Keith Randall [Tue, 4 Oct 2016 21:35:45 +0000 (14:35 -0700)]
cmd/compile: make CSE faster
To refine a set of possibly equivalent values, the old CSE algorithm
picked one value, compared it against all the others, and made two sets
out of the results (the values that match the picked value and the
values that didn't). Unfortunately, this leads to O(n^2) behavior. The
picked value ends up being equal to no other values, we make size 1 and
size n-1 sets, and then recurse on the size n-1 set.
Instead, sort the set by the equivalence classes of its arguments. Then
we just look for spots in the sorted list where the equivalence classes
of the arguments change. This lets us do a multi-way split for O(n lg
n) time.
This change makes cmpDepth unnecessary.
The refinement portion used to call the type comparator. That is
unnecessary as the type was already part of the initial partition.
Lowers time of 16361 from 8 sec to 3 sec.
Lowers time of 15112 from 282 sec to 20 sec. That's kind of unfair, as
CL 30257 changed it from 21 sec to 282 sec. But that CL fixed other bad
compile times (issue #17127) by large factors, so net still a big win.
For a simple application that I wrote, which reads in a CSV file (via
ReadAll) and outputs the number of rows read (15857625 rows), this change
reduces the total time on my notebook from ~58 seconds to ~48 seconds.
This reduces time and allocations (bytes) each by ~6% for a real world
CSV file at work (~230000 rows, 13 colums).
Michael Munday [Fri, 30 Sep 2016 18:20:42 +0000 (14:20 -0400)]
crypto/{aes,cipher}: add optimized implementation of AES-GCM for s390x
Also adds two tests: one to exercise the counter incrementing code
and one which checks the output of the optimized implementation
against that of the generic implementation for large/unaligned data
sizes.
Uses the KIMD instruction for GHASH and the KMCTR instruction for AES
in counter mode.
Brad Fitzpatrick [Wed, 5 Oct 2016 04:35:59 +0000 (04:35 +0000)]
sort: fix a slice benchmark not using the stable variant, add another
Change-Id: I9783d8023d453a72c4605a308064bef98168bcb8
Reviewed-on: https://go-review.googlesource.com/30360 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Ian Lance Taylor [Mon, 3 Oct 2016 23:58:34 +0000 (16:58 -0700)]
runtime: don't call cgocallback from signal handler
Calling cgocallback from a signal handler can fail when using the race
detector. Calling cgocallback will lead to a call to newextram which
will call oneNewExtraM which will call racegostart. The racegostart
function will set up some race detector data structures, and doing that
will sometimes call the C memory allocator. If we are running the signal
handler from a signal that interrupted the C memory allocator, we will
crash or hang.
Instead, change the signal handler code to call needm and dropm. The
needm function will grab allocated m and g structures and initialize the
g to use the current stack--the signal stack. That is all we need to
safely call code that allocates memory and checks whether it needs to
split the stack. This may temporarily leave us with no m available to
run a cgo callback, but that is OK in this case since the code we call
will quickly either crash or call dropm to return the m.
Implementing this required changing some of the setSignalstackSP
functions to avoid a write barrier. These functions never need a write
barrier but in some cases generated one anyhow because on some systems
the ss_sp field is a pointer.
Change-Id: I3893f47c3a66278f85eab7f94c1ab11d4f3be133
Reviewed-on: https://go-review.googlesource.com/30218
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Jeff R. Allen [Sat, 5 Dec 2015 15:06:05 +0000 (21:06 +0600)]
image/gif: check handling of truncated GIF files
All the prefixes of the testGIF produce errors today,
but they differ wildly in which errors: some are io.EOF,
others are io.ErrUnexpectedEOF, and others are gif-specific.
Make them all gif-specific to explain context, and make
any complaining about EOF be sure to mention the EOF
is unexpected.
