it was ever really
a bug, but it's confusing.
It might fail like this:
(Foo x y) -> @v.Args[0].Block (Bar y (Baz ...))
v.Op = Bar
v.Args[0] = y
v.Args[1] = v.Args[0].Block.NewValue(Baz, ...)
That new value is allocated in the block of y, not the
block of x.
Anyway, read the destination block first so this
potential bug can't happen.
David Chase [Wed, 24 Feb 2016 02:09:39 +0000 (21:09 -0500)]
[dev.ssa] cmd/compile: reduce line number churn in generated code
In regalloc, make LoadReg instructions use the line number
of their *use*, not their *source*. This reduces the
tendency of debugger stepping to "jump around" the program.
Change-Id: I59e2eeac4dca9168d8af3a93effbc5bdacac2881
Reviewed-on: https://go-review.googlesource.com/19836
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Todd Neal [Tue, 23 Feb 2016 23:52:17 +0000 (17:52 -0600)]
[dev.ssa] cmd/compile: speed up cse
Construct better initial partitions by recursively comparing values and
their arguments. This saves one second on compile of arithConst_ssa.go
(4.3s to 3.3s) and shows a 3-5% increase with compilebench.
Alexandru Moșoi [Tue, 23 Feb 2016 20:48:33 +0000 (21:48 +0100)]
[dev.ssa] cmd/compile/internal/ssa: simplify convert in more cases
Saves about 2k for binaries in pkg/tool/linux_amd64.
Also useful when opt runs after cse (as in 12960) which reorders
arguments for commutative operations such as Add64.
Change-Id: I49ad53afa53db9736bd35c425f4fb35fb511fd63
Reviewed-on: https://go-review.googlesource.com/19827
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com> Reviewed-by: Keith Randall <khr@golang.org>
Paul Marks [Tue, 9 Feb 2016 04:25:38 +0000 (20:25 -0800)]
net: use dialTCP cancelation for DualStack dialing.
The previous Happy Eyeballs implementation would intentionally leak
connections, because dialTCP could not be reliably terminated upon
losing the race.
Now that dialTCP supports cancelation (plan9 excluded), dialParallel can
wait for responses from both the primary and fallback racers, strictly
before returning control to the caller.
In dial_test.go, we no longer need Sleep to avoid leaks.
Also, fix a typo in the Benchmark IPv4 address.
Robert Griesemer [Tue, 23 Feb 2016 18:42:06 +0000 (10:42 -0800)]
spec: fix EBNF for slice syntax
The () parentheses grouped wrongly. Removed them completely in
favor of separate 2- and 3-index slice alternatives which is
clearer.
Fixes #14477.
Change-Id: I0b7521ac912130d9ea8740b8793b3b88e2609418
Reviewed-on: https://go-review.googlesource.com/19853 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alexandru Moșoi [Mon, 22 Feb 2016 16:14:53 +0000 (17:14 +0100)]
[dev.ssa] cmd/compile/internal/ssa: eliminate phis during deadcode removal
While investigating the differences between 19710 (remove
tautological controls) and 12960 (bounds and nil propagation)
I observed that part of the wins of 19710 come from missed
opportunities for deadcode elimination due to phis.
See for example runtime.stackcacherelease. 19710 happens much
later than 12960 and has more chances to eliminate bounds.
Robert Griesemer [Mon, 22 Feb 2016 19:53:20 +0000 (11:53 -0800)]
cmd/compile: move Io state into lexer and remove Io type
Pass lexer around so state is accessible and dependency is explicit.
In the process remove EOF -> '\n' conversion that has to be corrected
for when reporting errors.
Change-Id: If95564b70e7484dedc1f5348e585cd19acbc1243
Reviewed-on: https://go-review.googlesource.com/19819
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Brady Sullivan [Mon, 22 Feb 2016 23:19:18 +0000 (15:19 -0800)]
crypto/tls: Improve ambiguous comment in cipher_suites.go
A comment existed referencing RC4 coming before AES because of it's
vulnerability to the Lucky 13 attack. This clarifies that the Lucky 13 attack
only effects AES-CBC, and not AES-GCM.
