]> Cypherpunks repositories - gostls13.git/log
gostls13.git
8 years agocmd/compile/internal/ppc64, cmd/compile/internal/ssa: Remove OldArch checks
Carlos Eduardo Seo [Wed, 22 Mar 2017 15:25:44 +0000 (12:25 -0300)]
cmd/compile/internal/ppc64, cmd/compile/internal/ssa: Remove OldArch checks

Starting in go1.9, the minimum processor requirement for ppc64 is POWER8.
Therefore, the checks for OldArch and the code enabled by it are not necessary
anymore.

Updates #19074

Change-Id: I33d6a78b2462c80d57c5dbcba2e13424630afab4
Reviewed-on: https://go-review.googlesource.com/38404
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime/internal/atomic: Remove unnecessary checks for GOARCH_ppc64
Carlos Eduardo Seo [Wed, 22 Mar 2017 17:25:30 +0000 (14:25 -0300)]
runtime/internal/atomic: Remove unnecessary checks for GOARCH_ppc64

Starting in go1.9, the minimum processor requirement for ppc64 is POWER8. This
means the checks for GOARCH_ppc64 in asm_ppc64x.s can be removed, since we can
assume LBAR and STBCCC instructions (both from ISA 2.06) will always be
available.

Updates #19074

Change-Id: Ib4418169cd9fc6f871a5ab126b28ee58a2f349e2
Reviewed-on: https://go-review.googlesource.com/38406
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/gc: make SSAGenFPJump a method of SSAGenState
Matthew Dempsky [Wed, 22 Mar 2017 17:27:30 +0000 (10:27 -0700)]
cmd/compile/internal/gc: make SSAGenFPJump a method of SSAGenState

Change-Id: Ie22a08c93dfcfd4b336e7b158415448dd55b2c11
Reviewed-on: https://go-review.googlesource.com/38407
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/internal/obj: eliminate Link.Asmode
Josh Bleecher Snyder [Wed, 22 Mar 2017 17:17:46 +0000 (10:17 -0700)]
cmd/internal/obj: eliminate Link.Asmode

Asmode is always set to p.Mode,
which is always set based on the arch family.
Instead, use the arch family directly.

Passes toolstash-check -all.

Change-Id: Id982472dcc8eeb6dd22cac5ad2f116b54a44caee
Reviewed-on: https://go-review.googlesource.com/38451
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/internal/obj: eliminate Ctxt.Mode
Josh Bleecher Snyder [Wed, 22 Mar 2017 05:23:45 +0000 (22:23 -0700)]
cmd/internal/obj: eliminate Ctxt.Mode

Replace Ctxt.Mode with a method, Ctxt.RegWidth,
which is calculated directly off the arch info.

I believe that Prog.Mode can also be removed; future CL.

This is a step towards obj.Link immutability.

Passes toolstash-check -all.

Updates #15756

Change-Id: Ifd7f8f6ed0a2fdc032d1dd306fcd695a14aa5bc5
Reviewed-on: https://go-review.googlesource.com/38446
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/compile: funnel SSA Prog creation through SSAGenState
Josh Bleecher Snyder [Mon, 20 Mar 2017 15:01:28 +0000 (08:01 -0700)]
cmd/compile: funnel SSA Prog creation through SSAGenState

Step one in eliminating Prog-related globals.

Passes toolstash-check -all.

Updates #15756

Change-Id: I3b777fb5a7716f2d9da3067fbd94c28ca894a465
Reviewed-on: https://go-review.googlesource.com/38450
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/vet: -lostcancel: fix crash in ill-typed code
Alan Donovan [Wed, 22 Mar 2017 16:38:09 +0000 (12:38 -0400)]
cmd/vet: -lostcancel: fix crash in ill-typed code

Fixes golang/go#19656

Change-Id: Ied20d3f25b6e147cc693a1dd1aeb9480adc6687e
Reviewed-on: https://go-review.googlesource.com/38405
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>

8 years agonet/http: fix possible nil pointer dereference in TestOnlyWriteTimeout
Mikio Hara [Thu, 9 Feb 2017 21:22:39 +0000 (06:22 +0900)]
net/http: fix possible nil pointer dereference in TestOnlyWriteTimeout

TestOnlyWriteTimeout assumes wrongly that:
- the Accept method of trackLastConnListener is called only once
- the shared variable conn never becomes nil
and crashes on some circumstances.

Updates #19032.

Change-Id: I61de22618cd90b84a2b6401afdb6e5d9b3336b12
Reviewed-on: https://go-review.googlesource.com/36735
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: pass an explicit position to namedAuto
Josh Bleecher Snyder [Wed, 22 Mar 2017 03:58:00 +0000 (20:58 -0700)]
cmd/compile: pass an explicit position to namedAuto

To enable this, inline the call to nod and simplify.
Eliminates a reference to lineno from the backend.

Passes toolstash-check -all.

Updates #15756

Change-Id: I9c4bd77d10d727aa8f5e6c6bb16b0e05de165631
Reviewed-on: https://go-review.googlesource.com/38441
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/internal/obj: clean up brloop
Josh Bleecher Snyder [Wed, 22 Mar 2017 05:06:39 +0000 (22:06 -0700)]
cmd/internal/obj: clean up brloop

Add docs.
Reduce indentation.

Passes toolstash-check -all.

Change-Id: I968d1af25989886ae9945052e05e211a107dde9c
Reviewed-on: https://go-review.googlesource.com/38443
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agostrconv: optimize decimal ints formatting with smallsString
Aliaksandr Valialkin [Thu, 16 Mar 2017 17:19:29 +0000 (19:19 +0200)]
strconv: optimize decimal ints formatting with smallsString

Benchmark results for GOARCH=amd64:

name                                     old time/op  new time/op  delta
FormatInt-4                              2.51µs ± 2%  2.40µs ± 2%   -4.51%  (p=0.000 n=9+10)
AppendInt-4                              1.67µs ± 2%  1.61µs ± 3%   -3.74%  (p=0.000 n=9+9)
FormatUint-4                              698ns ± 2%   643ns ± 3%   -7.95%  (p=0.000 n=10+8)
AppendUint-4                              478ns ± 1%   418ns ± 2%  -12.61%  (p=0.000 n=8+10)
AppendUintVarlen/1-4                     9.30ns ± 6%  9.15ns ± 1%     ~     (p=0.199 n=9+10)
AppendUintVarlen/12-4                    9.12ns ± 0%  9.16ns ± 2%     ~     (p=0.307 n=9+9)
AppendUintVarlen/123-4                   18.6ns ± 2%  18.7ns ± 0%     ~     (p=0.091 n=10+6)
AppendUintVarlen/1234-4                  19.1ns ± 4%  17.7ns ± 1%   -7.35%  (p=0.000 n=10+9)
AppendUintVarlen/12345-4                 21.5ns ± 3%  20.7ns ± 3%   -3.78%  (p=0.002 n=9+10)
AppendUintVarlen/123456-4                23.5ns ± 3%  20.9ns ± 1%  -11.14%  (p=0.000 n=10+9)
AppendUintVarlen/1234567-4               25.0ns ± 2%  23.6ns ± 7%   -5.48%  (p=0.004 n=9+10)
AppendUintVarlen/12345678-4              26.8ns ± 2%  23.4ns ± 2%  -12.79%  (p=0.000 n=9+10)
AppendUintVarlen/123456789-4             29.8ns ± 3%  26.5ns ± 5%  -11.03%  (p=0.000 n=10+10)
AppendUintVarlen/1234567890-4            31.6ns ± 3%  26.9ns ± 3%  -14.95%  (p=0.000 n=10+9)
AppendUintVarlen/12345678901-4           33.8ns ± 3%  29.3ns ± 5%  -13.21%  (p=0.000 n=10+10)
AppendUintVarlen/123456789012-4          35.5ns ± 4%  29.2ns ± 4%  -17.82%  (p=0.000 n=10+10)
AppendUintVarlen/1234567890123-4         37.6ns ± 4%  31.4ns ± 3%  -16.48%  (p=0.000 n=10+10)
AppendUintVarlen/12345678901234-4        39.8ns ± 6%  32.0ns ± 7%  -19.60%  (p=0.000 n=10+10)
AppendUintVarlen/123456789012345-4       40.7ns ± 0%  34.4ns ± 4%  -15.55%  (p=0.000 n=6+10)
AppendUintVarlen/1234567890123456-4      45.4ns ± 6%  35.1ns ± 4%  -22.66%  (p=0.000 n=10+10)
AppendUintVarlen/12345678901234567-4     45.1ns ± 1%  36.7ns ± 4%  -18.77%  (p=0.000 n=9+10)
AppendUintVarlen/123456789012345678-4    46.9ns ± 0%  36.4ns ± 3%  -22.49%  (p=0.000 n=9+10)
AppendUintVarlen/1234567890123456789-4   50.6ns ± 6%  38.8ns ± 3%  -23.28%  (p=0.000 n=10+10)
AppendUintVarlen/12345678901234567890-4  51.3ns ± 2%  38.4ns ± 0%  -25.00%  (p=0.000 n=9+8)

