Matthew Dempsky [Fri, 17 Mar 2017 20:35:31 +0000 (13:35 -0700)]
cmd/compile: eliminate direct uses of gc.Thearch in backends
This CL changes the GOARCH.Init functions to take gc.Thearch as a
parameter, which gc.Main supplies.
Additionally, the x86 backend is refactored to decide within Init
whether to use the 387 or SSE2 instruction generators, rather than for
each individual SSA Value/Block.
Robert Griesemer [Fri, 17 Mar 2017 04:13:29 +0000 (21:13 -0700)]
strconv: replace small int string table with constant string
This reduces memory use yet still provides the significant
performance gain seen when using a fast path for small integers.
Improvement of this CL comparing to code without fast path:
name old time/op new time/op delta
FormatIntSmall-8 35.6ns ± 1% 4.5ns ± 1% -87.30% (p=0.008 n=5+5)
AppendIntSmall-8 17.4ns ± 1% 9.4ns ± 3% -45.70% (p=0.008 n=5+5)
For comparison, here's the improvement before this CL to code without
fast path (1% better for FormatIntSmall):
name old time/op new time/op delta
FormatIntSmall-8 35.6ns ± 1% 4.0ns ± 3% -88.64% (p=0.008 n=5+5)
AppendIntSmall-8 17.4ns ± 1% 8.2ns ± 1% -52.80% (p=0.008 n=5+5)
Thus, the code in this CL performs slower for small integers using fast
path then the prior version, but this is relative to an already very fast
version:
name old time/op new time/op delta
FormatIntSmall-8 4.05ns ± 3% 4.52ns ± 1% +11.81% (p=0.008 n=5+5)
AppendIntSmall-8 8.21ns ± 1% 9.45ns ± 3% +15.05% (p=0.008 n=5+5)
Measured on 2.3 GHz Intel Core i7 running macOS Sierra 10.12.3.
Overall, it's still ~88% faster than without fast path for small integers,
so probably worth it as it removes 100 global string slices in favor of
a single string.
Credits: This is based on the original (but cleaned up) version of the
code by Aliaksandr Valialkin (https://go-review.googlesource.com/c/37963/).
Change-Id: Icda78679c8c14666d46257894e9fa3d7f35e58b8
Reviewed-on: https://go-review.googlesource.com/38319 Reviewed-by: Martin Möhrmann <moehrmann@google.com>
Alexander Menzhinsky [Fri, 3 Mar 2017 12:57:19 +0000 (15:57 +0300)]
cmd/go: fix race libraries rebuilding by `go test -i`
`go test -i -race` adds the "sync/atomic" package to every package dependency tree
that makes buildIDs different from packages installed with `go install -race`
and causes cache rebuilding.
Fixes #19133
Fixes #19151
Change-Id: I0536c6fa41b0d20fe361b5d35b3c0937b146d07d
Reviewed-on: https://go-review.googlesource.com/37598 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Alexey Neganov [Tue, 14 Mar 2017 18:11:42 +0000 (21:11 +0300)]
mime: handling invalid mime media parameters
Sometimes it's necessary to deal with emails that do not follow the specification; in particular, it's possible to download such email via gmail.
When the existing implementation handle invalid mime media parameters, it returns nils and error, although there is a valid media type, which may be returned.
If this behavior changes, it may not affect any existing programs, but it will help to parse some emails.
Fixes #19498
Change-Id: Ieb2fdbddfd93857faee941d2aa49d59e286d57fd
Reviewed-on: https://go-review.googlesource.com/38190 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
cmd/compile: rearrange fields between ssa.Func, ssa.Cache, and ssa.Config
This makes ssa.Func, ssa.Cache, and ssa.Config fulfill
the roles laid out for them in CL 38160.
The only non-trivial change in this CL is how cached
values and blocks get IDs. Prior to this CL, their IDs were
assigned as part of resetting the cache, and only modified
IDs were reset. This required knowing how many values and
blocks were modified, which required a tight coupling between
ssa.Func and ssa.Config. To eliminate that coupling,
we now zero values and blocks during reset,
and assign their IDs when they are used.
Since unused values and blocks have ID == 0,
we can efficiently find the last used value/block,
to avoid zeroing everything.
Bulk zeroing is efficient, but not efficient enough
to obviate the need to avoid zeroing everything every time.