Rob Pike [Fri, 19 Feb 2016 05:36:03 +0000 (16:36 +1100)]
cmd/doc: handle embedded fields properly
The structure of the code meant that an embedded field was never
checked for export status. We need to check the name of the type,
which is either of type T or type *T, and T might be unexported.
Fixes #14356.
Change-Id: I56f468e9b8ae67e9ed7509ed0b91d860507baed2
Reviewed-on: https://go-review.googlesource.com/19701 Reviewed-by: Robert Griesemer <gri@golang.org>
Caio Marcelo de Oliveira Filho [Sun, 21 Feb 2016 02:23:01 +0000 (23:23 -0300)]
cmd/cover: don't overskip children nodes when adding counters
When visiting the AST to add counters, there are special cases in which
the code calls cuts the walking short by returning nil. In some cases
certain nodes are ignored, e.g. Init and Cond inside IfStmt.
The fix is to explicitly walk all the children nodes (not only
Body and Else) when cutting the current walk. Similar approach
was taken with SwitchStmt and TypeSwitchStmt.
While the existing test code doesn't handle different counters in the
same line, the generated HTML report does it correctly (because it takes
column into account).
The previous behavior caused lines in function literals to not be
tracked when those literals were inside Init or Cond of an IfStmt for
example.
Fixes #14039.
Change-Id: Iad591363330843ad833bd79a0388d709c8d0c8aa
Reviewed-on: https://go-review.googlesource.com/19775 Reviewed-by: Rob Pike <r@golang.org>
Robert Griesemer [Tue, 2 Feb 2016 18:14:36 +0000 (10:14 -0800)]
cmd/compile: bring vendored copy of math/big up-to-date
These files were not added to the repo. They contain conversion
routines and corresponding tests not used by the compiler and
thus are technically not needed.
However, future changes to math/big (and corresponding updates
of this vendored version) may require these files to exist.
Add them to avoid unnecessary confusion.
Change-Id: Ie390fb54f499463b2bba2fdc084967539afbeeb3
Reviewed-on: https://go-review.googlesource.com/19730 Reviewed-by: Alan Donovan <adonovan@google.com>
David Chase [Thu, 11 Feb 2016 20:09:43 +0000 (15:09 -0500)]
[dev.ssa] cmd/compile: double speed of CSE phase
Replaced comparison based on (*Type).String() with an
allocation-free structural comparison. Roughly doubles
speed of CSE, also reduces allocations.
Checked that roughly the same number of CSEs were detected
during make.bash (about a million) and that "new" CSEs
were caused by the effect described above.
Change-Id: Id205a9f6986efd518043e12d651f0b01206aeb1b
Reviewed-on: https://go-review.googlesource.com/19471 Reviewed-by: Keith Randall <khr@golang.org>
I tried something similar with Leq and Geq, but the results were
not great because it confuses the 'lowered cse' pass too much
which can no longer remove redundant comparisons from IsInBounds.
Change-Id: I2f928663a11320bfc51c7fa47e384b7411c420ba
Reviewed-on: https://go-review.googlesource.com/19727 Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
David Chase [Wed, 10 Feb 2016 22:43:31 +0000 (17:43 -0500)]
[dev.ssa] cmd/compile: memory allocation tweaks to regalloc and dom
Spotted a minor source of excess allocation in the register
allocator. Rearranged the dominator tree code to pull its
scratch memory from a reused buffer attached to Config.
Change-Id: I6da6e7b112f7d3eb1fd00c58faa8214cdea44e38
Reviewed-on: https://go-review.googlesource.com/19450 Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Todd Neal [Thu, 11 Feb 2016 01:39:32 +0000 (19:39 -0600)]
[dev.ssa] cmd/compile: add a zero arg cse pass
Add an initial cse pass that only operates on zero argument
values. This removes the need for a special case in cse for removing
OpSB and speeds up arithConst_ssa.go compilation by 9% while slowing
"test -c net/http" by 1.5%.