Benchmark results for GOARCH=386:

name                                     old time/op  new time/op  delta
FormatInt-4                              6.21µs ± 0%  6.14µs ± 0%  -1.11%  (p=0.008 n=5+5)
AppendInt-4                              4.95µs ± 0%  4.85µs ± 0%  -1.99%  (p=0.016 n=5+4)
FormatUint-4                             1.89µs ± 1%  1.83µs ± 1%  -2.94%  (p=0.008 n=5+5)
AppendUint-4                             1.59µs ± 0%  1.57µs ± 2%  -1.72%  (p=0.040 n=5+5)
FormatIntSmall-4                         8.48ns ± 0%  8.48ns ± 0%    ~     (p=0.905 n=5+5)
AppendIntSmall-4                         12.2ns ± 0%  12.2ns ± 0%    ~     (all equal)
AppendUintVarlen/1-4                     10.6ns ± 1%  10.7ns ± 0%    ~     (p=0.238 n=5+4)
AppendUintVarlen/12-4                    10.7ns ± 0%  10.7ns ± 1%    ~     (p=0.333 n=4+5)
AppendUintVarlen/123-4                   29.9ns ± 1%  30.2ns ± 0%  +1.07%  (p=0.016 n=5+4)
AppendUintVarlen/1234-4                  32.4ns ± 1%  30.4ns ± 0%  -6.30%  (p=0.008 n=5+5)
AppendUintVarlen/12345-4                 35.1ns ± 2%  34.9ns ± 0%    ~     (p=0.238 n=5+5)
AppendUintVarlen/123456-4                36.6ns ± 0%  35.3ns ± 0%  -3.55%  (p=0.029 n=4+4)
AppendUintVarlen/1234567-4               38.9ns ± 0%  39.6ns ± 0%  +1.80%  (p=0.029 n=4+4)
AppendUintVarlen/12345678-4              41.3ns ± 0%  40.1ns ± 0%  -2.91%  (p=0.000 n=5+4)
AppendUintVarlen/123456789-4             44.9ns ± 1%  44.8ns ± 0%    ~     (p=0.667 n=5+5)
AppendUintVarlen/1234567890-4            65.6ns ± 0%  66.2ns ± 1%  +0.88%  (p=0.016 n=4+5)
AppendUintVarlen/12345678901-4           77.9ns ± 0%  76.3ns ± 0%  -2.00%  (p=0.000 n=4+5)
AppendUintVarlen/123456789012-4          80.7ns ± 0%  79.1ns ± 1%  -2.01%  (p=0.008 n=5+5)
AppendUintVarlen/1234567890123-4         83.6ns ± 0%  80.2ns ± 1%  -4.07%  (p=0.008 n=5+5)
AppendUintVarlen/12345678901234-4        86.2ns ± 1%  83.3ns ± 0%  -3.39%  (p=0.008 n=5+5)
AppendUintVarlen/123456789012345-4       88.5ns ± 0%  83.7ns ± 0%  -5.42%  (p=0.008 n=5+5)
AppendUintVarlen/1234567890123456-4      90.6ns ± 0%  88.3ns ± 0%  -2.54%  (p=0.008 n=5+5)
AppendUintVarlen/12345678901234567-4     92.7ns ± 0%  89.0ns ± 1%  -4.01%  (p=0.008 n=5+5)
AppendUintVarlen/123456789012345678-4    95.6ns ± 1%  92.6ns ± 0%  -3.18%  (p=0.016 n=5+4)
AppendUintVarlen/1234567890123456789-4    118ns ± 0%   114ns ± 0%    ~     (p=0.079 n=4+5)
AppendUintVarlen/12345678901234567890-4   138ns ± 0%   136ns ± 0%  -1.45%  (p=0.008 n=5+5)

Updates #19445

Change-Id: Iafbe5c074898187c150dc3854e5b9fc19c10be05
Reviewed-on: https://go-review.googlesource.com/38255
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/internal/obj: clean up checkaddr
Josh Bleecher Snyder [Wed, 22 Mar 2017 04:51:10 +0000 (21:51 -0700)]
cmd/internal/obj: clean up checkaddr

Coalesce identical cases.
Give it a proper doc comment.
Fix comment locations.
Update/delete old comments.

Passes toolstash-check -all.

Change-Id: I88d9cf20e6e04b0c1c6583e92cd96335831f183f
Reviewed-on: https://go-review.googlesource.com/38442
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/internal/obj: eliminate AMODE
Josh Bleecher Snyder [Wed, 22 Mar 2017 05:21:33 +0000 (22:21 -0700)]
cmd/internal/obj: eliminate AMODE

AMODE appears to have been intended to allow
a Prog to switch between 16 (!), 32, or 64 bit x86.
It is unused anywhere in the tree.

Passes toolstash-check -all.

Updates #15756

Change-Id: Ic57b257cfe580f29dad81d97e4193bf3c330c598
Reviewed-on: https://go-review.googlesource.com/38445
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/internal/obj: eliminate unnecessary ctxt.Cursym assignment
Josh Bleecher Snyder [Wed, 22 Mar 2017 05:07:23 +0000 (22:07 -0700)]
cmd/internal/obj: eliminate unnecessary ctxt.Cursym assignment

None of the following code uses it.

Passes toolstash-check -all.

Updates #15756

Change-Id: Ieeaaca8ba31e5c345c0c8a758d520b24be88e173
Reviewed-on: https://go-review.googlesource.com/38444
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
8 years agocmd/doc: implement "go doc struct.field"
Rob Pike [Wed, 22 Mar 2017 03:27:47 +0000 (20:27 -0700)]
cmd/doc: implement "go doc struct.field"

By analogy with the handling of methods on types, show the documentation
for a single field of a struct.

% go doc ast.structtype.fields
struct StructType {
    Fields *FieldList  // list of field declarations
}
%

Fixes #19169.

Change-Id: I002f992e4aa64bee667e2e4bccc7082486149842
Reviewed-on: https://go-review.googlesource.com/38438
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agotest/bench/go1: fix typo in parserdata_test.go comment
Jason Travis [Wed, 22 Mar 2017 03:12:47 +0000 (20:12 -0700)]
test/bench/go1: fix typo in parserdata_test.go comment

Change-Id: Iaca02660bdc8262db2b003a94aca661b5cec5576
Reviewed-on: https://go-review.googlesource.com/38437
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: disable typPtr caching in the backend
Josh Bleecher Snyder [Sat, 18 Mar 2017 15:11:49 +0000 (08:11 -0700)]
cmd/compile: disable typPtr caching in the backend

The only new Types that the backend introduces
are pointers to Types generated by the frontend.
Usually, when we generate a *T,
we cache the resulting Type in T,
to avoid recreating it later.
However, that caching is not concurrency safe.
Rather than add mutexes, this CL disables that
caching before starting the backend.
The backend generates few enough new *Ts that the
performance impact of this is small, particularly
if we pre-create some commonly used *Ts.

Updates #15756

name       old alloc/op    new alloc/op    delta
Template      40.3MB ± 0%     40.4MB ± 0%  +0.18%  (p=0.001 n=10+10)
Unicode       29.8MB ± 0%     29.8MB ± 0%  +0.11%  (p=0.043 n=10+9)
GoTypes        114MB ± 0%      115MB ± 0%  +0.33%  (p=0.000 n=9+10)
SSA            855MB ± 0%      859MB ± 0%  +0.40%  (p=0.000 n=10+10)
Flate         25.7MB ± 0%     25.8MB ± 0%  +0.35%  (p=0.000 n=10+10)
GoParser      31.9MB ± 0%     32.1MB ± 0%  +0.58%  (p=0.000 n=10+10)
Reflect       79.6MB ± 0%     79.9MB ± 0%  +0.31%  (p=0.000 n=10+10)
Tar           26.9MB ± 0%     26.9MB ± 0%  +0.21%  (p=0.000 n=10+10)
XML           42.5MB ± 0%     42.7MB ± 0%  +0.52%  (p=0.000 n=10+9)

name       old allocs/op   new allocs/op   delta
Template        394k ± 1%       393k ± 0%    ~     (p=0.529 n=10+10)
Unicode         319k ± 1%       319k ± 0%    ~     (p=0.720 n=10+9)
GoTypes        1.15M ± 0%      1.15M ± 0%  +0.14%  (p=0.035 n=10+10)
SSA            7.53M ± 0%      7.56M ± 0%  +0.45%  (p=0.000 n=9+10)
Flate           238k ± 0%       238k ± 1%    ~     (p=0.579 n=10+10)
GoParser        318k ± 1%       320k ± 1%  +0.64%  (p=0.001 n=10+10)
Reflect        1.00M ± 0%      1.00M ± 0%    ~     (p=0.393 n=10+10)
Tar             254k ± 0%       254k ± 1%    ~     (p=0.075 n=10+10)
XML             395k ± 0%       397k ± 0%  +0.44%  (p=0.001 n=10+9)

Change-Id: I6c031ed4f39108f26969c5712b73aa2fc08cd10a
Reviewed-on: https://go-review.googlesource.com/38417
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agosyscall, os: fix FreeBSD 9 build
Brad Fitzpatrick [Tue, 21 Mar 2017 22:16:16 +0000 (22:16 +0000)]
syscall, os: fix FreeBSD 9 build

I broke FreeBSD 9 in https://golang.org/cl/38426 by using Pipe2.

We still want to support FreeBSD 9 for one last release (Go 1.9 will
be the last), and FreeBSD 9 doesn't have Pipe2.

So this still uses Pipe2, but falls back to Pipe on error.

Updates #18854
Updates #19072

Change-Id: I1de90fb83606c93fb84b4b86fba31e207a702835
Reviewed-on: https://go-review.googlesource.com/38430
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/gc: remove unneeded effects cache fields
Matthew Dempsky [Tue, 21 Mar 2017 20:14:35 +0000 (13:14 -0700)]
cmd/compile/internal/gc: remove unneeded effects cache fields

Must have been lost when rebasing the SSA liveness CLs.

Change-Id: Iaac33158cc7c92ea44a023c242eb914a7d6979c6
Reviewed-on: https://go-review.googlesource.com/38427
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agobytes: add optimized countByte for amd64
Josselin Costanzi [Sun, 19 Mar 2017 11:18:08 +0000 (12:18 +0100)]
bytes: add optimized countByte for amd64

Use SSE/AVX2 when counting a single byte.
Inspired from runtime indexbyte implementation.

