s390x doesn't introduce any new assembly syntax. There are a few
instructions which require the operands to be reordered, notably
the storage-storage instructions that put the length into From3 so
that the memory operands can be put into From and To.
The assembly test currently covers a subset of instructions but
tries to hit edge cases as much as possible. Unlike the other ports
it can be linked as an executable to make disassembling it easy.
It would be nice to autogenerate it at some point in the future.
Change-Id: I8dd542c34b9e450b8129d46693a5acb0ded791ce
Reviewed-on: https://go-review.googlesource.com/21253
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
substAny needs access to many internal details
of gc.Type. substArgTypes comes along for the ride.
Change-Id: I430a4edfd54a1266522f7a9818e5e7b5da72479c
Reviewed-on: https://go-review.googlesource.com/21250
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Based on the ppc64 port.
s390x supports 2, 4 and 6 byte instructions and Go assembly
instructions sometimes map to several s390x instructions. The
assembler loops until a fixed point is reached in order to use
branch instructions that can only handle a short offset in a
similar way to other ports.
Change-Id: I4278bf46aca35a96ca9cea0857e6229643c9c1e3
Reviewed-on: https://go-review.googlesource.com/20942
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Previously if we were only using the low bits of AuxInt,
the high bits were ignored and could be junk. This CL
changes that behavior to define the high bits to be the
sign-extended version of the low bits for all cases.
There are 2 main benefits:
- Deterministic representation. This helps with CSE.
(Const8 [0x1]) and (Const8 [0x101]) used to be the same "value"
but CSE couldn't see them as such.
- Testability. We can check that all ops leave AuxInt in a state
consistent with the new rule. In the old scheme, it was hard
to check whether a rule correctly used only the low-order bits.
Side benefits:
- ==0 and !=0 tests are easier.
Drawbacks:
- This differs from the runtime representation in registers,
where it is important that we allow upper bits to be undefined
(so we're not sign/zero-extending all the time).
- Ops that treat AuxInt as unsigned (shifts, mostly) need to be
a bit more careful.
Change-Id: I9a685ff27e36dc03287c9ab1cecd6c0b4045c819
Reviewed-on: https://go-review.googlesource.com/21256
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Flip around the composition order of the http.Response.Body's
gzip.Reader vs. the reader which keeps track of waiting to see the end
of the HTTP/1 response framing (whether that's a Content-Length or
HTTP/1.1 chunking).
Previously:
user -> http.Response.Body
-> bodyEOFSignal
-> gzipReader
-> gzip.Reader
-> bufio.Reader
[ -> http/1.1 de-chunking reader ] optional
-> http1 framing *body
But because bodyEOFSignal was waiting to see an EOF from the
underlying gzip.Reader before reusing the connection, and gzip.Reader
(or more specifically: the flate.Reader) wasn't returning an early
io.EOF with the final chunk, the bodyEOfSignal was never releasing the
connection, because the EOF from the http1 framing was read by a party
who didn't care about it yet: the helper bufio.Reader created to do
byte-at-a-time reading in the flate.Reader.
Flip the read composition around to:
user -> http.Response.Body
-> gzipReader
-> gzip.Reader
-> bufio.Reader
-> bodyEOFSignal
[ -> http/1.1 de-chunking reader ] optional
-> http1 framing *body
Now when gzip.Reader does its byte-at-a-time reading via the
bufio.Reader, the bufio.Reader will do its big reads against the
bodyEOFSignal reader instead, which will then see the underlying http1
framing EOF, and be able to reuse the connection.
Updates google/go-github#317
Updates #14867
And related abandoned fix to flate.Reader: https://golang.org/cl/21290
Change-Id: I3729dfdffe832ad943b84f4734b0f59b0e834749
Reviewed-on: https://go-review.googlesource.com/21291
Reviewed-by: David Symonds <dsymonds@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Record total number of relocations, pcdata, automatics, funcdata and files in
object file and use these numbers in the linker to allocate contiguous
slices to later be filled by the defined symbols.