Matthew Dempsky [Sun, 21 Feb 2016 05:36:12 +0000 (21:36 -0800)]
cmd/compile: make cmpstackvarlt properly asymmetric
Previously, given two Nodes n1 and n2 of different non-PAUTO classes
(e.g., PPARAM and PPARAMOUT), cmpstackvarlt(n1, n2) and
cmpstackvarlt(n2, n1) both returned true, which is nonsense.
This doesn't seem to cause any visible miscompilation problems, but
notably fixing it does cause toolstash/buildall to fail.
Change-Id: I33b2c66e902c5eced875d8fbf18b7cfdc81e8aed
Reviewed-on: https://go-review.googlesource.com/19778
Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Sun, 21 Feb 2016 02:49:22 +0000 (18:49 -0800)]
cmd/compile: replace Order's use of NodeLists with slices
Order's "temp" and "free" fields use NodeLists in a rather
non-idiomatic way. Instead of using the "list" or "concat" functions,
it manipulates them directly and without the normal invariants (e.g.,
it doesn't maintain the "End" field).
Rather than convert it to more typical usage, just replace with a
slice, which ends up much simpler anyway.
Passes toolstash/buildall.
Change-Id: Ibd0f24324bd674c0d5bb1bc40d073b01e7824ad5
Reviewed-on: https://go-review.googlesource.com/19776
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Shenghou Ma [Sun, 21 Feb 2016 04:24:27 +0000 (23:24 -0500)]
runtime: fix missing word in comment
Change-Id: I6cb8ac7b59812e82111ab3b0f8303ab8194a5129
Reviewed-on: https://go-review.googlesource.com/19791 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Sun, 21 Feb 2016 04:33:34 +0000 (20:33 -0800)]
net: fix TestUpdateResolvConf after CL 18860
When writing a fake dnsConfig to conf.dnsConfig, set lastChecked to an
hour into the future. This causes dnsclient_unix.go's
tryUpdate("/etc/resolv.conf") calls to short-circuit and ignore that
/etc/resolv.conf's mtime differs from the test's fake resolv.conf
file. We only need to zero out lastChecked in teardown.
While here, this makes two other tryUpdate(conf.path) test calls
pointless, since they'll now short circuit too.
Robert Griesemer [Sat, 20 Feb 2016 20:53:34 +0000 (12:53 -0800)]
cmd/compile: set lexer nlsemi state directly
The old code used an extra function call and switch to inspect the
current token and determine the new state of curio.nlsemi. However,
the lexer knows the token w/o the need of an extra test and thus
can set curio.nlsemi directly:
- removed need for extra function call in next
- renamed _yylex to next
- set nlsemi at the point a token is identified
- moved nlsemi from curio to lexer - it's really part of the lexer state
This change makes the lexer call sequence less convoluted and should
also speed up the lexing a bit.
Change-Id: Iaf2683081f04231cb62c94e1400d455f98f6f82a
Reviewed-on: https://go-review.googlesource.com/19765 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Mikio Hara [Fri, 19 Feb 2016 08:34:54 +0000 (17:34 +0900)]
net/internal/socktest: add missing support for AcceptEx
Change-Id: I37faedc6fa316fffac80093b01e15429995b0f5b
Reviewed-on: https://go-review.googlesource.com/19705 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Matthew Dempsky [Sat, 20 Feb 2016 08:00:53 +0000 (00:00 -0800)]
cmd/compile: simplify if statement parsing
Somewhat notably, this means long if statement chains are now parsed
recursively, rather than iteratively. This shouldn't be a concern
though, as several other functions (e.g., gen, typecheck, walk)
already use recursion to process the parsed if statement Node trees.
Change-Id: Ic8c12ace9021c870d60c06f5db86a48c4ec57084
Reviewed-on: https://go-review.googlesource.com/19756 Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Sat, 20 Feb 2016 02:39:25 +0000 (18:39 -0800)]
cmd/compile: change two pushedio.bin tests to use importpkg instead
pushedio.bin and importpkg are both non-nil iff we're parsing an
package's export data, so "pushedio.bin == nil" and "importpkg == nil"
are equivalent tests.