Benchmark against previous implementation, where
1 byte in every 8 is the one we are looking for:

* On a machine without AVX2
name               old time/op   new time/op     delta
CountSingle/10-4    61.8ns ±10%     15.6ns ±11%    -74.83%  (p=0.000 n=10+10)
CountSingle/32-4     100ns ± 4%       17ns ±10%    -82.54%  (p=0.000 n=10+9)
CountSingle/4K-4    9.66µs ± 3%     0.37µs ± 6%    -96.21%  (p=0.000 n=10+10)
CountSingle/4M-4    11.0ms ± 6%      0.4ms ± 4%    -96.04%  (p=0.000 n=10+10)
CountSingle/64M-4    194ms ± 8%        8ms ± 2%    -95.64%  (p=0.000 n=10+10)

name               old speed     new speed       delta
CountSingle/10-4   162MB/s ±10%    645MB/s ±10%   +297.00%  (p=0.000 n=10+10)
CountSingle/32-4   321MB/s ± 5%   1844MB/s ± 9%   +474.79%  (p=0.000 n=10+9)
CountSingle/4K-4   424MB/s ± 3%  11169MB/s ± 6%  +2533.10%  (p=0.000 n=10+10)
CountSingle/4M-4   381MB/s ± 7%   9609MB/s ± 4%  +2421.88%  (p=0.000 n=10+10)
CountSingle/64M-4  346MB/s ± 7%   7924MB/s ± 2%  +2188.78%  (p=0.000 n=10+10)

* On a machine with AVX2
name               old time/op   new time/op     delta
CountSingle/10-8    37.1ns ± 3%      8.2ns ± 1%    -77.80%  (p=0.000 n=10+10)
CountSingle/32-8    66.1ns ± 3%      9.8ns ± 2%    -85.23%  (p=0.000 n=10+10)
CountSingle/4K-8    7.36µs ± 3%     0.11µs ± 1%    -98.54%  (p=0.000 n=10+10)
CountSingle/4M-8    7.46ms ± 2%     0.15ms ± 2%    -97.95%  (p=0.000 n=10+9)
CountSingle/64M-8    124ms ± 2%        6ms ± 4%    -95.09%  (p=0.000 n=10+10)

name               old speed     new speed       delta
CountSingle/10-8   269MB/s ± 3%   1213MB/s ± 1%   +350.32%  (p=0.000 n=10+10)
CountSingle/32-8   484MB/s ± 4%   3277MB/s ± 2%   +576.66%  (p=0.000 n=10+10)
CountSingle/4K-8   556MB/s ± 3%  37933MB/s ± 1%  +6718.36%  (p=0.000 n=10+10)
CountSingle/4M-8   562MB/s ± 2%  27444MB/s ± 3%  +4783.43%  (p=0.000 n=10+9)
CountSingle/64M-8  543MB/s ± 2%  11054MB/s ± 3%  +1935.81%  (p=0.000 n=10+10)

Fixes #19411

Change-Id: Ieaf20b1fabccabe767c55c66e242e86f3617f883
Reviewed-on: https://go-review.googlesource.com/38258
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agosyscall, os: use pipe2 syscall on FreeBSD instead of pipe
Brad Fitzpatrick [Tue, 21 Mar 2017 17:23:08 +0000 (10:23 -0700)]
syscall, os: use pipe2 syscall on FreeBSD instead of pipe

The pipe2 syscall exists in all officially supported FreeBSD
versions: 10, 11 and future 12.
The pipe syscall no longer exists in 11 and 12. To build and
run Go on these versions, kernel needs COMPAT_FREEBSD10 option.

Based on Gleb Smirnoff's https://golang.org/cl/38422

Fixes #18854

Change-Id: I8e201ee1b15dca10427c3093b966025d160aaf61
Reviewed-on: https://go-review.googlesource.com/38426
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: add regress test for issue 19632
Matthew Dempsky [Tue, 21 Mar 2017 19:38:23 +0000 (12:38 -0700)]
cmd/compile: add regress test for issue 19632

Updates #19632.

Change-Id: I1411dd997c8c6a789d17d0dcc0bfbd2281447b16
Reviewed-on: https://go-review.googlesource.com/38401
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: remove function-name-based debuglive tweaks
Josh Bleecher Snyder [Tue, 21 Mar 2017 16:57:47 +0000 (09:57 -0700)]
cmd/compile: remove function-name-based debuglive tweaks

It's easier to grep output than recompile the compiler anyway.

For concurrent compilation.

Updates #15756

Change-Id: I151cb5dc77056469cd9019d516f86454e931a197
Reviewed-on: https://go-review.googlesource.com/38424
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: eliminate a Curfn reference in plive
Josh Bleecher Snyder [Tue, 21 Mar 2017 19:13:12 +0000 (12:13 -0700)]
cmd/compile: eliminate a Curfn reference in plive

I think this got lost in a rebase somewhere.

Updates #15756

Change-Id: Ia3e7c60d1b9254f2877217073732b46c91059ade
Reviewed-on: https://go-review.googlesource.com/38425
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/gc: fix liveness regression
Matthew Dempsky [Tue, 21 Mar 2017 17:15:14 +0000 (10:15 -0700)]
cmd/compile/internal/gc: fix liveness regression

During AllocFrame, we drop unused variables from Curfn.Func.Dcl, but
there might still be OpVarFoo instructions that reference those
variables. This wasn't a problem though because gvardefx used to emit
ANOP for unused variables instead of AVARFOO.

As an easy fix, if we see OpVarFoo (or OpKeepAlive) referencing an
unused variable, we can ignore it.

Fixes #19632.

Change-Id: I4e9ffabdb4058f7cdcc4663b540f5a5a692daf8b
Reviewed-on: https://go-review.googlesource.com/38400
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agoruntime: remove unused *chantype parameters
Daniel Martí [Sat, 18 Mar 2017 15:55:41 +0000 (15:55 +0000)]
runtime: remove unused *chantype parameters

The chanrecv funcs don't use it at all. The chansend ones do, but the
element type is now part of the hchan struct, which is already a
parameter.

hchan can be nil in chansend when sending to a nil channel, so when
instrumenting we must copy to the stack to be able to read the channel
type.

name             old time/op  new time/op  delta
ChanUncontended  6.42µs ± 1%  6.22µs ± 0%  -3.06%  (p=0.000 n=19+18)

Initially found by github.com/mvdan/unparam.

Fixes #19591.

Change-Id: I3a5e8a0082e8445cc3f0074695e3593fd9c88412
Reviewed-on: https://go-review.googlesource.com/38351
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoflag: validate Int and Uint values to be in range
Brad Fitzpatrick [Tue, 21 Mar 2017 06:34:35 +0000 (06:34 +0000)]
flag: validate Int and Uint values to be in range

Fixes #19230

Change-Id: I38df9732b88f0328506e74f1a46f52adf47db1e5
Reviewed-on: https://go-review.googlesource.com/38419
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile,runtime: fix atomic And8 for mipsle
Vladimir Stefanovic [Tue, 21 Mar 2017 13:57:58 +0000 (14:57 +0100)]
cmd/compile,runtime: fix atomic And8 for mipsle

Removing stray xori that came from big endian copy/paste.
Adding atomicand8 check to runtime.check() that would have revealed
this error.
Might fix #19396.

Change-Id: If8d6f25d3e205496163541eb112548aa66df9c2a
Reviewed-on: https://go-review.googlesource.com/38257
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/compile: improve LoweredZero performance for ppc64x
Lynn Boger [Mon, 13 Mar 2017 14:16:30 +0000 (10:16 -0400)]
cmd/compile: improve LoweredZero performance for ppc64x

This change improves the performance of the LoweredZero rule
on ppc64x.

The improvement can be seen in the runtime ClearFat
benchmarks:

BenchmarkClearFat12-16       2.40          0.69          -71.25%
BenchmarkClearFat16-16       9.98          0.93          -90.68%
BenchmarkClearFat24-16       4.75          0.93          -80.42%
BenchmarkClearFat32-16       6.02          0.93          -84.55%
BenchmarkClearFat40-16       7.19          1.16          -83.87%
BenchmarkClearFat48-16       15.0          1.39          -90.73%
BenchmarkClearFat56-16       9.95          1.62          -83.72%
BenchmarkClearFat64-16       18.0          1.86          -89.67%
BenchmarkClearFat128-16      30.0          8.08          -73.07%
BenchmarkClearFat256-16      52.5          11.3          -78.48%
BenchmarkClearFat512-16      97.0          19.0          -80.41%
BenchmarkClearFat1024-16     244           34.2          -85.98%

Fixes: #19532
Change-Id: If493e28bc1d8e61bc79978498be9f5336a36cd3f
Reviewed-on: https://go-review.googlesource.com/38096
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <munday@ca.ibm.com>
8 years agonet/http/cookiejar: fix out-of-bounds errors on malformed domains
Volker Dobler [Mon, 6 Mar 2017 09:07:25 +0000 (10:07 +0100)]
net/http/cookiejar: fix out-of-bounds errors on malformed domains

The old implementation of Jar made the assumption that the host names
in the URLs given to SetCookies() and Cookies() methods are well-formed.
This is not an unreasonable assumption as malformed host names do not
trigger calls to SetCookies or Cookies (at least not from net/http)
as the HTTP request themselves are not executed. But there can be other
invocations of these methods and at least on Linux it was possible to
make DNS lookup to domain names with two trailing dots (see issue #7122).

This is an old bug and this CL revives an old change (see
https://codereview.appspot.com/52100043) to fix the issue. The discussion
around 52100043 focused on the interplay between the jar and the
public suffix list and who is responsible for which type if domain name
canonicalization. The new bug report in issue #19384 used a nil public
suffix list which demonstrates that the package cookiejar alone exhibits
this problem and any solution cannot be fully delegated to the
implementation of the used PublicSuffixList: Package cookiejar itself
needs to protect against host names of the form ".." which triggered an
out-of-bounds error.

This CL does not address the issue of host name canonicalization and
the question who is responsible for it. This CL just prevents the
out-of-bounds error: It is a very conservative change, i.e. one might
still set and retrieve cookies for host names like "weird.stuf...".
Several more test cases document how the current code works.

Fixes #19384.