name old secs new secs delta
LinkCmdGo 0.52 ± 3% 0.49 ± 3% -4.21% (p=0.000 n=91+92)
LinkJuju 4.48 ± 4% 4.21 ± 7% -6.08% (p=0.000 n=96+100)
name old MaxRSS new MaxRSS delta
LinkCmdGo 122k ± 2% 120k ± 4% -1.66% (p=0.000 n=98+93)
LinkJuju 799k ± 5% 865k ± 8% +8.29% (p=0.000 n=89+99)
GOGC=off
name old secs new secs delta
LinkCmdGo 0.42 ± 2% 0.41 ± 0% -2.98% (p=0.000 n=89+70)
LinkJuju 3.61 ± 0% 3.52 ± 1% -2.46% (p=0.000 n=80+89)
name old MaxRSS new MaxRSS delta
LinkCmdGo 130k ± 1% 128k ± 1% -1.33% (p=0.000 n=100+100)
LinkJuju 1.00M ± 0% 0.99M ± 0% -1.70% (p=0.000 n=100+100)
Change-Id: Ie08f6ccd4311bb78d8950548c678230a58635c73
Reviewed-on: https://go-review.googlesource.com/21026
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The CMP* family of instructions are longer than their TEST counterparts by one byte.
After this change, my go tool has 13 cmp.*$0x0 instructions, compared to 5612 before.
Change-Id: Ieb87d65657917e494c0e4b711a7ba2918ae27610
Reviewed-on: https://go-review.googlesource.com/21255
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Simplify the handling of zero padding in fmt_integer and
fmt_float to not require any adjustment of the format flags.
Note that f.zero can only be true when padding to the left
and f.wid is always greater than or equal to 0.
Change-Id: I204b57d103c0eac13d86995992f2b26209196925
Reviewed-on: https://go-review.googlesource.com/21185
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
These are the first of several convenience
constructors for types.
They are part of type field encapsulation.
This removes most external writes to TARRAY Type and Bound fields.
substAny still directly fiddles with the .Type field.
substAny generally needs access to Type internals.
It will be moved to type.go in a future CL.
bimport still directly writes the .Type field.
This is hard to change.
Also of note:
* inl.go contains an (apparently irrelevant) bug fix:
as.Right was given the wrong type.
vararrtype was previously unused.
* I believe that aindex (subr.go) never creates slices,
but it is safer to keep existing behavior.
The removal of -1 as a constant there is part
of hiding that implementation detail.
Future CLs will finish that job.
Passes toolstash -cmp.
Change-Id: If09bf001a874d7dba08e9ad0bcd6722860af4b91
Reviewed-on: https://go-review.googlesource.com/21249
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Previously format argument was detected via scanning func type args.
This didn't work when func type couldn't be determined if the func
is declared in the external package. Fall back to scanning for
the first string call argument in this case.
Fixes#14754
Change-Id: I571cc29684cc641bc87882002ef474cf1481e9e2
Reviewed-on: https://go-review.googlesource.com/21023
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
This is a change improving consistency in the source tree.
The pattern foo &= ^bar, was only used six times in src/ directory.
The usage of the supported &^ (bit clear / AND NOT) operator is way more
common, about factor 10x.
Change-Id: If26a2994fd81d23d42189bee00245eb84e672cf3
Reviewed-on: https://go-review.googlesource.com/21224
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
RFC 2047 recommends a maximum length of 75 characters for
encoded-words. Due to a bug, encoded-words were limited to 77
characters instead of 75.
Change-Id: I2ff9d013ab922df6fd542464ace70b1c46dc7ae7
Reviewed-on: https://go-review.googlesource.com/20918
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Add a "HuffmanOnly" compression level, where the input is
only entropy encoded.
The output is fully inflate compatible. Typical compression
is reduction is about 50% of typical level 1 compression, however
the compression time is very stable, and does not vary as much as
nearly as much level 1 compression (or Snappy).
This mode is useful for:
* HTTP compression in a CPU limited environment.
* Entropy encoding Snappy compressed data, for archiving, etc.
* Compression where compression time needs to be predictable.
* Fast network transfer.
Snappy "usually" performs inbetween this and level 1 compression-wise,
but at the same speed as "Huffman", so this is not a replacement,
but a good supplement for Snappy, since it usually can compress
Snappy output further.
This is implemented as level -2, since this would be too much of a
compression reduction to replace level 1.