Joe Tsai [Wed, 3 Aug 2016 07:31:17 +0000 (00:31 -0700)]
cmd/doc: ensure summaries truly are only one line
The documentation for doc says:
> Doc prints the documentation comments associated with the item identified by its
> arguments (a package, const, func, type, var, or method) followed by a one-line
> summary of each of the first-level items "under" that item (package-level
> declarations for a package, methods for a type, etc.).
Certain variables (and constants, functions, and types) have value specifications
that are multiple lines long. Prior to this change, doc would print out all of the
lines necessary to display the value. This is inconsistent with the documented
behavior, which guarantees a one-line summary for all first-level items.
We fix this here by writing a general oneLineNode method that always returns
a one-line summary (guaranteed!) of any input node.
Packages like image/color/palette and unicode now become much
more readable since large slices are now a single line.
$ go doc image/color/palette
<<<
// Before:
var Plan9 = []color.Color{
color.RGBA{0x00, 0x00, 0x00, 0xff},
color.RGBA{0x00, 0x00, 0x44, 0xff},
color.RGBA{0x00, 0x00, 0x88, 0xff},
... // Hundreds of more lines!
}
var WebSafe = []color.Color{
color.RGBA{0x00, 0x00, 0x00, 0xff},
color.RGBA{0x00, 0x00, 0x33, 0xff},
color.RGBA{0x00, 0x00, 0x66, 0xff},
... // Hundreds of more lines!
}
// After:
var Plan9 = []color.Color{ ... }
var WebSafe = []color.Color{ ... }
>>>
In order to test this, I ran `go doc` and `go doc -u` on all of the
standard library packages and diff'd the output with and without the
change to ensure that all differences were intended.
Fixes #13072
Change-Id: Ida10b7796b7e4e174a929b55c60813a9eb7158fe
Reviewed-on: https://go-review.googlesource.com/25420
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rob Pike <r@golang.org>
Robert Griesemer [Mon, 3 Oct 2016 20:28:25 +0000 (13:28 -0700)]
go/ast, go/parser: parse alias declarations
For now, we also accept "type p = p.T" (using = instead of =>, for
type aliases only), so we can experiment with an approach that only
uses type aliases. This concession is implemened in the parser.
For #16339
Change-Id: I88b5522a8b6cfc2e97ca146ede8b32af340220f8
Reviewed-on: https://go-review.googlesource.com/30211 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Carl Mastrangelo [Mon, 3 Oct 2016 19:45:12 +0000 (12:45 -0700)]
runtime: make append only clear uncopied memory
Also add a benchmark that shows off the new behavior. The
existing benchmarks reuse the same slice, and thus don't ever have
to clear memory. Running the Append|Grow benchmarks in runtime:
Change-Id: I69a8afbd942c787c591d95b9d9439bd6db4d1e49
Reviewed-on: https://go-review.googlesource.com/30192 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Quentin Renard [Sun, 6 Mar 2016 16:27:50 +0000 (17:27 +0100)]
net/http: improve performance for parsePostForm
Remove the use of io.ReadAll in http.parsePostForm to avoid converting
the whole input from []byte to string and not performing well
space-allocated-wise.
Instead a new function called parsePostFormURLEncoded is used and is
fed directly an io.Reader that is parsed using a bufio.Reader.
Benchmark:
name old time/op new time/op delta
PostQuery-4 2.90µs ± 6% 2.82µs ± 4% ~ (p=0.094 n=9+9)
name old alloc/op new alloc/op delta
PostQuery-4 1.05kB ± 0% 0.90kB ± 0% -14.49% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
PostQuery-4 6.00 ± 0% 7.00 ± 0% +16.67% (p=0.000 n=10+10)
Matthew Dempsky [Mon, 3 Oct 2016 23:09:13 +0000 (16:09 -0700)]
cmd/compile: cleanup artifacts from previous CL
Does not pass toolstash, but only because it causes ATYPE instructions
to be emitted in a different order, and it avoids emitting type
metadata for unused variables.