Change-Id: I14be080e8a2a0b266ced779f2aeb18841b730610
Reviewed-on: https://go-review.googlesource.com/37843
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
8 years agoruntime: add mapdelete_fast*
Hugues Bruant [Tue, 14 Mar 2017 18:11:28 +0000 (11:11 -0700)]
runtime: add mapdelete_fast*

Add benchmarks for map delete with int32/int64/string key

Benchmark results on darwin/amd64

name                 old time/op  new time/op  delta
MapDelete/Int32/1-8   151ns ± 8%    99ns ± 3%  -34.39%  (p=0.008 n=5+5)
MapDelete/Int32/2-8   128ns ± 2%   111ns ±15%  -13.40%  (p=0.040 n=5+5)
MapDelete/Int32/4-8   128ns ± 5%   114ns ± 2%  -10.82%  (p=0.008 n=5+5)
MapDelete/Int64/1-8   144ns ± 0%   104ns ± 3%  -27.53%  (p=0.016 n=4+5)
MapDelete/Int64/2-8   153ns ± 1%   126ns ± 3%  -17.17%  (p=0.008 n=5+5)
MapDelete/Int64/4-8   178ns ± 3%   136ns ± 2%  -23.60%  (p=0.008 n=5+5)
MapDelete/Str/1-8     187ns ± 3%   171ns ± 3%   -8.54%  (p=0.008 n=5+5)
MapDelete/Str/2-8     221ns ± 3%   206ns ± 4%   -7.18%  (p=0.016 n=5+4)
MapDelete/Str/4-8     256ns ± 5%   232ns ± 2%   -9.36%  (p=0.016 n=4+5)

name                     old time/op    new time/op    delta
BinaryTree17-8              2.78s ± 7%     2.70s ± 1%    ~     (p=0.151 n=5+5)
Fannkuch11-8                3.21s ± 2%     3.19s ± 1%    ~     (p=0.310 n=5+5)
FmtFprintfEmpty-8          49.1ns ± 3%    50.2ns ± 2%    ~     (p=0.095 n=5+5)
FmtFprintfString-8         78.6ns ± 4%    80.2ns ± 5%    ~     (p=0.460 n=5+5)
FmtFprintfInt-8            79.7ns ± 1%    81.0ns ± 3%    ~     (p=0.103 n=5+5)
FmtFprintfIntInt-8          117ns ± 2%     119ns ± 0%    ~     (p=0.079 n=5+4)
FmtFprintfPrefixedInt-8     153ns ± 1%     146ns ± 3%  -4.19%  (p=0.024 n=5+5)
FmtFprintfFloat-8           239ns ± 1%     237ns ± 1%    ~     (p=0.246 n=5+5)
FmtManyArgs-8               506ns ± 2%     509ns ± 2%    ~     (p=0.238 n=5+5)
GobDecode-8                7.06ms ± 4%    6.86ms ± 1%    ~     (p=0.222 n=5+5)
GobEncode-8                6.01ms ± 5%    5.87ms ± 2%    ~     (p=0.222 n=5+5)
Gzip-8                      246ms ± 4%     236ms ± 1%  -4.12%  (p=0.008 n=5+5)
Gunzip-8                   37.7ms ± 4%    37.3ms ± 1%    ~     (p=0.841 n=5+5)
HTTPClientServer-8         64.9µs ± 1%    64.4µs ± 0%  -0.80%  (p=0.032 n=5+4)
JSONEncode-8               16.0ms ± 2%    16.2ms ±11%    ~     (p=0.548 n=5+5)
JSONDecode-8               53.2ms ± 2%    53.1ms ± 4%    ~     (p=1.000 n=5+5)
Mandelbrot200-8            4.33ms ± 2%    4.32ms ± 2%    ~     (p=0.841 n=5+5)
GoParse-8                  3.24ms ± 2%    3.27ms ± 4%    ~     (p=0.690 n=5+5)
RegexpMatchEasy0_32-8      86.2ns ± 1%    85.2ns ± 3%    ~     (p=0.286 n=5+5)
RegexpMatchEasy0_1K-8       198ns ± 2%     199ns ± 1%    ~     (p=0.310 n=5+5)
RegexpMatchEasy1_32-8      82.6ns ± 2%    81.8ns ± 1%    ~     (p=0.294 n=5+5)
RegexpMatchEasy1_1K-8       359ns ± 2%     354ns ± 1%  -1.39%  (p=0.048 n=5+5)
RegexpMatchMedium_32-8      123ns ± 2%     123ns ± 1%    ~     (p=0.905 n=5+5)
RegexpMatchMedium_1K-8     38.2µs ± 2%    38.6µs ± 8%    ~     (p=0.690 n=5+5)
RegexpMatchHard_32-8       1.92µs ± 2%    1.91µs ± 5%    ~     (p=0.460 n=5+5)
RegexpMatchHard_1K-8       57.6µs ± 1%    57.0µs ± 2%    ~     (p=0.310 n=5+5)
Revcomp-8                   483ms ± 7%     441ms ± 1%  -8.79%  (p=0.016 n=5+4)
Template-8                 58.0ms ± 1%    58.2ms ± 7%    ~     (p=0.310 n=5+5)
TimeParse-8                 324ns ± 6%     312ns ± 2%    ~     (p=0.087 n=5+5)
TimeFormat-8                330ns ± 1%     329ns ± 1%    ~     (p=0.968 n=5+5)

name                     old speed      new speed      delta
GobDecode-8               109MB/s ± 4%   112MB/s ± 1%    ~     (p=0.222 n=5+5)
GobEncode-8               128MB/s ± 5%   131MB/s ± 2%    ~     (p=0.222 n=5+5)
Gzip-8                   78.9MB/s ± 4%  82.3MB/s ± 1%  +4.25%  (p=0.008 n=5+5)
Gunzip-8                  514MB/s ± 4%   521MB/s ± 1%    ~     (p=0.841 n=5+5)
JSONEncode-8              121MB/s ± 2%   120MB/s ±10%    ~     (p=0.548 n=5+5)
JSONDecode-8             36.5MB/s ± 2%  36.6MB/s ± 4%    ~     (p=1.000 n=5+5)
GoParse-8                17.9MB/s ± 2%  17.7MB/s ± 4%    ~     (p=0.730 n=5+5)
RegexpMatchEasy0_32-8     371MB/s ± 1%   375MB/s ± 3%    ~     (p=0.310 n=5+5)
RegexpMatchEasy0_1K-8    5.15GB/s ± 1%  5.13GB/s ± 1%    ~     (p=0.548 n=5+5)
RegexpMatchEasy1_32-8     387MB/s ± 2%   391MB/s ± 1%    ~     (p=0.310 n=5+5)
RegexpMatchEasy1_1K-8    2.85GB/s ± 2%  2.89GB/s ± 1%    ~     (p=0.056 n=5+5)
RegexpMatchMedium_32-8   8.07MB/s ± 2%  8.06MB/s ± 1%    ~     (p=0.730 n=5+5)
RegexpMatchMedium_1K-8   26.8MB/s ± 2%  26.6MB/s ± 7%    ~     (p=0.690 n=5+5)
RegexpMatchHard_32-8     16.7MB/s ± 2%  16.7MB/s ± 5%    ~     (p=0.421 n=5+5)
RegexpMatchHard_1K-8     17.8MB/s ± 1%  18.0MB/s ± 2%    ~     (p=0.310 n=5+5)
Revcomp-8                 527MB/s ± 6%   577MB/s ± 1%  +9.44%  (p=0.016 n=5+4)
Template-8               33.5MB/s ± 1%  33.4MB/s ± 7%    ~     (p=0.310 n=5+5)

Updates #19495

Change-Id: Ib9ece1690813d9b4788455db43d30891e2138df5
Reviewed-on: https://go-review.googlesource.com/38172
Reviewed-by: Hugues Bruant <hugues.bruant@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: import os package in BenchmarkRunningGoProgram
Alex Brainman [Wed, 8 Mar 2017 03:58:41 +0000 (14:58 +1100)]
runtime: import os package in BenchmarkRunningGoProgram

I would like to use BenchmarkRunningGoProgram to measure
changes for issue #15588. So the program in the benchmark
should import "os" package.

It is also reasonable that basic Go program includes
"os" package.

For #15588.

Change-Id: Ida6712eab22c2e79fbe91b6fdd492eaf31756852
Reviewed-on: https://go-review.googlesource.com/37914
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/compile: fix pos of typenames created during SSA construction
Josh Bleecher Snyder [Tue, 21 Mar 2017 02:00:14 +0000 (19:00 -0700)]
cmd/compile: fix pos of typenames created during SSA construction

Prior to this CL, the function's position was used.
The dottype Node's position is clearly better.

I'm not thrilled about introducing a reference to
lineno in the middle of SSA construction;
I will have to remove it later.
My immediate goal is stability and correctness of positions,
though, since that aids refactoring, so this is an improvement.

An example from package io:

func (t *multiWriter) WriteString(s string) (n int, err error) {
var p []byte // lazily initialized if/when needed
for _, w := range t.writers {
if sw, ok := w.(stringWriter); ok {
n, err = sw.WriteString(s)

The w.(stringWriter) type assertion includes loading
the address of static type data for stringWriter:

LEAQ type."".stringWriter(SB), R10

Prior to this CL, this instruction was given the line number
of the function declaration.
After this CL, this instruction is given the line number
of the type assertion itself.

Change-Id: Ifcca274b581a5a57d7e3102c4d7b7786bf307210
Reviewed-on: https://go-review.googlesource.com/38389
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoencoding/gob: document the extra byte after a singleton
Rob Pike [Tue, 21 Mar 2017 03:24:26 +0000 (20:24 -0700)]
encoding/gob: document the extra byte after a singleton

This paragraph has been added, as the notion was missing from the
documentation.

If a value is passed to Encode and the type is not a struct (or pointer to struct,
etc.), for simplicity of processing it is represented as a struct of one field.
The only visible effect of this is to encode a zero byte after the value, just as
after the last field of an encoded struct, so that the decode algorithm knows when
the top-level value is complete.

Fixes #16978

Change-Id: I5f008e792d1b6fe80d2e026a7ff716608889db32
Reviewed-on: https://go-review.googlesource.com/38414
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agotext/template,html/template: state that Funcs must happen before parsing
Rob Pike [Tue, 21 Mar 2017 03:18:02 +0000 (20:18 -0700)]
text/template,html/template: state that Funcs must happen before parsing

Any method that affects the parse must happen before parsing.
This obvious point is clear, but it's not clear to some that the
set of defined functions affect the parse.

Fixes #18971

Change-Id: I8b7f8c8cf85b028c18e5ca3b9797de92ea910669
Reviewed-on: https://go-review.googlesource.com/38413
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agocmd/compile: check for missing function body earlier
Josh Bleecher Snyder [Mon, 20 Mar 2017 20:16:07 +0000 (13:16 -0700)]
cmd/compile: check for missing function body earlier

Tested by fixedbugs/issue3705.go.

This removes a dependency on lineno
from near the backend.

Change-Id: I228bd0ad7295cf881b9bdeb0df9d18483fb96821
Reviewed-on: https://go-review.googlesource.com/38382
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: use autogenerated position for init functions
Josh Bleecher Snyder [Mon, 20 Mar 2017 22:32:00 +0000 (15:32 -0700)]
cmd/compile: use autogenerated position for init functions

This eliminates an old TODO,
and stabilizes the position information
for init functions.