>go test -bench=Encode -cpu=1
BenchmarkEncodeDigitsHuffman1e4 30000 52334 ns/op 191.08 MB/s
BenchmarkEncodeDigitsHuffman1e5 3000 518343 ns/op 192.92 MB/s
BenchmarkEncodeDigitsHuffman1e6 300 5356884 ns/op 186.68 MB/s
BenchmarkEncodeDigitsSpeed1e4 5000 324214 ns/op 30.84 MB/s
BenchmarkEncodeDigitsSpeed1e5 500 3952614 ns/op 25.30 MB/s
BenchmarkEncodeDigitsSpeed1e6 30 40760350 ns/op 24.53 MB/s
BenchmarkEncodeDigitsDefault1e4 5000 387056 ns/op 25.84 MB/s
BenchmarkEncodeDigitsDefault1e5 300 5950614 ns/op 16.80 MB/s
BenchmarkEncodeDigitsDefault1e6 20 63842195 ns/op 15.66 MB/s
BenchmarkEncodeDigitsCompress1e4 5000 391859 ns/op 25.52 MB/s
BenchmarkEncodeDigitsCompress1e5 300 5707112 ns/op 17.52 MB/s
BenchmarkEncodeDigitsCompress1e6 20 59839465 ns/op 16.71 MB/s
BenchmarkEncodeTwainHuffman1e4 20000 73498 ns/op 136.06 MB/s
BenchmarkEncodeTwainHuffman1e5 2000 595892 ns/op 167.82 MB/s
BenchmarkEncodeTwainHuffman1e6 200 6059016 ns/op 165.04 MB/s
BenchmarkEncodeTwainSpeed1e4 5000 321212 ns/op 31.13 MB/s
BenchmarkEncodeTwainSpeed1e5 500 2823873 ns/op 35.41 MB/s
BenchmarkEncodeTwainSpeed1e6 50 27237864 ns/op 36.71 MB/s
BenchmarkEncodeTwainDefault1e4 3000 454634 ns/op 22.00 MB/s
BenchmarkEncodeTwainDefault1e5 200 6859537 ns/op 14.58 MB/s
BenchmarkEncodeTwainDefault1e6 20 71547405 ns/op 13.98 MB/s
BenchmarkEncodeTwainCompress1e4 3000 462307 ns/op 21.63 MB/s
BenchmarkEncodeTwainCompress1e5 200 7534992 ns/op 13.27 MB/s
BenchmarkEncodeTwainCompress1e6 20 80353365 ns/op 12.45 MB/s
PASS
ok compress/flate 55.333s
Change-Id: I8e12ad13220e50d4cf7ddba6f292333efad61b0c
Reviewed-on: https://go-review.googlesource.com/20982
Reviewed-by: Joe Tsai <joetsai@digital-static.net>
Reviewed-by: Nigel Tao <nigeltao@golang.org>
Patch originally from Steven Hartland. Tweaked a bit & added a test.
Fixes#7197
Change-Id: I09012b4674e7c641dba31a24e9758cedb898d3ee
Reviewed-on: https://go-review.googlesource.com/21196
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
This was the only unconverted instance.
Change-Id: Ic0ba75824614fcd1e055316e62e26acd06801dd1
Reviewed-on: https://go-review.googlesource.com/21247
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TestEvalSymlinksCanonicalNames fails on system where 8dot3 name creation
is disabled. Add new test that temporarily changes 8dot3 name creation
file system setting and runs TestEvalSymlinksCanonicalNames under that
setting. New test requires administrator access and modifies important
file system setting, so don't run the test unless explicitly requested
by specifying new test flag.
Updates #13980
Change-Id: I598b5b956e6bd0ed556e79d350cb244808c89c0b
Reviewed-on: https://go-review.googlesource.com/20863
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
The previous rules to combine indexed loads produced addresses like:
From: obj.Addr{
Type: TYPE_MEM,
Reg: REG_CX,
Name: NAME_AUTO,
Offset: 121,
...
}
which are erroneous because NAME_AUTO implies a base register of
REG_SP, and cmd/internal/obj/x86 makes many assumptions to this
effect. Note that previously we were also producing an extra "ADDQ
SP, CX" instruction, so indexing off of SP was already handled.
The approach taken by this CL to address the problem is to instead
produce addresses like:
From: obj.Addr{
Type: TYPE_MEM,
Reg: REG_SP,
Name: NAME_AUTO,
Offset: 121,
Index: REG_CX,
Scale: 1,
}
and to omit the "ADDQ SP, CX" instruction.
Downside to this approach is it requires adding a lot of new
MOV[WLQ]loadidx1 instructions that nearly duplicate functionality of
the existing MOV[WLQ]loadidx[248] instructions, but with a different
Scale.
Fixes#15001.
Change-Id: Iad9a1a41e5e2552f8d22e3ba975e4ea0862dffd2
Reviewed-on: https://go-review.googlesource.com/21245
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
See #14874
This change adds a compiler optimization for pointer shaped convT2I.
Since itab symbols are now emitted by the compiler, the itab address can
be directly moved into the iface structure.
Change-Id: I311483af544519ca682c5f872960717ead772f26
Reviewed-on: https://go-review.googlesource.com/20901
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
See #14874
This change tells the linker to collect all the itablink symbols and
collect them so that moduledata can have a slice of all compiler
generated itabs.