Robert Griesemer [Fri, 3 Jun 2016 20:21:55 +0000 (13:21 -0700)]
spec: ignore struct tags when converting structs
This is a backwards-compatible language change.
Per the proposal (#16085), the rules for conversions are relaxed
such that struct tags in any of the structs involved in the conversion
are ignored (recursively).
Because this is loosening the existing rules, code that compiled so
far will continue to compile.
Dhananjay Nakrani [Sun, 2 Oct 2016 01:01:30 +0000 (18:01 -0700)]
cmd/cover: Fix compiler directives handling.
Currently, it separates comments from rest of the AST. This causes problems when
long counter increment statements are added before compiler directives.
See Issue #17315.
This change moves comments handling into AST Visitor so that when printer prints
code from AST, position of compiler directives relative to the associated function
is preserved.
Nick Craig-Wood [Tue, 4 Oct 2016 13:04:18 +0000 (14:04 +0100)]
runtime: correct function name in throw message
Change-Id: I8fd271066925734c3f7196f64db04f27c4ce27cb
Reviewed-on: https://go-review.googlesource.com/30274 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Filippo Valsorda [Wed, 23 Dec 2015 02:03:44 +0000 (02:03 +0000)]
crypto/tls: implement countermeasures against CBC padding oracles
The aim is to make the decrypt() timing profile constant, irrespective of
the CBC padding length or correctness. The old algorithm, on valid padding,
would only MAC bytes up to the padding length threshold, making CBC
ciphersuites vulnerable to plaintext recovery attacks as presented in the
"Lucky Thirteen" paper.
The new algorithm Write()s to the MAC all supposed payload, performs a
constant time Sum()---which required implementing a constant time Sum() in
crypto/sha1, see the "Lucky Microseconds" paper---and then Write()s the rest
of the data. This is performed whether the padding is good or not.
This should have no explicit secret-dependent timings, but it does NOT
attempt to normalize memory accesses to prevent cache timing leaks.
Change-Id: Id838fcc234e71f735bb2609073f4c2214b48a970
Reviewed-on: https://go-review.googlesource.com/29891 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Mikio Hara [Sat, 1 Oct 2016 10:20:32 +0000 (19:20 +0900)]
syscall: fix Send{msg,msgN}, Recvmsg and control message handling on solaris
This change switches the use of socket implementation from the
conventional SUS-based one to the latest POSIX-based one to make
socket control message work correctly on Solaris.
It looks like those two implementations, Socket over TLI/XTI and
Socket, have different semantics in details but it wouldn't hurt
the existing applications because the exposed syscall API doesn't
support socket properties related to such a protocol independent
application framework.
Fixes #7402.
Change-Id: I45a4e782d606bfbebe1404086c50a8c69af53461
Reviewed-on: https://go-review.googlesource.com/30171 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Michael Munday [Fri, 30 Sep 2016 16:04:46 +0000 (12:04 -0400)]
cmd/compile: remove unnecessary OpSB checks in S390X.rules
Reversed, indexed and multi-register stores/loads cannot accept SB
inputs. Therefore if one of these Ops is an input to a rule any
pointer that is an argument to that Op cannot be OpSB.
Change-Id: Ib8048362d1c6277122afec0d13a1c905290d69cb
Reviewed-on: https://go-review.googlesource.com/30131
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
In particular, it wasn't obvious that some values are special (unless
you also found those special values), so document that it isn't
necessarily a hash value.
gcDumpObject is often used on a stack pointer (for example, when
checkmark finds an unmarked object on the stack), but since stack
spans don't have an elemsize, it doesn't print any of the memory from
the frame. Make it at least slightly more useful by printing
everything between obj and obj+off (inclusive). While we're here, also
print out the span state.
Keith Randall [Fri, 30 Sep 2016 17:12:32 +0000 (10:12 -0700)]
cmd/compile: Use Sreedhar+Gao phi building algorithm
Should be more asymptotically happy.