Change-Id: Idf2d9a16a60e097ee08f42541b87e170da2f9d3a
Reviewed-on: https://go-review.googlesource.com/38388
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agogo/build: remove unused returnImports parameter
Daniel Martí [Fri, 17 Mar 2017 16:05:31 +0000 (16:05 +0000)]
go/build: remove unused returnImports parameter

The code uses the filename suffix instead of the bool parameter to
determine what to do.

Fixes #19474.

Change-Id: Ic552a54e50194592a4b4ae7f74d3109af54e6d36
Reviewed-on: https://go-review.googlesource.com/38265
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
8 years agocmd/compile/internal/gc: remove unneeded Type.Pos field
Matthew Dempsky [Mon, 20 Mar 2017 19:27:25 +0000 (12:27 -0700)]
cmd/compile/internal/gc: remove unneeded Type.Pos field

Change-Id: I9ab650d9d2d0a99186009362454e1eabc9f6bad6
Reviewed-on: https://go-review.googlesource.com/38393
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile/internal/gc: export interface embedding information
Matthew Dempsky [Mon, 20 Mar 2017 19:14:16 +0000 (12:14 -0700)]
cmd/compile/internal/gc: export interface embedding information

Fixes #16369.

Change-Id: I23f8c36370d0da37ac5b5126d012d22f78782782
Reviewed-on: https://go-review.googlesource.com/38392
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/compile/internal/gc: handle recursive interfaces better
Matthew Dempsky [Mon, 20 Mar 2017 18:56:15 +0000 (11:56 -0700)]
cmd/compile/internal/gc: handle recursive interfaces better

Previously, we handled recursive interfaces by deferring typechecking
of interface methods, while eagerly expanding interface embeddings.

This CL switches to eagerly evaluating interface methods, and
deferring expanding interface embeddings to dowidth. This allows us to
detect recursive interface embeddings with the same mechanism used for
detecting recursive struct embeddings.

Updates #16369.

Change-Id: If4c0320058047f8a2d9b52b9a79de47eb9887f95
Reviewed-on: https://go-review.googlesource.com/38391
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agonet/http: fix ProxyFromEnvironment panic on invalid $NO_PROXY value
Jakob Borg [Mon, 13 Mar 2017 23:21:51 +0000 (08:21 +0900)]
net/http: fix ProxyFromEnvironment panic on invalid $NO_PROXY value

Given an entry in $no_proxy like ":1" we would interpret it as an empty
host name and a port number, then check the first character of the host
name for dots. This would then cause an index out of range panic. This
change simply skips these entries, as the following checks would anyway
have returned false.

Fixes #19536

Change-Id: Iafe9c7a77ad4a6278c8ccb00a1575b56e4bdcd79
Reviewed-on: https://go-review.googlesource.com/38067
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoencoding/json: reduce unmarshal mallocs for unmapped fields
Pascal S. de Kloe [Wed, 16 Nov 2016 15:35:42 +0000 (16:35 +0100)]
encoding/json: reduce unmarshal mallocs for unmapped fields

JSON decoding performs poorly for unmapped and ignored fields. We noticed better
performance when unmarshalling unused fields. The loss comes mostly from calls
to scanner.error as described at #17914.

benchmark                 old ns/op     new ns/op     delta
BenchmarkIssue10335-8     431           408           -5.34%
BenchmarkUnmapped-8       1744          1314          -24.66%

benchmark                 old allocs     new allocs     delta
BenchmarkIssue10335-8     4              3              -25.00%
BenchmarkUnmapped-8       18             4              -77.78%

benchmark                 old bytes     new bytes     delta
BenchmarkIssue10335-8     320           312           -2.50%
BenchmarkUnmapped-8       568           344           -39.44%

Fixes #17914, improves #10335

Change-Id: I7d4258a94eb287c0fe49e7334795209b90434cd0
Reviewed-on: https://go-review.googlesource.com/33276
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/internal/obj: remove unneeded Addr.Node and Prog.Opt fields
Matthew Dempsky [Sun, 12 Mar 2017 01:17:15 +0000 (17:17 -0800)]
cmd/internal/obj: remove unneeded Addr.Node and Prog.Opt fields

Change-Id: I218b241c32a5948b66ad0d95ecc368648cf4ddf5
Reviewed-on: https://go-review.googlesource.com/38130
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/internal/obj: remove unneeded AVARFOO ops
Matthew Dempsky [Fri, 10 Mar 2017 23:52:26 +0000 (15:52 -0800)]
cmd/internal/obj: remove unneeded AVARFOO ops

Change-Id: I10e36046ebce8a8741ef019cfe266b9ac9fa322d
Reviewed-on: https://go-review.googlesource.com/38088
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: remove ProgInfo tables
Matthew Dempsky [Fri, 10 Mar 2017 23:03:25 +0000 (15:03 -0800)]
cmd/compile: remove ProgInfo tables

Change-Id: Id807c702ad71edddd23f2eb6f5e69e9a62e60bcd
Reviewed-on: https://go-review.googlesource.com/38089
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agoruntime: clear signal stack on main thread
Ian Lance Taylor [Fri, 17 Mar 2017 20:48:56 +0000 (13:48 -0700)]
runtime: clear signal stack on main thread

This is a workaround for a FreeBSD kernel bug. It can be removed when
we are confident that all people are using the fixed kernel. See #15658.

Updates #15658.

Change-Id: I0ecdccb77ddd0c270bdeac4d3a5c8abaf0449075
Reviewed-on: https://go-review.googlesource.com/38325
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: port liveness analysis to SSA
Matthew Dempsky [Fri, 10 Mar 2017 02:32:17 +0000 (18:32 -0800)]
cmd/compile: port liveness analysis to SSA

Passes toolstash-check -all.

Change-Id: I92c3c25d6c053f971f346f4fa3bbc76419b58183
Reviewed-on: https://go-review.googlesource.com/38087
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: sort CFG blocks in PC order during liveness
Matthew Dempsky [Sat, 11 Mar 2017 00:03:56 +0000 (16:03 -0800)]
cmd/compile: sort CFG blocks in PC order during liveness

This CL changes the order that liveness analysis visits CFG blocks to
PC order, rather than RPO. This doesn't meaningfully change anything
except that the PCDATA_StackMapIndex values will be assigned in PC
order too.

However, this does have the benefit that the subsequent CL to port
liveness analysis to the SSA CFG (which has blocks in PC order) will
now pass toolstash-check.

Change-Id: I1de5a2eecb8027723a6e422d46186d0c63d48c8d
Reviewed-on: https://go-review.googlesource.com/38086
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/internal/obj: convert Debug* Link fields into bools
Josh Bleecher Snyder [Mon, 20 Mar 2017 22:01:20 +0000 (15:01 -0700)]
cmd/internal/obj: convert Debug* Link fields into bools

Change-Id: I9ac274dbfe887675a7820d2f8f87b5887b1c9b0e
Reviewed-on: https://go-review.googlesource.com/38383
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile/internal/gc: split SetInterface from SetFields
Matthew Dempsky [Mon, 20 Mar 2017 09:32:42 +0000 (02:32 -0700)]
cmd/compile/internal/gc: split SetInterface from SetFields

Change-Id: I4e568414faf64d3d47b1795382f0615f6caf53bc
Reviewed-on: https://go-review.googlesource.com/38390
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/internal/obj/s390x: cleanup objz.go
Michael Munday [Mon, 20 Mar 2017 21:31:35 +0000 (17:31 -0400)]
cmd/internal/obj/s390x: cleanup objz.go

This CL deletes some unnecessary code in objz.go that existed to
support instruction scheduling. It's likely instruction scheduling
will never be done in this part of the backend so this code can
just be deleted.

This file can probably be cleaned up a bit more, but I think this
is a good start.

Passes: go build -toolexec 'toolstash -cmp' -a std.

Change-Id: I1645632ac551a90a4f4be418045c046b488e9469
Reviewed-on: https://go-review.googlesource.com/38394
Run-TryBot: Michael Munday <munday@ca.ibm.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/ssa: recognize constant pointer comparison
philhofer [Wed, 15 Mar 2017 22:34:52 +0000 (15:34 -0700)]
cmd/compile/internal/ssa: recognize constant pointer comparison

Teach the backend to recognize that the address of a symbol
is equal with itself, and that the addresses of two different
symbols are different.

Some examples of where this rule hits in the standard library:

 - inlined uses of (*time.Time).setLoc (e.g. time.UTC)
 - inlined uses of bufio.NewReader (via type assertion)

Change-Id: I23dcb068c2ec333655c1292917bec13bbd908c24
Reviewed-on: https://go-review.googlesource.com/38338
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/go, cmd/compile: always optimize when building runtime
Josh Bleecher Snyder [Mon, 20 Mar 2017 16:56:50 +0000 (09:56 -0700)]
cmd/go, cmd/compile: always optimize when building runtime

When optimizations are disabled, the compiler
cannot eliminate enough write barriers to satisfy
the runtime's nowritebarrier and nowritebarrierrec
annotations.

Enforce that requirement, and for convenience,
have cmd/go elide -N when compiling the runtime.

This came up in practice for me when running
toolstash -cmp. When toolstash -cmp detected
mismatches, it recompiled with -N, which caused
runtime compilation failures.

Change-Id: Ifcdef22c725baf2c59a09470f00124361508a8f3
Reviewed-on: https://go-review.googlesource.com/38380
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/internal/obj: reduce variable scope
Josh Bleecher Snyder [Mon, 20 Mar 2017 05:31:02 +0000 (22:31 -0700)]
cmd/internal/obj: reduce variable scope

Minor cleanup, to make it clearer
that the two p's are unrelated.

Change-Id: Icb6386c626681f60e5e631b33aa3a0fc84f40e4a
Reviewed-on: https://go-review.googlesource.com/38381
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile/internal/gc: eliminate two uses of Type.Pos
Matthew Dempsky [Mon, 20 Mar 2017 09:21:50 +0000 (02:21 -0700)]
cmd/compile/internal/gc: eliminate two uses of Type.Pos

Instead we can use t.nod.Pos.