As a happy side-effect, ssa.Func.Free is no longer necessary.
DebugHashMatch and friends now belong in func.go.
They have been left in place for clarity and review.
I will move them in a subsequent CL.
Passes toolstash -cmp. No compiler performance impact.
No change in 'go test cmd/compile/internal/ssa' execution time.
We could likely also eliminate the "assignment count mismatch" prefix now
without losing any information, but that string is matched by a number of
tests.
Change-Id: Ie6fc8a7bbd0ebe841d53e66e5c2f49868decf761
Reviewed-on: https://go-review.googlesource.com/38313 Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
8-bit args is a special case - the Go code is really fast because
it is just a single table lookup. So I've disabled that for now.
Intrinsics were actually slower:
LeadingZeros8-4 1.22ns ± 3% 1.58ns ± 1% +29.56% (p=0.000 n=10+10)
Update #18616
Change-Id: Ia9c289b9ba59c583ea64060470315fd637e814cf
Reviewed-on: https://go-review.googlesource.com/38311
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Steve Francia [Thu, 9 Feb 2017 20:33:13 +0000 (15:33 -0500)]
doc: reorganize the contribution guidelines into a guide
Updates #17802
Change-Id: I65ea0f4cde973604c04051e7eb25d12e4facecd3
Reviewed-on: https://go-review.googlesource.com/36626 Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Chris Broadfoot <cbro@golang.org>
Cherry Zhang [Tue, 14 Mar 2017 01:51:08 +0000 (21:51 -0400)]
cmd/compile: use type information in Aux for Store size
Remove size AuxInt in Store, and alignment in Move/Zero. We still
pass size AuxInt to Move/Zero, as it is used for partial Move/Zero
lowering (e.g. cmd/compile/internal/ssa/gen/386.rules:288).
SizeAndAlign is gone.
Cherry Zhang [Tue, 14 Feb 2017 02:27:53 +0000 (21:27 -0500)]
cmd/compile: add a test for writebarrier pass with single-block loop
The old writebarrier implementation fails to handle single-block
loop where a memory Phi value depends on the write barrier store
in the same block. The new implementation (CL 36834) doesn't have
this problem. Add a test to ensure it.
Fix #19067.
Change-Id: Iab13c6817edc12be8a048d18699b4450fa7ed712
Reviewed-on: https://go-review.googlesource.com/36940 Reviewed-by: David Chase <drchase@google.com>
Cherry Zhang [Mon, 6 Feb 2017 04:43:31 +0000 (23:43 -0500)]
cmd/compile: move write barrier insertion to SSA
When the compiler insert write barriers, the frontend makes
conservative decisions at an early stage. This sometimes have
false positives because of the lack of information, for example,
writes on stack. SSA's writebarrier pass identifies writes on
stack and eliminates write barriers for them.
This CL moves write barrier insertion into SSA. The frontend no
longer makes decisions about write barriers, and simply does
normal assignments and emits normal Store ops when building SSA.
SSA writebarrier pass inserts write barrier for Stores when needed.
There, it has better information about the store because Phi and
Copy propagation are done at that time.
This CL only changes StoreWB to Store in gc/ssa.go. A followup CL
simplifies SSA building code.
Cherry Zhang [Thu, 9 Feb 2017 14:46:44 +0000 (09:46 -0500)]
cmd/compile: pass types on SSA Store/Move/Zero ops
For SSA Store/Move/Zero ops, attach the type of the value being
stored to the op as the Aux field. This type will be used for
write barrier insertion (in a followup CL). Since SSA passes
do not accurately propagate types of values (because of type
casting), we can't simply use type of the store's arguments
for write barrier insertion.
Keith Randall [Tue, 14 Mar 2017 20:25:12 +0000 (13:25 -0700)]
cmd/compile: intrinsics for math/bits.TrailingZerosX
Implement math/bits.TrailingZerosX using intrinsics.
Generally reorganize the intrinsic spec a bit.
The instrinsics data structure is now built at init time.
This will make doing the other functions in math/bits easier.
Update sys.CtzX to return int instead of uint{64,32} so it
matches math/bits.TrailingZerosX.
Improve the intrinsics a bit for amd64. We don't need the CMOV
for <64 bit versions.