We process each variable in turn to find all the
locations where it needs a phi (the dominance frontier
of all of its definitions). Then we add all those phis.
This takes O(n * #variables), although hopefully much less.
Then we do a single tree walk to match all the
FwdRefs with the nearest definition or phi.
This takes O(n) time.
The one remaining inefficiency is that we might end up
introducing a bunch of dead phis in the first step.
A TODO is to introduce phis only where they might be
used by a read.
The old algorithm is still faster on small functions,
so there's a cutover size (currently 500 blocks).
This algorithm supercedes the David's sparse phi
placement algorithm for large functions.
Lowers compile time of example from #14934 from
~10 sec to ~4 sec.
Lowers compile time of example from #16361 from
~4.5 sec to ~3 sec.
Lowers #16407 from ~20 min to ~30 sec.
Update #14934
Update #16361
Fixes #16407
Change-Id: I1cff6364e1623c143190b6a924d7599e309db58f
Reviewed-on: https://go-review.googlesource.com/30163 Reviewed-by: David Chase <drchase@google.com>
Ian Lance Taylor [Mon, 3 Oct 2016 18:25:25 +0000 (11:25 -0700)]
doc: add PKG_CONFIG and GIT_ALLOW_PROTOCOL env vars to go1.8.txt
Change-Id: I592b87f49fc636b89807d911132f69257d718afd
Reviewed-on: https://go-review.googlesource.com/30168 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Billy Lynch [Fri, 30 Sep 2016 20:24:31 +0000 (16:24 -0400)]
cmd/go: add support for GIT_ALLOW_PROTOCOL
Allows users to override the default secure protocol list by setting the
GIT_ALLOW_PROTOCOL environment variable.
Addresses #17299 for vcs.go.
Change-Id: If575861d2b1b04b59029fed7e5d12b49690af50a
Reviewed-on: https://go-review.googlesource.com/30135 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Boris Nagaev [Sat, 25 Jun 2016 11:51:06 +0000 (13:51 +0200)]
cmd/dist, cmd/go: add environment variable override for pkg-config
Allow overriding default name of `pkg-config` tool via environment
variable PKG_CONFIG (same as used by autoconf pkg.m4 macros). This
facilitates easy cross-compilation of cgo code.
Original patch against Go <= 1.4 was written by
xnox_canonical <dimitri.ledkov@canonical.com> in 2014.
Source: https://codereview.appspot.com/104960043/
Fixes #16253
Change-Id: I31c33ffc3ecbff65da31421e6188d092ab4fe7e4
Reviewed-on: https://go-review.googlesource.com/29991 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Than McIntosh [Thu, 14 Jul 2016 17:23:11 +0000 (13:23 -0400)]
cmd/compile: relax liveness restrictions on ambiguously live
Update gc liveness to remove special conservative treatment
of ambiguously live vars, since there is no longer a need to
protect against GCDEBUG=gcdead.
Change-Id: Id6e2d03218f7d67911e8436d283005a124e6957f
Reviewed-on: https://go-review.googlesource.com/24896 Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Brad Fitzpatrick [Sun, 2 Oct 2016 03:28:09 +0000 (20:28 -0700)]
net: clarify that Conn deadlines also affect currently-blocked I/O
All implementations have always implemented this behavior, it's
tested, and it's depended on by other packages. (notably, by net/http)
The one exception is Plan 9 which doesn't support I/O deadlines at all
(tracked in #11932). As a result, a bunch of tests fail on plan9
(#7237). But once Plan 9 adds I/O deadline support, it'll also need
this behavior.