Change-Id: I643ee3226e402e38d4c77e8f328cbe83e55eac5c
Reviewed-on: https://go-review.googlesource.com/38309
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/compile: enable CSE of constant strings
Josh Bleecher Snyder [Wed, 31 Aug 2016 02:11:19 +0000 (19:11 -0700)]
cmd/compile: enable CSE of constant strings

CL 27254 changed a constant string to a byte array
in encoding/hex and got significant performance
improvements.

hex.Encode used the string twice in a single function.
The rewrite rules lower constant strings into components.
The pointer component requires an aux symbol.
The existing implementation created a new aux symbol every time.
As a result, constant string pointers were never CSE'd.
Tighten then moved the pointer calculation next to the uses, i.e.
into the loop.

The re-use of aux syms enabled by this CL
occurs 3691 times during make.bash.

This CL should not go in without CL 38338
or something like it.

Change-Id: Ibbf5b17283c0e31821d04c7e08d995c654de5663
Reviewed-on: https://go-review.googlesource.com/28219
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/gofmt: clarify doc string even more
Robert Griesemer [Mon, 20 Mar 2017 20:08:26 +0000 (13:08 -0700)]
cmd/gofmt: clarify doc string even more

Since "columns of alignment" are terminated whenever indentation
changes from one line to the next, alignment with spaces will work
independent of the actually chosen tab width. Don't mention tab width
anymore.

Follow-up on https://golang.org/cl/38374/.

For #19618.

Change-Id: I58e47dfde57834f56a98d9119670757a12fb9c41
Reviewed-on: https://go-review.googlesource.com/38379
Reviewed-by: Rob Pike <r@golang.org>
8 years agocmd/compile: don't permit declarations in post statement of for loop
Robert Griesemer [Mon, 20 Mar 2017 18:36:15 +0000 (11:36 -0700)]
cmd/compile: don't permit declarations in post statement of for loop

Report syntax error that was missed when moving to new parser.

Fixes #19610.

Change-Id: Ie5625f907a84089dc56fcccfd4f24df546042783
Reviewed-on: https://go-review.googlesource.com/38375
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: emit fused multiply-{add,subtract} on ppc64x
Michael Munday [Mon, 13 Mar 2017 18:39:17 +0000 (14:39 -0400)]
cmd/compile: emit fused multiply-{add,subtract} on ppc64x

A follow on to CL 36963 adding support for ppc64x.

Performance changes (as posted on the issue):

poly1305:
benchmark               old ns/op new ns/op delta
Benchmark64-16          172       151       -12.21%
Benchmark1K-16          1828      1523      -16.68%
Benchmark64Unaligned-16 172       151       -12.21%
Benchmark1KUnaligned-16 1827      1523      -16.64%

math:
BenchmarkAcos-16        43.9      39.9      -9.11%
BenchmarkAcosh-16       57.0      45.8      -19.65%
BenchmarkAsin-16        35.8      33.0      -7.82%
BenchmarkAsinh-16       68.6      60.8      -11.37%
BenchmarkAtan-16        19.8      16.2      -18.18%
BenchmarkAtanh-16       65.5      57.5      -12.21%
BenchmarkAtan2-16       45.4      34.2      -24.67%
BenchmarkGamma-16       37.6      26.0      -30.85%
BenchmarkLgamma-16      40.0      28.2      -29.50%
BenchmarkLog1p-16       35.1      29.1      -17.09%
BenchmarkSin-16         22.7      18.4      -18.94%
BenchmarkSincos-16      31.7      23.7      -25.24%
BenchmarkSinh-16        146       131       -10.27%
BenchmarkY0-16          130       107       -17.69%
BenchmarkY1-16          127       107       -15.75%
BenchmarkYn-16          278       235       -15.47%

Updates #17895.

Change-Id: I1c16199715d20c9c4bd97c4a950bcfa69eb688c1
Reviewed-on: https://go-review.googlesource.com/38095
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
8 years agocmd/gofmt: clarify documentation re: tab width
Robert Griesemer [Mon, 20 Mar 2017 17:58:40 +0000 (10:58 -0700)]
cmd/gofmt: clarify documentation re: tab width

Fixes #19618.

Change-Id: I0ac450ff717ec1f16eb12758c6bf5e98b5de20e8
Reviewed-on: https://go-review.googlesource.com/38374
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agonet/http: deflake TestServerAllowsBlockingRemoteAddr more
Brad Fitzpatrick [Mon, 20 Mar 2017 16:44:34 +0000 (16:44 +0000)]
net/http: deflake TestServerAllowsBlockingRemoteAddr more

As noted in https://github.com/golang/go/issues/19161#issuecomment-287554171,
CL 37771 (adding use of the new httptest.Server.Client to all net/http
tests) accidentally reverted DisableKeepAlives for this test. For
many tests, DisableKeepAlives was just present to prevent goroutines
from staying active after the test exited.  In this case it might
actually be important. (We'll see)

Updates #19161

Change-Id: I11f889f86c932b51b11846560b68dbe5993cdfc3
Reviewed-on: https://go-review.googlesource.com/38373
Reviewed-by: Michael Munday <munday@ca.ibm.com>
8 years agocrypto/*: fix spelling of 'below'
Martin Lindhe [Mon, 20 Mar 2017 11:14:47 +0000 (12:14 +0100)]
crypto/*: fix spelling of 'below'

Change-Id: Ic9d65206ec27f6d54bb71395802929e9c769e80a
Reviewed-on: https://go-review.googlesource.com/38355
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: disallow malloc or panic in scavenge
Austin Clements [Thu, 16 Mar 2017 21:02:24 +0000 (17:02 -0400)]
runtime: disallow malloc or panic in scavenge

Mallocs and panics in the scavenge path are particularly nasty because
they're likely to silently self-deadlock on the mheap.lock. Avoid
sinking lots of time into debugging these issues in the future by
turning these into immediate throws.

Change-Id: Ib36fdda33bc90b21c32432b03561630c1f3c69bc
Reviewed-on: https://go-review.googlesource.com/38293
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agoruntime: introduce a type for lfstacks
Austin Clements [Tue, 7 Mar 2017 21:38:29 +0000 (16:38 -0500)]
runtime: introduce a type for lfstacks

The lfstack API is still a C-style API: lfstacks all have unhelpful
type uint64 and the APIs are package-level functions. Make the code
more readable and Go-style by creating an lfstack type with methods
for push, pop, and empty.

Change-Id: I64685fa3be0e82ae2d1a782a452a50974440a827
Reviewed-on: https://go-review.googlesource.com/38290
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agocmd/compile: replace all uses of ptrto by typPtr
Martin Möhrmann [Sun, 19 Mar 2017 08:51:22 +0000 (09:51 +0100)]
cmd/compile: replace all uses of ptrto by typPtr

This makes the overall naming and use of the functions
to create a Type more consistent.

Passes toolstash -cmp.

Change-Id: Ie0d40b42cc32b5ecf5f20502675a225038ea40e4
Reviewed-on: https://go-review.googlesource.com/38354
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: reduce allocs when appending to Node slices
Martin Möhrmann [Fri, 24 Feb 2017 23:34:21 +0000 (00:34 +0100)]
cmd/compile: reduce allocs when appending to Node slices

Rewrite Append function such that the *Node slice argument does not escape.

Passes toolstash -cmp.

name      old alloc/op    new alloc/op    delta
Template     40.8MB ± 0%     40.8MB ± 0%  -0.17%  (p=0.000 n=20+19)
Unicode      30.3MB ± 0%     30.2MB ± 0%  -0.11%  (p=0.000 n=19+20)
GoTypes       115MB ± 0%      115MB ± 0%  -0.20%  (p=0.000 n=20+20)
Compiler      492MB ± 0%      491MB ± 0%  -0.25%  (p=0.000 n=20+20)
SSA           858MB ± 0%      858MB ± 0%  -0.08%  (p=0.000 n=20+20)
Flate        26.2MB ± 0%     26.2MB ± 0%  -0.13%  (p=0.000 n=20+19)
GoParser     32.5MB ± 0%     32.4MB ± 0%  -0.14%  (p=0.000 n=20+20)
Reflect      80.6MB ± 0%     80.4MB ± 0%  -0.27%  (p=0.000 n=20+20)
Tar          27.3MB ± 0%     27.3MB ± 0%  -0.12%  (p=0.000 n=20+19)
XML          43.1MB ± 0%     43.0MB ± 0%  -0.14%  (p=0.000 n=20+20)

name      old allocs/op   new allocs/op   delta
Template       400k ± 1%       397k ± 0%  -0.81%  (p=0.000 n=20+18)
Unicode        321k ± 1%       320k ± 0%  -0.43%  (p=0.000 n=20+20)
GoTypes       1.17M ± 0%      1.16M ± 0%  -0.89%  (p=0.000 n=20+20)
Compiler      4.59M ± 0%      4.54M ± 0%  -1.26%  (p=0.000 n=20+19)
SSA           7.68M ± 0%      7.65M ± 0%  -0.37%  (p=0.000 n=18+18)
Flate          242k ± 1%       240k ± 1%  -0.70%  (p=0.000 n=20+20)
GoParser       323k ± 1%       321k ± 1%  -0.64%  (p=0.000 n=20+20)
Reflect       1.01M ± 0%      1.00M ± 0%  -0.92%  (p=0.000 n=20+19)
Tar            258k ± 1%       256k ± 1%  -0.60%  (p=0.000 n=20+19)
XML            403k ± 1%       400k ± 0%  -0.78%  (p=0.000 n=20+20)

Change-Id: Ie1eb603dc46f729574f6a76c08085b2619249be4
Reviewed-on: https://go-review.googlesource.com/37437
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/pprof: use proxy from environment
Michel Lespinasse [Sun, 19 Mar 2017 02:02:20 +0000 (19:02 -0700)]
cmd/pprof: use proxy from environment

See #18736

Change-Id: I9c16357c05c16db677125d3077ee466b71559c7a
Reviewed-on: https://go-review.googlesource.com/38343
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: clean up SSA test API
Josh Bleecher Snyder [Sun, 19 Mar 2017 05:00:28 +0000 (22:00 -0700)]
cmd/compile: clean up SSA test API

I noted in CL 38327 that the SSA test API felt a bit
clunky after the ssa.Func/ssa.Cache/ssa.Config refactoring,
and promised to clean it up once the dust settled.
The dust has settled.

Along the way, this CL fixes a potential latent bug,
in which the amd64 test context was used for all dummy Syslook calls.
The lone SSA test using the s390x context did not depend on the
Syslook context being correct, so the bug did not arise in practice.