Update #18616
Change-Id: Ic1c5339c943f961d830ae56f12674d7b29d4ff39
Reviewed-on: https://go-review.googlesource.com/38155
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Martin Möhrmann [Sat, 25 Feb 2017 22:50:56 +0000 (23:50 +0100)]
runtime: make complex division c99 compatible
- changes tests to check that the real and imaginary part of the go complex
division result is equal to the result gcc produces for c99
- changes complex division code to satisfy new complex division test
- adds float functions isNan, isFinite, isInf, abs and copysign
in the runtime package
Austin Clements [Fri, 10 Mar 2017 15:59:39 +0000 (10:59 -0500)]
runtime: print user stack on other threads during GOTRACBEACK=crash
Currently, when printing tracebacks of other threads during
GOTRACEBACK=crash, if the thread is on the system stack we print only
the header for the user goroutine and fail to print its stack. This
happens because we passed the g0 to traceback instead of curg. The g0
never has anything set in its gobuf, so traceback doesn't print
anything.
Fix this by passing _g_.m.curg to traceback instead of the g0.
Fixes #19494.
Change-Id: Idfabf94d6a725e9cdf94a3923dead6455ef3b217
Reviewed-on: https://go-review.googlesource.com/38012
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Austin Clements [Wed, 15 Mar 2017 18:48:23 +0000 (14:48 -0400)]
runtime: make GOTRACEBACK=crash crash promptly in cgo binaries
GOTRACEBACK=crash works by bouncing a SIGQUIT around the process
sched.mcount times. However, sched.mcount includes the extra Ms
allocated by oneNewExtraM for cgo callbacks. Hence, if there are any
extra Ms that don't have real OS threads, we'll try to send SIGQUIT
more times than there are threads to catch it. Since nothing will
catch these extra signals, we'll fall back to blocking for five
seconds before aborting the process.
Avoid this five second delay by subtracting out the number of extra Ms
when sending SIGQUITs.
Of course, in a cgo binary, it's still possible for the SIGQUIT to go
to a cgo thread and cause some other failure mode. This does not fix
that.
Change-Id: I4fbf3c52dd721812796c4c1dcb2ab4cb7026d965
Reviewed-on: https://go-review.googlesource.com/38182
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
cmd/compile: check labels and gotos before building SSA
This CL introduces yet another compiler pass,
which checks for correct control flow constructs
prior to converting from AST to SSA form.
It cannot be integrated with walk, since walk rewrites
switch and select statements on the fly.
To reduce code duplication, this CL also does some
minor refactoring.
With this pass in place, the AST to SSA converter
can now stop generating SSA for any known-dead code.
This minor savings pays for the minor cost of the new pass.
Cherry Zhang [Mon, 13 Mar 2017 12:24:46 +0000 (08:24 -0400)]
cmd/link: on PPC64, put plt stubs at beginning of Textp
Put call stubs at the beginning (instead of the end). So the
trampoline pass knows the addresses of the stubs, and it can
insert trampolines when necessary.
cmd/compile: define roles for ssa.Func, ssa.Config, and ssa.Cache
The line between ssa.Func and ssa.Config has blurred.
Concurrent compilation in the backend will require more precision.
This CL lays out an (aspirational) organization.
The implementation will come in follow-up CLs,
once the organization is settled.
ssa.Config holds basic compiler configuration,
mostly arch-specific information.
It is configured once, early on, and is readonly,
so it is safe for concurrent use.
ssa.Func is a single-shot object used for
compiling a single Func. It is not concurrency-safe
and not re-usable.
ssa.Cache is a multi-use object used to avoid
expensive allocations during compilation.
Each ssa.Func is given an ssa.Cache to use.
ssa.Cache is not concurrency-safe.
Change-Id: Id02809b6f3541541cac6c27bbb598834888ce1cc
Reviewed-on: https://go-review.googlesource.com/38160 Reviewed-by: Keith Randall <khr@golang.org>
David Chase [Tue, 7 Mar 2017 19:45:46 +0000 (14:45 -0500)]
cmd/compile: put spills in better places
Previously we always issued a spill right after the op
that was being spilled. This CL pushes spills father away
from the generator, hopefully pushing them into unlikely branches.
For example:
x = ...
if unlikely {
call ...
}
... use x ...