Change-Id: Idb71767f0c99279c66dce29f7bdc78ef467e47aa
Reviewed-on: https://go-review.googlesource.com/30164 Reviewed-by: Sam Whited <sam@samwhited.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Austin Clements [Mon, 3 Oct 2016 15:52:20 +0000 (11:52 -0400)]
runtime: weaken claim about SetFinalizer panicking
Currently the SetFinalizer documentation makes a strong claim that
SetFinalizer will panic if the pointer is not to an object allocated
by calling new, to a composite literal, or to a local variable. This
is not true. For example, it doesn't panic when passed the address of
a package-level variable. Nor can we practically make it true. For
example, we can't distinguish between passing a pointer to a composite
literal and passing a pointer to its first field.
Hence, weaken the guarantee to say that it "may" panic.
Updates #17311. (Might fix it, depending on what we want to do with
package-level variables.)
Florian Uekermann [Wed, 17 Aug 2016 16:45:57 +0000 (18:45 +0200)]
math/rand: add Rand.Uint64
This adds Uint64 methods to Rand and rngSource.
Rand.Uint64 uses Source.Uint64 directly if it is present.
rngSource.Uint64 provides access to all 64 bits generated by the
underlying ALFG. To ensure high seed quality a 64th bit has been added
to all elements of the array of "cooked" random numbers that are used
for seeding. gen_cooked.go generates both the 63 bit and 64 bit array.
Adam Langley [Fri, 30 Sep 2016 21:48:11 +0000 (14:48 -0700)]
crypto/ecdsa: correct code comment.
The code comment mixed up max and min. In this case, min is correct
because this entropy is only used to make the signature scheme
probabilistic. (I.e. if it were fixed then the scheme would still be
secure except that key.Sign(foo) would always give the same result for a
fixed key and foo.)
Adam Langley [Fri, 30 Sep 2016 21:16:12 +0000 (14:16 -0700)]
crypto/x509: return better error when a certificate contains no names.
Currently, if a certificate contains no names (that we parsed),
verification will return the confusing error:
x509: certificate is valid for , not example.com.
This change improves the error for that situation.
Adam Langley [Fri, 30 Sep 2016 23:54:54 +0000 (16:54 -0700)]
crypto/x509: fix name constraints handling.
This change brings the behaviour of X.509 name constraints into line
with NSS[1]. In this area, the behavior specified by the RFC and by NIST
differs and this code follows the NIST behaviour.
cmd/compile: improve error message for wrong number of arguments to return
Fixes #4215.
Fixes #6750.
Improves the error message for wrong number of arguments by comparing
the signature of the return call site arguments, versus the function's
expected return arguments.
In this CL, the signature representation of:
+ ideal numbers(TIDEAL) ie float*, complex*, rune, int is
"number" instead of "untyped number".
+ idealstring is "string" instead of "untyped string".
+ idealbool is "bool" instead of "untyped bool".
However, the representation of other types remains as the compiler
would produce.
* Example 1(in the error messages, if all lines were printed):
$ cat main.go && go run main.go
package main
func foo() (int, int) {
return 2.3
}
func foo2() {
return int(2), 2
}
func foo3(v int) (a, b, c, d int) {
if v >= 5 {
return 1
}
return 2, 3
}
func foo4(name string) (string, int) {
switch name {
case "cow":
return "moo"
case "dog":
return "dog", 10, true
case "fish":
return ""
default:
return "lizard", 10
}
}
Adam Langley [Fri, 30 Sep 2016 19:55:25 +0000 (12:55 -0700)]
crypto/tls: simplify keylog tests.
Since there's no aspect of key logging that OpenSSL can check for us,
the tests for it might as well just connect to another goroutine as this
is lower-maintainance.
strings, bytes: panic if Repeat overflows or if given a negative count
Panic if Repeat is given a negative count or
if the value of (len(*) * count) is detected
to overflow.
We panic because we cannot change the
signature of Repeat to return an error.
Mike Appleby [Wed, 28 Sep 2016 21:01:27 +0000 (16:01 -0500)]
runtime: sleep on CLOCK_MONOTONIC in futexsleep1 on freebsd
In FreeBSD 10.0, the _umtx_op syscall was changed to allow sleeping on
any supported clock, but the default clock was switched from a monotonic
clock to CLOCK_REALTIME.