Change-Id: If964251d1807976073ad7f47da0b1f1f77c58413
Reviewed-on: https://go-review.googlesource.com/38346
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: canonicalize empty interface types
Josh Bleecher Snyder [Sat, 18 Mar 2017 21:39:48 +0000 (14:39 -0700)]
cmd/compile: canonicalize empty interface types

Mapping all empty interfaces onto the same Type
allows better reuse of the ptrTo and sliceOf
Type caches for *interface{} and []interface{}.

This has little compiler performance impact now,
but it will be helpful in the future,
when we will eagerly populate some of those caches.

Passes toolstash-check.

Change-Id: I17daee599a129b0b2f5f3025c1be43d569d6782c
Reviewed-on: https://go-review.googlesource.com/38344
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: add more types to ssa.Types
Josh Bleecher Snyder [Sat, 18 Mar 2017 17:16:03 +0000 (10:16 -0700)]
cmd/compile: add more types to ssa.Types

This reduces the number of calls back into the
gc Type routines, which will help performance
in a concurrent backend.
It also reduces the number of callsites
that must be considered in making the transition.

Passes toolstash-check -all. No compiler performance changes.

Updates #15756

Change-Id: Ic7a8f1daac7e01a21658ae61ac118b2a70804117
Reviewed-on: https://go-review.googlesource.com/38340
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: separate ssa.Frontend and ssa.TypeSource
Josh Bleecher Snyder [Fri, 17 Mar 2017 23:04:46 +0000 (16:04 -0700)]
cmd/compile: separate ssa.Frontend and ssa.TypeSource

Prior to this CL, the ssa.Frontend field was responsible
for providing types to the backend during compilation.
However, the types needed by the backend are few and static.
It makes more sense to use a struct for them
and to hang that struct off the ssa.Config,
which is the correct home for readonly data.
Now that Types is a struct, we can clean up the names a bit as well.

This has the added benefit of allowing early construction
of all types needed by the backend.
This will be useful for concurrent backend compilation.

Passes toolstash-check -all. No compiler performance change.

Updates #15756

Change-Id: I021658c8cf2836d6a22bbc20cc828ac38c7da08a
Reviewed-on: https://go-review.googlesource.com/38336
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/gofmt: unindent the second line of a BUG note
Damien Lespiau [Sat, 18 Mar 2017 19:44:37 +0000 (19:44 +0000)]
cmd/gofmt: unindent the second line of a BUG note

Currently, this second line is treated a pre-formatted text as it's
indented relatively to the BUG() line.

The current state can be seen at:

  https://golang.org/cmd/gofmt/#pkg-note-BUG

Unindenting makes the rest of the sentence part of the same paragraph.

Change-Id: I6dee55c9c321b1a03b41c7124c6a1ea15772c878
Reviewed-on: https://go-review.googlesource.com/38353
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoencoding/gob: speedup floats encoding and decoding
Alberto Donizetti [Sat, 18 Mar 2017 17:43:20 +0000 (18:43 +0100)]
encoding/gob: speedup floats encoding and decoding

By replacing bytes-reversing routines with bits.ReverseBytes64 calls.

name                     old time/op  new time/op  delta
EncodeComplex128Slice-4  35.1µs ± 1%  23.2µs ± 2%  -33.94%  (p=0.000 n=20+20)
EncodeFloat64Slice-4     17.9µs ± 1%  11.0µs ± 1%  -38.36%  (p=0.000 n=17+18)

name                     old time/op  new time/op  delta
DecodeComplex128Slice-4  79.7µs ± 0%  69.9µs ± 1%  -12.31%  (p=0.000 n=20+20)
DecodeFloat64Slice-4     47.3µs ± 1%  42.2µs ± 1%  -10.65%  (p=0.000 n=17+17)

Change-Id: I91a6401c6009b5712fca6258dd1e57c6fe68ea64
Reviewed-on: https://go-review.googlesource.com/38352
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: use testConfig consistently in SSA tests
Josh Bleecher Snyder [Fri, 17 Mar 2017 23:59:32 +0000 (16:59 -0700)]
cmd/compile: use testConfig consistently in SSA tests

Change-Id: Iae41e14ee55eb4068fcb2189a77b345a7c5468b4
Reviewed-on: https://go-review.googlesource.com/38333
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agonet: fix tests for /etc/hosts with entries named "test"
Matthew Dempsky [Fri, 17 Mar 2017 20:34:56 +0000 (13:34 -0700)]
net: fix tests for /etc/hosts with entries named "test"

Fixes #19592.

Change-Id: I8946b33fd36ae1f39bdcc4bf0bd4b5b99618efe8
Reviewed-on: https://go-review.googlesource.com/38300
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: make stkptrsize local
Josh Bleecher Snyder [Fri, 17 Mar 2017 16:19:56 +0000 (09:19 -0700)]
cmd/compile: make stkptrsize local

While we're here, also eliminate a few more Curfn uses.

Passes toolstash -cmp. No compiler performance impact.

Updates #15756

Change-Id: Ib8db9e23467bbaf16cc44bf62d604910f733d6b8
Reviewed-on: https://go-review.googlesource.com/38331
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: make Stksize local
Josh Bleecher Snyder [Fri, 17 Mar 2017 16:10:57 +0000 (09:10 -0700)]
cmd/compile: make Stksize local

Passes toolstash -cmp. No compiler performance impact.

Updates #15756

Change-Id: I85b45244453ae28d4da76be4313badddcbf3f5dc
Reviewed-on: https://go-review.googlesource.com/38330
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: make Maxarg local
Josh Bleecher Snyder [Fri, 17 Mar 2017 15:58:36 +0000 (08:58 -0700)]
cmd/compile: make Maxarg local

Passes toolstash -cmp. No compiler performance impact.

Updates #15756

Change-Id: I1294058716d83dd1be495d399ed7ab2277754dc6
Reviewed-on: https://go-review.googlesource.com/38329
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: pass frame size to defframe
Josh Bleecher Snyder [Fri, 17 Mar 2017 15:55:40 +0000 (08:55 -0700)]
cmd/compile: pass frame size to defframe

Preparation for de-globalizing Stksize and MaxArg.

Passes toolstash -cmp. No compiler performance impact.

Updates #15756

Change-Id: I312f0bbd15587a6aebf472cd66c8e62b89e55c8a
Reviewed-on: https://go-review.googlesource.com/38328
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoRevert "go/types: enforce Check path restrictions via panics"
Brad Fitzpatrick [Fri, 17 Mar 2017 23:53:19 +0000 (23:53 +0000)]
Revert "go/types: enforce Check path restrictions via panics"

This reverts commit b744a11a966ad3999c190fea9909ec8df0570b87.

Reason for revert: Broke trybots. (misc-vetall builder is busted)

Change-Id: I651d1c18db2fb3cb6ec12c2ae62024627baf8d77
Reviewed-on: https://go-review.googlesource.com/38332
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: thread Curfn through SSA
Josh Bleecher Snyder [Fri, 17 Mar 2017 14:49:22 +0000 (07:49 -0700)]
cmd/compile: thread Curfn through SSA

This is a first step towards eliminating the
Curfn global in the backend.
There's more to do.

Passes toolstash -cmp. No compiler performance impact.

Updates #15756

Change-Id: Ib09f550a001e279a5aeeed0f85698290f890939c
Reviewed-on: https://go-review.googlesource.com/38232
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: using a single Store op for non-pointer non-skip store
Cherry Zhang [Fri, 17 Mar 2017 15:53:24 +0000 (11:53 -0400)]
cmd/compile: using a single Store op for non-pointer non-skip store

This makes fewer Values around until decomposing, reducing
allocation in compiler.

name       old alloc/op    new alloc/op    delta
Template      41.4MB ± 0%     40.8MB ± 0%  -1.29%  (p=0.000 n=10+10)
Unicode       30.3MB ± 0%     30.2MB ± 0%  -0.24%  (p=0.000 n=10+10)
GoTypes        118MB ± 0%      115MB ± 0%  -2.23%  (p=0.000 n=10+10)
Compiler       505MB ± 0%      493MB ± 0%  -2.47%  (p=0.000 n=10+10)
SSA            881MB ± 0%      872MB ± 0%  -1.03%  (p=0.000 n=10+10)

name       old allocs/op   new allocs/op   delta
Template        401k ± 1%       400k ± 1%    ~     (p=0.631 n=10+10)
Unicode         321k ± 0%       321k ± 1%    ~     (p=0.684 n=10+10)
GoTypes        1.18M ± 0%      1.17M ± 0%  -0.34%  (p=0.000 n=10+10)
Compiler       4.63M ± 0%      4.61M ± 0%  -0.43%  (p=0.000 n=10+10)
SSA            7.83M ± 0%      7.82M ± 0%  -0.13%  (p=0.000 n=10+10)

Change-Id: I8f736396294444248a439bd4c90be1357024ce88
Reviewed-on: https://go-review.googlesource.com/38294
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: move Frontend field from ssa.Config to ssa.Func
Josh Bleecher Snyder [Fri, 17 Mar 2017 05:42:10 +0000 (22:42 -0700)]
cmd/compile: move Frontend field from ssa.Config to ssa.Func

Suggested by mdempsky in CL 38232.
This allows us to use the Frontend field
to associate frontend state and information
with a function.
See the following CL in the series for examples.

This is a giant CL, but it is almost entirely routine refactoring.

The ssa test API is starting to feel a bit unwieldy.
I will clean it up separately, once the dust has settled.

Passes toolstash -cmp.

Updates #15756

Change-Id: I71c573bd96ff7251935fce1391b06b1f133c3caf
Reviewed-on: https://go-review.googlesource.com/38327
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: evaluate config as needed in rewrite rules
Josh Bleecher Snyder [Fri, 17 Mar 2017 17:50:20 +0000 (10:50 -0700)]
cmd/compile: evaluate config as needed in rewrite rules

Prior to this CL, config was an explicit argument
to the SSA rewrite rules, and rules that needed
a Frontend got at it via config.
An upcoming CL moves Frontend from Config to Func,
so rules can no longer reach Frontend via Config.
Passing a Frontend as an argument to the rewrite rules
causes a 2-3% regression in compile times.
This CL takes a different approach:
It treats the variable names "config" and "fe"
as special and calculates them as needed.
The "as needed part" is also important to performance:
If they are calculated eagerly, the nilchecks themselves
cause a regression.