Used to compile to
x = ...
spill x
if unlikely {
call ...
restore x
}
It now compiles to
x = ...
if unlikely {
spill x
call ...
restore x
}
This is particularly useful for code which appends, as the only
call is an unlikely call to growslice. It also helps for the
spills needed around write barrier calls.
The basic algorithm is walk down the dominator tree following a
path where the block still dominates all of the restores. We're
looking for a block that:
1) dominates all restores
2) has the value being spilled in a register
3) has a loop depth no deeper than the value being spilled
The walking-down code is iterative. I was forced to limit it to
searching 100 blocks so it doesn't become O(n^2). Maybe one day
we'll find a better way.
I had to delete most of David's code which pushed spills out of loops.
I suspect this CL subsumes most of the cases that his code handled.
Generally positive performance improvements, but hard to tell for sure
with all the noise. (compilebench times are unchanged.)
Philip Hofer [Tue, 14 Mar 2017 21:00:38 +0000 (14:00 -0700)]
cmd/compile/internal/gc: mark generated wrappers as DUPOK
Interface wrapper functions now get compiled eagerly in some cases.
Consequently, they may be present in multiple translation units.
Mark them as DUPOK, just like closures.
Fixes #19548
Fixes #19550
Change-Id: Ibe74adb5a62dbf6447db37fde22dcbb3479969ef
Reviewed-on: https://go-review.googlesource.com/38156 Reviewed-by: David Chase <drchase@google.com>
Matthew Dempsky [Thu, 9 Mar 2017 20:15:41 +0000 (12:15 -0800)]
cmd/compile: refactor liveness analysis for moving to SSA
In the SSA CFG, TEXT, RET, and JMP instructions correspond to Blocks,
not Values. Rework liveness analysis so that progeffects only cares
about Progs that result from Values, and handle Blocks separately.
Daniel Martí [Fri, 10 Mar 2017 17:15:46 +0000 (17:15 +0000)]
*.bash: always use the same string equality operator
POSIX Shell only supports = to compare variables inside '[' tests. But
this is Bash, where == is an alias for =. In practice they're the same,
but the current form is inconsisnent and breaks POSIX for no good
reason.
Change-Id: I38fa7a5a90658dc51acc2acd143049e510424ed8
Reviewed-on: https://go-review.googlesource.com/38031 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
cmd/compile: eliminate fmtmode and fmtpkgpfx globals
The fmtmode and fmtpkgpfx globals stand in the
way of making the compiler more concurrent (#15756).
This CL removes them.
The natural way to eliminate a global is to explicitly
thread it as a parameter through all function calls.
However, most of the functions in gc/fmt.go
get called indirectly, by way of fmt format strings,
so there's nowhere natural to add a parameter.
Since there are only a few fmtmode modes,
use named types to distinguish between modes.
For example, fmtNodeErr, fmtNodeDbg, and fmtNodeTypeId
are all gc.Node, but they print in different modes.
Varying the type allows us to thread mode through fmt.
Handle fmtpkgpfx by converting it to a printing mode,
FTypeIdName, and using the same type-based approach.
To avoid a loss of readability and danger of bugs
from introducing conversions at all call sites,
instead add a helper that systematically modifies the args.
The only remaining gc/fmt.go global is dumpdepth.
Since that is used for debugging only,
it that can be handled with a global mutex,
or some similarly basic, if inefficient, protection.
Passes toolstash -cmp. No compiler performance impact.
For future reference, other options for threading state
that were considered and rejected:
* Wrapping values in structs, such as:
type fmtNode struct {
n *Node
mode fmtMode
}
This reduces the proliferation of types, and supports
easily adding extra local parameters.
However, putting such a struct into an interface{} allocates.
This is unacceptable in this particular area of code.
* Passing state via precision, such as:
fmt.Fprintf("%*v", mode, n)
where mode is the state encoded as an integer.
This avoids extra allocations, but it is out of keeping
with the intended semantics of precision, and is less readable.
* Modify the fmt package to support setting/getting context
via fmt.State. Unavailable due to Go 1 compatibility,
and probably the wrong solution anyway.
* Give up on package fmt. This would be a huge readability
regression and cause high code churn.
* Attempt a de-novo rewrite that circumvents these problems.