Prior to 10.0, the __umtx_op_wait* functions ignored the fourth argument
to _umtx_op (uaddr1), expected the fifth argument (uaddr2) to be a
struct timespec pointer, and used a monotonic clock (nanouptime(9)) for
timeout calculations.
Since 10.0, if callers want a clock other than CLOCK_REALTIME, they must
call _umtx_op with uaddr1 set to a value greater than sizeof(struct
timespec), and with uaddr2 as pointer to a struct _umtx_time, rather
than a timespec. Callers can set the _clockid field of the struct
_umtx_time to request the clock they want.
The relevant FreeBSD commit:
https://svnweb.freebsd.org/base?view=revision&revision=232144
Fixes #17168
Change-Id: I3dd7b32b683622b8d7b4a6a8f9eb56401bed6bdf
Reviewed-on: https://go-review.googlesource.com/30154 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Mike Appleby [Sat, 1 Oct 2016 00:43:42 +0000 (19:43 -0500)]
time: update comment to reflect correct file path.
Update cross-reference in the comment for runtimeTimer to point to the
new go file instead of the old .h file.
Change-Id: Iddb3614c41e1989096d6caf77d6c0d5781005181
Reviewed-on: https://go-review.googlesource.com/30157 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Joe Tsai [Wed, 3 Aug 2016 01:42:58 +0000 (18:42 -0700)]
cmd/doc: perform type grouping for constants and variables
In golang.org/cl/22354, we added functionality to group functions under the
type that they construct to. In this CL, we extend the same concept to
constants and variables. This makes the doc tool more consistent with what
the godoc website does.
$ go doc reflect | egrep "ChanDir|Kind|SelectDir"
<<<
// Before:
const RecvDir ChanDir = 1 << iota ...
const Invalid Kind = iota ...
type ChanDir int
type Kind uint
type SelectDir int
func ChanOf(dir ChanDir, t Type) Type
// After:
type ChanDir int
const RecvDir ChanDir = 1 << iota ...
type Kind uint
const Invalid Kind = iota ...
type SelectDir int
const SelectSend SelectDir ...
func ChanOf(dir ChanDir, t Type) Type
>>>
Furthermore, a fix was made to ensure that the type was printed in constant
blocks when the iota was applied on an unexported field.
$ go doc reflect SelectSend
<<<
// Before:
const (
SelectSend // case Chan <- Send
SelectRecv // case <-Chan:
SelectDefault // default
)
// After:
const (
SelectSend SelectDir // case Chan <- Send
SelectRecv // case <-Chan:
SelectDefault // default
)
>>>
Fixes #16569
Change-Id: I26124c3d19e50caf9742bb936803a665e0fa6512
Reviewed-on: https://go-review.googlesource.com/25419 Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
David Benjamin [Fri, 1 Jul 2016 20:41:09 +0000 (16:41 -0400)]
crypto/tls: Fix c.in.decrypt error handling.
readRecord was not returning early if c.in.decrypt failed and ran
through the rest of the function. It does set c.in.err, so the various
checks in the callers do ultimately notice before acting on the result,
but we should avoid running the rest of the function at all.
Also rename 'err' to 'alertValue' since it isn't actually an error.
Change-Id: I6660924716a85af704bd3fe81521b34766238695
Reviewed-on: https://go-review.googlesource.com/24709
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Adam Langley <agl@golang.org>
Michael Munday [Wed, 28 Sep 2016 00:30:01 +0000 (20:30 -0400)]
cmd/compile: improve load/store merging on s390x
This commit makes the process of load/store merging more incremental
for both big and little endian operations. It also adds support for
32-bit shifts (needed to merge 16- and 32-bit loads/stores).
In addition, the merging of little endian stores is now supported.