Change-Id: I571ee908fef867117ef72c5da1eb24fe9b3fd12d
Reviewed-on: https://go-review.googlesource.com/19751 Reviewed-by: Robert Griesemer <gri@golang.org>
Matthew Dempsky [Tue, 1 Dec 2015 20:19:36 +0000 (12:19 -0800)]
cmd/compile: refactor export data parsing
Merge push_parser and pop_parser into a single parse_import function
and inline unimportfile. Shake out function boundaries a little bit so
that the symmetry is readily visible.
Move the import_package call into parse_import (and inline
import_there into import_package). This means importfile no longer
needs to provide fake import data to be needlessly lexed/parsed every
time it's called.
Also, instead of indicating import success/failure by whether the next
token is "package", import_spec can just check whether importpkg is
non-nil.
Tangentially, this somehow alters the diagnostics produced for
test/fixedbugs/issue11610.go. However, the new diagnostics are more
consistent with those produced when the empty import statement is
absent, which seems more desirable than maintaining the previous
errors.
Change-Id: I5cd1c22aa14da8a743ef569ff084711d137279d5
Reviewed-on: https://go-review.googlesource.com/19650 Reviewed-by: Robert Griesemer <gri@golang.org>
Martin Möhrmann [Fri, 19 Feb 2016 21:45:38 +0000 (22:45 +0100)]
fmt: remove math package dependency and avoid float operations
Remove floating point comparisons and rely only on the information
directly provided by appendFloat.
Make restoring the zero padding flag explicit instead of using a defer.
Rearrange some case distinctions to remove duplicated code.
Add more test cases for zero padded floating point numbers with sign.
benchmark old ns/op new ns/op delta
BenchmarkSprintfFloat-4 187 180 -3.74%
Change-Id: Ifa2ae85257909f40b1b18118c92b516933271729
Reviewed-on: https://go-review.googlesource.com/19721 Reviewed-by: Rob Pike <r@golang.org>
Matthew Dempsky [Tue, 1 Dec 2015 20:15:25 +0000 (12:15 -0800)]
cmd/compile: refactor import statement parsing
Combine parser's import_stmt and import_here methods as a single new
importdcl method, and cleanup conditional logic slightly to make the
code easier to follow.
Also, eliminate importfile's unused line parameter, and get rid of all
of its duplicate type assertions.
Change-Id: Ic37ae8490afedc533f98ead9feef383e3599bc01
Reviewed-on: https://go-review.googlesource.com/19629 Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Tue, 1 Dec 2015 20:05:30 +0000 (12:05 -0800)]
cmd/compile: make -A and -newexport compatible
Packages compiled with -A may reference the builtin "any" type, so it
needs to be included in the list of predeclared types for binary
import/export.
Also, when -A is used, mark all symbols as SymExport instead of
SymPackage in importsym. This parallels the logic in autoexport and
is necessary to prevent a "export/package mismatch" errors in
exportsym during dumpexport's verifyExport pass.
Change-Id: Iff5ec5fbfe2219525ec9d1a975307fa8936af9b9
Reviewed-on: https://go-review.googlesource.com/19627 Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Tue, 1 Dec 2015 20:02:42 +0000 (12:02 -0800)]
cmd/compile: load builtin export data only once
Previously, the builtin runtime export data was reparsed before every
Go source file, and the unsafe export data was reparsed for every
import of package unsafe. Now, we parse both of them just once ahead
of time.
This does mean package unsafe's export data will be loaded even when
compiling packages that don't import it, but it's tiny anyway.
David Chase [Fri, 19 Feb 2016 17:06:31 +0000 (12:06 -0500)]
cmd/compile: don't walk field-name syntax in esc.go
Walking the field name as if it were an expression
caused a called to haspointers with a TFIELD, which panics.
Trigger was a field at a large offset within a large struct,
combined with a struct literal expression mentioning that
field.
Damien Neil [Fri, 19 Feb 2016 17:58:22 +0000 (09:58 -0800)]
cmd/go: don't assume cc supports -gno-record-gcc-switches
NetBSD's C compiler appears to support -fdebug-prefix-map but
not -gno-record-gcc-switches. Remove assumption that support
for the former implies the latter.
Change-Id: Iecad9e4f497ea4edc1ce440010e6fe19dc3e0566
Reviewed-on: https://go-review.googlesource.com/19686 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>