Too high a risk of bugs, with insufficient reward for the effort,
particularly since long term plans call for elimination
of gc.Node.
philhofer [Mon, 13 Mar 2017 22:03:17 +0000 (15:03 -0700)]
cmd/compile: de-virtualize interface calls
With this change, code like
h := sha1.New()
h.Write(buf)
sum := h.Sum()
gets compiled into static calls rather than
interface calls, because the compiler is able
to prove that 'h' is really a *sha1.digest.
The InterCall re-write rule hits a few dozen times
during make.bash, and hundreds of times during all.bash.
The most common pattern identified by the compiler
is a constructor like
func New() Interface { return &impl{...} }
where the constructor gets inlined into the caller,
and the result is used immediately. Examples include
{sha1,md5,crc32,crc64,...}.New, base64.NewEncoder,
base64.NewDecoder, errors.New, net.Pipe, and so on.
Some existing benchmarks that change on darwin/amd64:
On architectures like amd64, the reduction in code size
appears to contribute more to benchmark improvements than just
removing the indirect call, since that branch gets predicted
accurately when called in a loop.
Updates #19361
Change-Id: I57d4dc21ef40a05ec0fbd55a9bb0eb74cdc67a3d
Reviewed-on: https://go-review.googlesource.com/38139
Run-TryBot: Philip Hofer <phofer@umich.edu>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
Matthew Dempsky [Thu, 9 Mar 2017 22:46:43 +0000 (14:46 -0800)]
cmd/compile/internal/ssa: populate SymEffects for SSA Ops
Changes to ${GOARCH}Ops.go files were mechanically produced using
github.com/mdempsky/ssa-symops, a one-off tool that inserts
"SymEffect: X" elements by pattern matching against the Op names.
Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf
Reviewed-on: https://go-review.googlesource.com/38084
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Reason for revert: The presence-of-optimization test program is fragile, breaks under noopt, and might break if the Go libraries are tweaked. It needs to be (re)written without reference to other packages.
Change-Id: I3aaf1ab006a1a255f961a978e9c984341740e3c7
Reviewed-on: https://go-review.googlesource.com/38097 Reviewed-by: Keith Randall <khr@golang.org>
This abstracts creation of ACALL Progs into package gc. The main
benefit of this today is we can refactor away a lot of common
boilerplate code.
Later, once liveness analysis happens on the SSA graph, this will also
provide an easy insertion point for emitting the PCDATA Progs
immediately before call instructions.
Passes toolstash-check -all.
Change-Id: Ia15108ace97201cd84314f1ca916dfeb4f09d61c
Reviewed-on: https://go-review.googlesource.com/38081 Reviewed-by: Keith Randall <khr@golang.org>
Alberto Donizetti [Mon, 13 Mar 2017 19:35:07 +0000 (20:35 +0100)]
encoding/gob: add Encode-Decode Int slices tests
Tinkering with the gob package shows that is currently possible to
*completely destroy* Int slices encoding without triggering a single
test failure.
The various encInt{8,16,32,64}Slice methods are only called during the
execution of the GobMapInterfaceEncode test, which only encodes a few
slices of length exactly 1 and then just checks that the error
returned by Encode is nil (without trying to Decode back the data).
This patch adds a few tests for signed integer slices encoding.
khr [Thu, 9 Mar 2017 18:38:45 +0000 (10:38 -0800)]
cmd/compile: zero return parameters earlier
Move the zeroing of results earlier. In particular, they need to
come before any move-to-heap operations, as those require allocation.
Those allocations are points at which the GC can see the uninitialized
result slots.
For the function:
func f() (x, y, z *int) {
defer(){}()
escape(&y)
return
}
We used to generate code like this:
x = nil
y = nil
&y = new(int)
z = nil
Now we will generate:
x = nil
y = nil
z = nil
&y = new(int)
Since the fix for #18860, the return slots are always live if there
is a defer, so the former ordering allowed the GC to see junk
in the z slot.
Fixes #19078
Change-Id: I71554ae437549725bb79e13b2c100b2911d47ed4
Reviewed-on: https://go-review.googlesource.com/38133
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Philip Hofer [Fri, 3 Mar 2017 18:57:53 +0000 (10:57 -0800)]
cmd/compile: de-virtualize interface calls
With this change, code like
h := sha1.New()
h.Write(buf)
sum := h.Sum()
gets compiled into static calls rather than
interface calls, because the compiler is able
to prove that 'h' is really a *sha1.digest.