Little endian stores are now up to 30 times faster.
Change-Id: Iefdd81eda4a65b335f23c3ff222146540083ad9c
Reviewed-on: https://go-review.googlesource.com/29956
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Alberto Donizetti [Fri, 30 Sep 2016 08:35:03 +0000 (10:35 +0200)]
cmd/compile: delete unused Convconst function
Convconst is not used in the new backend, and all its callers
were deleted in CL 29168 (cmd/compile: delete lots of the legacy
backend). iconv was an helper function for Convconst.
Updates bundled http2 to x/net git rev a333c53 for:
http2: add Transport support for IdleConnTimeout
https://golang.org/cl/30075
And add tests.
The bundled http2 also includes a change adding a Ping method to
http2.ClientConn, but that type isn't exposed in the standard
library. Nevertheless, the code gets compiled and adds a dependency on
"crypto/rand", requiring an update to go/build's dependency
test. Because net/http already depends on crypto/tls, which uses
crypto/rand, it's not really a new dependency.
Fixes #16808
Change-Id: I1ec8666ea74762f27c70a6f30a366a6647f923f7
Reviewed-on: https://go-review.googlesource.com/30078
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
David Crawshaw [Fri, 30 Sep 2016 00:21:52 +0000 (20:21 -0400)]
runtime: remove defer from standard cgo call
The endcgo function call is currently deferred in case a cgo
callback into Go panics and unwinds through cgocall. Typical cgo
calls do not have callbacks into Go, and even fewer panic, so we
pay the cost of this defer for no typical benefit.
Amazingly, there is another defer on the cgocallback path also used
to cleanup in case the Go code called by cgo panics. This CL folds
the first defer into the second, to reduce the cost of typical cgo
calls.
This reduces the overhead for a no-op cgo call significantly:
name old time/op new time/op delta
CgoNoop-8 93.5ns ± 0% 51.1ns ± 1% -45.34% (p=0.016 n=4+5)
The total effect between Go 1.7 and 1.8 is even greater, as CL 29656
reduced the cost of defer recently. Hopefully a future Go release
will drop the cost of defer to nothing, making this optimization
unnecessary. But until then, this is nice.
Change-Id: Id1a5648f687a87001d95bec6842e4054bd20ee4f
Reviewed-on: https://go-review.googlesource.com/30080
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Matthew Dempsky [Fri, 30 Sep 2016 04:34:44 +0000 (21:34 -0700)]
runtime, syscall: use FP instead of SP for parameters
Consistently access function parameters using the FP pseudo-register
instead of SP (e.g., x+0(FP) instead of x+4(SP) or x+8(SP), depending
on register size). Two reasons: 1) doc/asm says the SP pseudo-register
should use negative offsets in the range [-framesize, 0), and 2)
cmd/vet only validates parameter offsets when indexed from the FP
pseudo-register.
No binary changes to the compiled object files for any of the affected
package/OS/arch combinations.
Matthew Dempsky [Fri, 30 Sep 2016 02:09:36 +0000 (19:09 -0700)]
cmd/compile: eliminate stkdelta
At this point in the compiler we haven't assigned Xoffset values for
PAUTO variables anyway, so just immediately store the stack offsets
into Xoffset rather than into a global map.
Change-Id: I61eb471c857c8b145fd0895cbd98fd4e8d3c3365
Reviewed-on: https://go-review.googlesource.com/30081
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Matthew Dempsky [Thu, 29 Sep 2016 23:22:43 +0000 (16:22 -0700)]
cmd/compile: split Addrconst out of Naddr
There are only three Prog types that we were creating with an OLITERAL
Node: ATEXT, ATYPE, and AFUNCDATA. ATEXT's value we later overwrite in
defframe, and ATYPE's we don't even need. AFUNCDATA only needs integer
constants, so get rid of all the non-int constant logic and skip
creating a Node representation for the constant.