This introduces a little bit of magic into the rewrite
generator. However, from the perspective of the rules,
the config variable was already more or less magic.
And it makes the upcoming changes much clearer.

Passes toolstash -cmp.

Change-Id: I173f2bcc124cba43d53138bfa3775e21316a9107
Reviewed-on: https://go-review.googlesource.com/38326
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile/internal/gc: rename Thearch to thearch
Matthew Dempsky [Fri, 17 Mar 2017 20:35:36 +0000 (13:35 -0700)]
cmd/compile/internal/gc: rename Thearch to thearch

Prepared using gorename.

Change-Id: Id55dac9ae5446a8bfeac06e7995b35f4c249eeca
Reviewed-on: https://go-review.googlesource.com/38302
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: eliminate direct uses of gc.Thearch in backends
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.

Passes toolstash-check -all.

Change-Id: Ie6305a6cd6f6ab4e89ecbb3cbbaf5ffd57057a24
Reviewed-on: https://go-review.googlesource.com/38301
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agostrconv: replace small int string table with constant string
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>
8 years agogo/types: enforce Check path restrictions via panics
Daniel Martí [Fri, 17 Mar 2017 15:05:53 +0000 (15:05 +0000)]
go/types: enforce Check path restrictions via panics

Its godoc says that path must not be empty or dot, while the existing
implementation happily accepts both.

Change-Id: I64766271c35152dc7adb21ff60eb05c52237e6b6
Reviewed-on: https://go-review.googlesource.com/38262
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>

8 years agoencoding/gob: make integers encoding faster
Alberto Donizetti [Fri, 17 Mar 2017 17:42:58 +0000 (18:42 +0100)]
encoding/gob: make integers encoding faster

name                old time/op  new time/op  delta
EncodeInt32Slice-4  14.6µs ± 2%  12.2µs ± 1%  -16.65%  (p=0.000 n=19+18)

Change-Id: I078a171f1633ff81d7e3f981dc9a398309ecb2c0
Reviewed-on: https://go-review.googlesource.com/38269
Reviewed-by: Rob Pike <r@golang.org>
8 years agocmd/compile: intrinsic for math/bits.Reverse on ARM64
Keith Randall [Fri, 17 Mar 2017 05:34:38 +0000 (22:34 -0700)]
cmd/compile: intrinsic for math/bits.Reverse on ARM64

I don't know that it exists for any other architectures.

Update #18616

Change-Id: Idfe5dee251764d32787915889ec0be4bebc5be24
Reviewed-on: https://go-review.googlesource.com/38323
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
8 years agocmd/go: fix race libraries rebuilding by `go test -i`
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>

8 years agomime: handling invalid mime media parameters
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>

8 years agohash/crc32: improve performance for ppc64le
Lynn Boger [Mon, 13 Mar 2017 20:25:15 +0000 (16:25 -0400)]
hash/crc32: improve performance for ppc64le

This change improves the performance of crc32 for ppc64le by using
vpmsum and other vector instructions in the algorithm.

The testcase was updated to test more sizes.

Fixes #19570

BenchmarkCRC32/poly=IEEE/size=15/align=0-8             90.5          81.8          -9.61%
BenchmarkCRC32/poly=IEEE/size=15/align=1-8             89.7          81.7          -8.92%
BenchmarkCRC32/poly=IEEE/size=40/align=0-8             93.2          61.1          -34.44%
BenchmarkCRC32/poly=IEEE/size=40/align=1-8             92.8          60.9          -34.38%
BenchmarkCRC32/poly=IEEE/size=512/align=0-8            501           55.8          -88.86%
BenchmarkCRC32/poly=IEEE/size=512/align=1-8            502           132           -73.71%
BenchmarkCRC32/poly=IEEE/size=1kB/align=0-8            947           69.9          -92.62%
BenchmarkCRC32/poly=IEEE/size=1kB/align=1-8            946           144           -84.78%
BenchmarkCRC32/poly=IEEE/size=4kB/align=0-8            3602          186           -94.84%
BenchmarkCRC32/poly=IEEE/size=4kB/align=1-8            3603          263           -92.70%
BenchmarkCRC32/poly=IEEE/size=32kB/align=0-8           28404         1338          -95.29%
BenchmarkCRC32/poly=IEEE/size=32kB/align=1-8           28856         1405          -95.13%
BenchmarkCRC32/poly=Castagnoli/size=15/align=0-8       89.7          81.8          -8.81%
BenchmarkCRC32/poly=Castagnoli/size=15/align=1-8       89.8          81.9          -8.80%
BenchmarkCRC32/poly=Castagnoli/size=40/align=0-8       93.8          61.4          -34.54%
BenchmarkCRC32/poly=Castagnoli/size=40/align=1-8       94.3          61.3          -34.99%
BenchmarkCRC32/poly=Castagnoli/size=512/align=0-8      503           56.4          -88.79%
BenchmarkCRC32/poly=Castagnoli/size=512/align=1-8      502           132           -73.71%
BenchmarkCRC32/poly=Castagnoli/size=1kB/align=0-8      941           70.2          -92.54%
BenchmarkCRC32/poly=Castagnoli/size=1kB/align=1-8      943           145           -84.62%
BenchmarkCRC32/poly=Castagnoli/size=4kB/align=0-8      3588          186           -94.82%
BenchmarkCRC32/poly=Castagnoli/size=4kB/align=1-8      3595          264           -92.66%
BenchmarkCRC32/poly=Castagnoli/size=32kB/align=0-8     28266         1323          -95.32%
BenchmarkCRC32/poly=Castagnoli/size=32kB/align=1-8     28344         1404          -95.05%

Change-Id: Ic4d8274c66e0e87bfba5f609f508a3877aee6bb5
Reviewed-on: https://go-review.googlesource.com/38184
Reviewed-by: David Chase <drchase@google.com>
8 years agoimage/png: decode Gray8 transparent images.
Nigel Tao [Fri, 17 Mar 2017 04:34:39 +0000 (15:34 +1100)]
image/png: decode Gray8 transparent images.

Fixes #19553.

Change-Id: I414cb3b1c2dab20f41a7f4e7aba49c534ff19942
Reviewed-on: https://go-review.googlesource.com/38271
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: relocate code from config.go to func.go
Josh Bleecher Snyder [Thu, 16 Mar 2017 05:43:40 +0000 (22:43 -0700)]
cmd/compile: relocate code from config.go to func.go

This is a follow-up to CL 38167.
Pure code movement.

Change-Id: I13e58f7eac6718c77076d89e13fc721a5205ec57
Reviewed-on: https://go-review.googlesource.com/38322
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: rearrange fields between ssa.Func, ssa.Cache, and ssa.Config
Josh Bleecher Snyder [Wed, 15 Mar 2017 18:15:13 +0000 (11:15 -0700)]
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.

Change-Id: I2eb7af58da067ef6a36e815a6f386cfe8634d098
Reviewed-on: https://go-review.googlesource.com/38167
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: avoid calling unnecessary Sym format routine
Josh Bleecher Snyder [Thu, 16 Mar 2017 06:01:31 +0000 (23:01 -0700)]
cmd/compile: avoid calling unnecessary Sym format routine

Minor cleanup only.

No reason to go through String() when it is
just as easy to do a direct string comparison.

Eliminates a surprising number of allocations.

name       old alloc/op    new alloc/op    delta
Template      40.9MB ± 0%     40.9MB ± 0%    ~     (p=0.190 n=10+10)
Unicode       30.3MB ± 0%     30.3MB ± 0%    ~     (p=0.218 n=10+10)
GoTypes        116MB ± 0%      116MB ± 0%  -0.09%  (p=0.000 n=10+10)
SSA            871MB ± 0%      869MB ± 0%  -0.14%  (p=0.000 n=10+9)
Flate         26.2MB ± 0%     26.2MB ± 0%  -0.15%  (p=0.002 n=10+10)
GoParser      32.5MB ± 0%     32.5MB ± 0%    ~     (p=0.165 n=10+10)
Reflect       80.5MB ± 0%     80.4MB ± 0%  -0.12%  (p=0.003 n=9+10)
Tar           27.3MB ± 0%     27.3MB ± 0%  -0.13%  (p=0.008 n=10+9)
XML           43.1MB ± 0%     43.1MB ± 0%    ~     (p=0.218 n=10+10)

name       old allocs/op   new allocs/op   delta
Template        402k ± 1%       400k ± 1%  -0.64%  (p=0.002 n=10+10)
Unicode         322k ± 1%       321k ± 1%    ~     (p=0.075 n=10+10)
GoTypes        1.19M ± 0%      1.18M ± 0%  -0.90%  (p=0.000 n=10+10)
SSA            7.94M ± 0%      7.81M ± 0%  -1.66%  (p=0.000 n=10+9)
Flate           246k ± 0%       242k ± 1%  -1.42%  (p=0.000 n=10+10)
GoParser        325k ± 1%       323k ± 1%  -0.84%  (p=0.000 n=10+10)
Reflect        1.02M ± 0%      1.01M ± 0%  -0.99%  (p=0.000 n=10+10)
Tar             259k ± 0%       257k ± 1%  -0.72%  (p=0.009 n=10+10)
XML             406k ± 1%       403k ± 1%  -0.69%  (p=0.001 n=10+10)

Change-Id: Ia129a4cd272027d627e1f3b27e9f07f93e3aa27e
Reviewed-on: https://go-review.googlesource.com/38230
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: move hasdefer to Func
Josh Bleecher Snyder [Thu, 16 Mar 2017 05:55:21 +0000 (22:55 -0700)]
cmd/compile: move hasdefer to Func

Passes toolstash -cmp.

Updates #15756

Change-Id: Ia071dbbd7f2ee0f8433d8c37af4f7b588016244e
Reviewed-on: https://go-review.googlesource.com/38231
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/internal/obj/ppc64: remove stackbarrier function check
Josh Bleecher Snyder [Thu, 16 Mar 2017 07:24:51 +0000 (00:24 -0700)]
cmd/internal/obj/ppc64: remove stackbarrier function check

Stack barriers were removed in CL 36620.

Change-Id: If124d65a73a7b344a42be2a4b386a14d7a0a428b
Reviewed-on: https://go-review.googlesource.com/38169
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
Reviewed-by: David Chase <drchase@google.com>