The InterCall re-write rule hits a few dozen times
during make.bash, and hundreds of times during all.bash.
The most common pattern identified by the compiler
is a constructor like
func New() Interface { return &impl{...} }
where the constructor gets inlined into the caller,
and the result is used immediately. Examples include
{sha1,md5,crc32,crc64,...}.New, base64.NewEncoder,
base64.NewDecoder, errors.New, net.Pipe, and so on.
Some existing benchmarks that change on darwin/amd64:
On architectures like amd64, the reduction in code size
appears to contribute more to benchmark improvements than just
removing the indirect call, since that branch gets predicted
accurately when called in a loop.
Updates #19361
Change-Id: Ia9d30afdd5f6b4d38d38b14b88f308acae8ce7ed
Reviewed-on: https://go-review.googlesource.com/37751
Run-TryBot: Philip Hofer <phofer@umich.edu>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Robert Griesemer [Sat, 11 Mar 2017 01:17:23 +0000 (17:17 -0800)]
spec: remove superfluous terms "delimiter" and "special tokens"
The (original) section on "Operators and Delimiters" introduced
superfluous terminology ("delimiter", "special token") which
didn't matter and was used inconsistently.
Removed any mention of "delimiter" or "special token" and now
simply group the special character tokens into "operators"
(clearly defined via links), and "punctuation" (everything else).
Fixes #19450.
Change-Id: Ife31b24b95167ace096f93ed180b7eae41c66808
Reviewed-on: https://go-review.googlesource.com/38073 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Rob Pike <r@golang.org>
In Go 1.7 and earlier, gc.exportsize tracked the number of bytes
written through exportf. With the removal of the old exporter in Go 1.8
exportf is only used for printing the build id, and the header and
trailer of the binary export format. The size of the export data is
now returned directly from the exporter and exportsize is never
referenced. Remove it.
Martin Möhrmann [Sat, 11 Mar 2017 07:48:56 +0000 (08:48 +0100)]
strconv: fix performance regression in integer formatting on 32bit platforms
Some of the changes in CL golang.org/cl/38071/ assumed that / and %
could always be combined to use only one DIV instruction. However,
this is not the case for 64bit operands on a 32bit platform which use
seperate runtime functions to calculate division and modulo.
This CL restores the original optimizations that help on 32bit platforms
with negligible impact on 64bit platforms.
Ian Lance Taylor [Sat, 11 Mar 2017 01:03:52 +0000 (17:03 -0800)]
cmd/go: if we get a C compiler dwarf2 warning, try without -g
This avoids a problem that occurs on FreeBSD 11, in which the clang
3.8 assembler issues a pointless warning when invoked with -g on a
file that contains an empty .note.GNU-stack section.
No test because there is no reasonable way to write one, but should
fix the build on FreeBSD 11.
Fixes #14705.
Change-Id: I8c49bbf79a2c715c0e75495da19045fc92280e81
Reviewed-on: https://go-review.googlesource.com/38072
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
nat.setUint64 is nicely generic.
By assuming 32- or 64-bit words, however,
we can write simpler code,
and eliminate some shifts
in dead code that vet complains about.
Generated code for 64 bit systems is unaltered.
Generated code for 32 bit systems is much better.
For 386, the routine length drops from 325
bytes of code to 271 bytes of code, with fewer loops.
Robert Griesemer [Fri, 10 Mar 2017 23:14:54 +0000 (15:14 -0800)]
strconv: use % instead of computing the remainder from the quotient
The compiler recognizes that in a sequence q = x/y; r = x%y only
one division is required. Remove prior work-arounds and write
more readable straight-line code (this also results in fewer
instructions, though it doesn't appear to affect the benchmarks
significantly).
Also:
- use uint instead of uintptr where we want to guarantee single-
register operations
- remove some unnecessary conversions (before indexing)
- add more comments and fix some comments
David NewHamlet [Fri, 10 Mar 2017 20:13:20 +0000 (09:13 +1300)]
runtime: use cpuset_getaffinity for runtime.NumCPU() on FreeBSD
In FreeBSD when run Go proc under a given sub-list of
processors(e.g. 'cpuset -l 0 ./a.out' in multi-core system),
runtime.NumCPU() still return all physical CPUs from sysctl
hw.ncpu instead of account from sub-list.