While here, there are a few other Naddr code paths that are no longer
needed, so turn those into Fatalfs.
Matthew Dempsky [Thu, 29 Sep 2016 22:43:10 +0000 (15:43 -0700)]
cmd/compile: make Afunclit the default/only behavior for Naddr
Naddr used to translate PFUNC Nodes into references to the function
literal wrapper, and then Afunclit could be used to rewrite it to
reference the function text itself. But now everywhere we use Naddr on
PFUNC Nodes, we immediately call Afunclit anyway. So just merge
Afunclit's behavior into Naddr.
net/http: use atomic.Value for Transport's alternate protocol map
Fix an old TODO and use atomic.Value for holding the Transport's
alternate protocol map. It is very frequently accessed and almost
never set or updated.
Currently any script tag is treated as a javascript container, although
<script type="text/template"> must not be. Check "type" attribute of
"script" tag. If it is present and it is not a JS MIME type, do not
transition to elementScript state.
Fixes #12149, where // inside text template was treated as regexp.
Fixes #6701
All other architectures merge stack-zeroing ranges if there are at
most two pointers/registers of memory between them, but x86 is
erroneously coded to require *exactly* two.
Shaves a tiny amount of text size off cmd/go when building for
GOARCH=386 and eliminates an unnecessary inconsistency between x86's
defframe and the other GOARCHes'.
text data bss dec hex filename 5241015 191051 93336 5525402 544f9a go.before 5240224 191051 93336 5524611 544c83 go.after
Introduced in CL 9263 (prepare to unexport gc.Mp*) and CL 9267
(prepare Node.Val to be unexported), their only callers were in
the old backend and all got deleted in CL 29168 (cmd/compile:
delete lots of the legacy backend).
Joe Tsai [Sat, 3 Sep 2016 03:15:12 +0000 (20:15 -0700)]
archive/tar: move parse/format functionality into strconv.go
Move all parse/format related functionality into strconv.go
and thoroughly test them. This also reduces the amount of noise
inside reader.go and writer.go.
There was zero functionality change other than moving code around.
Joe Tsai [Fri, 16 Sep 2016 23:46:19 +0000 (16:46 -0700)]
net/http: document how Request.Cookie deals with duplicate cookies
RFC 6265, section 4.2.2 says:
<<<
Although cookies are serialized linearly in the Cookie header,
servers SHOULD NOT rely upon the serialization order. In particular,
if the Cookie header contains two cookies with the same name (e.g.,
that were set with different Path or Domain attributes), servers
SHOULD NOT rely upon the order in which these cookies appear in the
header.
>>>
This statement seems to indicate that cookies should conceptually
be thought of as a map of keys to sets of values (map[key][]value).
However, in practice, everyone pretty much treats cookies as a
map[key]value and the API for Request.Cookie seems to indicate that.
We should update the documentation for Request.Cookie to warn the
user what happens when there is are multiple cookies with the same
key. I deliberately did not want to say *which* cookie is returned.
net: make proto and port lookups fall back to baked-in maps on Windows
In https://golang.org/cl/28951 I cleaned up the lookupProtocol and
lookupPort paths to be consistently case-insensitive across operating
systems and to share the same baked-in maps of port & proto values
that can be relied on to exist on any platform.
I missed the fallback to the baked-in maps on Windows, though, which
broke Windows XP. This should fix it.
Fixes #17175
Change-Id: Iecd434fb684304137ee27f5521cfaa8c351a1bde
Reviewed-on: https://go-review.googlesource.com/29968
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alberto Donizetti [Thu, 29 Sep 2016 16:46:24 +0000 (18:46 +0200)]
cmd/compile: delete unused IntLiteral function
IntLiteral was only called by the gins functions in
cmd/compile/internal/{arm64,mips64,ppc64}/gsubr.go
but CL 29220 (cmd/compile: remove gins) deleted them,
so IntLiteral is now unused.