Fix by use syscall cpuset_getaffinity to account the number of sub-list.
Fixes #15206
Change-Id: If87c4b620e870486efa100685db5debbf1210a5b
Reviewed-on: https://go-review.googlesource.com/29341 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Ian Lance Taylor [Fri, 10 Mar 2017 18:04:42 +0000 (10:04 -0800)]
cmd/go: when expanding "cmd", skip vendored main packages
We are vendoring pprof from github.com/google/pprof, which comes with
a main package. If we don't explicitly skip that main package, then
`go install cmd` will install the compiled program in $GOROOT/bin.
Bryan C. Mills [Thu, 9 Mar 2017 21:45:14 +0000 (16:45 -0500)]
expvar: add benchmark for (*Map).Set with per-goroutine keys
Change-Id: I0fa68ca9812fe5e82ffb9d0b9598e95b47183eb8
Reviewed-on: https://go-review.googlesource.com/38011 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
net/http: change TestServerAllowsBlockingRemoteAddr to non-parallel
It appears that this test is particularly
sensitive to resource starvation.
Returning it to non-parallel should reduce flakiness,
by giving it the full system resources to run.
cmd/compile: add OpOffPtr [c] SP to constant cache
They accounted for almost 30% of all CSE'd values.
By never creating the duplicates in the first place,
we reduce the high water mark of Value IDs,
which in turn makes all SSA phases cheaper,
particularly regalloc.
Robert Griesemer [Wed, 8 Mar 2017 22:26:23 +0000 (14:26 -0800)]
cmd/compile: print columns (not just lines) in error messages
Compiler errors now show the exact line and line byte offset (sometimes
called "column") of where an error occured. For `go tool compile x.go`:
package p
const c int = false
//line foo.go:123
type t intg
reports
x.go:2:7: cannot convert false to type int
foo.go:123[x.go:4:8]: undefined: intg
(Some errors use the "wrong" position for the error message; arguably
the byte offset for the first error should be 15, the position of 'false',
rathen than 7, the position of 'c'. But that is an indepedent issue.)
The byte offset (column) values are measured in bytes; they start at 1,
matching the convention used by editors and IDEs.
Positions modified by //line directives show the line offset only for the
actual source location (in square brackets), not for the "virtual" file and
line number because that code is likely generated and the //line directive
only provides line information.
Because the new format might break existing tools or scripts, printing
of line offsets can be disabled with the new compiler flag -C. We plan
to remove this flag eventually.
Fixes #10324.
Change-Id: I493f5ee6e78457cf7b00025aba6b6e28e50bb740
Reviewed-on: https://go-review.googlesource.com/37970 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Robert Griesemer [Thu, 9 Mar 2017 21:38:10 +0000 (13:38 -0800)]
cmd/compile/internal/syntax: start line offset (column) numbers at 1
We could leave it alone and fix line offset (column) numbers when
reporting errors, but that is likely to cause confusion (internal
numbers don't match reported numbers). Instead, switch to default
numbering starting at 1.
For package syntax-internal use only, introduced constants defining
the line and column bases, and use them throughout the code and its
tests. It is possible to change these constants and package syntax
will continue to work. But changing them is going to break any client
that makes explicit assumptions about line and column numbers (which
is "all of them").
Change-Id: Ia3d136a8ec8d9372ed9c05ca47d3dff222cf030e
Reviewed-on: https://go-review.googlesource.com/37996 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Paul Marks [Wed, 2 Nov 2016 04:01:08 +0000 (21:01 -0700)]
net: add Resolver.StrictErrors
When LookupIP is performing multiple subqueries, this option causes a
timeout/servfail affecting a single query to abort the whole operation,
instead of returning a partial (IPv4/IPv6-only) result.
Similarly, operations that walk the DNS search list will also abort when
encountering one of these errors.
Daniel Martí [Thu, 9 Mar 2017 20:50:24 +0000 (20:50 +0000)]
src/*.bash: use tabs consistently
make.bash used mostly tabs and buildall.bash used mostly spaces, but
they were both mixing them. Be consistent and use tabs, as that's what's
more common and what the Go code uses.