The logic is shamelessly adapted from what is done with typelink symbols.
Change-Id: Ie93b59acf0fcba908a876d506afbf796f222dbac
Reviewed-on: https://go-review.googlesource.com/20889
Reviewed-by: Keith Randall <khr@golang.org>
See #14874
This change tells the compiler to emit itab and itablink symbols in
situations where they could be useful; however the compiled code does
not actually make use of the new symbols yet.
Change-Id: I0db3e6ec0cb1f3b7cebd4c60229e4a48372fe586
Reviewed-on: https://go-review.googlesource.com/20888
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Michel Lespinasse <walken@google.com>
See #14874
This change makes the runtime register all compiler generated itabs
(as obtained from the moduledata) during init.
Change-Id: I9969a0985b99b8bda820a631f7fe4c78f1174cdf
Reviewed-on: https://go-review.googlesource.com/20900
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Michel Lespinasse <walken@google.com>
In syscall.forkAndExecInChild, blocks of code labelled Pass 1
and Pass 2 permute the file descriptors (if necessary) which are
passed to the child process. If Pass 1 begins with fds = {0,2,1},
nextfd = 4 and pipe = 4, then the statement labelled "don't stomp
on pipe" is too late -- the pipe (which will be needed to pass
exec status back to the parent) will have been closed by the
preceding DUP call.
Moving the "don't stomp" test earlier ensures that the pipe is
protected.
Fixes#14979
Change-Id: I890c311527f6aa255be48b3277c1e84e2049ee22
Reviewed-on: https://go-review.googlesource.com/21184
Run-TryBot: David du Colombier <0intro@gmail.com>
Reviewed-by: David du Colombier <0intro@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This mostly a mechanical change.
However, the change in assignop (subr.go) is a bug fix.
The code didn’t match the comment,
and the comment was correct.
Nevertheless, this CL passes toolstash -cmp.
The last direct reference to dddBound outside
type.go (in typecheck.go) will go away
in a future CL.
Change-Id: Ifb1691e0a07f906712c18c4a4cd23060807a5da5
Reviewed-on: https://go-review.googlesource.com/21235
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Packed encoding is the default on the proto3 format. Profiles generated
in the profile.proto format by third parties cannot be decoded by the
Go pprof tool, since its proto decoder does not recognize packed
encoding for repeated fields.
In particular this issue prevents go tool pprof from reading profiles
generated by the version of pprof in github.com/google/pprof
Profiles generated by go tool pprof after this change will use packed
repeating fields, so older versions of pprof will not be able to read
them. pprof will continue to be able to read profiles generated before
this change.
Change-Id: Ife0b353a535ae1e495515b9bcec588dd967e171b
Reviewed-on: https://go-review.googlesource.com/21240
Reviewed-by: David Symonds <dsymonds@golang.org>
Run-TryBot: David Symonds <dsymonds@golang.org>
When building make.bash, calling Nodes.Set(s) where len(s) == 0 occurs
4738678 times vs 1465415 calls where len(s) > 0; i.e., it is over 3x
more common to set Nodes.slice to nil rather than to s.
Make a copy of slice (header) and take address of that copy instead
to avoid allocating the argument slice on the heap always even when
not needed.
Saves 4738678 slice header allocations and slice header value copies.
Change-Id: I88e8e919ea9868ceb2df46173d187af4109bd947
Reviewed-on: https://go-review.googlesource.com/21241
Reviewed-by: Alan Donovan <adonovan@google.com>
Now that structs use a slice to store their fields, this code can be
simplified somewhat.
Passes toolstash -cmp.
Change-Id: If17b1c89871fa06f34938fa67df0f8c6bcf1a86b
Reviewed-on: https://go-review.googlesource.com/21219
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This reverts commit 85bbabd9c4.
The reverted CL broke all builds, because it depends on other CLs
that haven't been reviewed or landed yet.
Change-Id: I936f969431e0ac77133e43de2bf63042cef6b777
Reviewed-on: https://go-review.googlesource.com/21238
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
s390x doesn't introduce any new assembly syntax. There are a few
instructions which require the operands to be reordered, notably
the storage-storage instructions that put the length into From3 so
that the memory operands can be put into From and To.
The assembly test currently covers a subset of instructions but
tries to hit edge cases as much as possible. Unlike the other ports
it can be linked as an executable to make disassembling it easy.
It would be nice to autogenerate it at some point in the future.
Change-Id: I7615ac6ecf239e3f347fad9ae1f8eede91742859
Reviewed-on: https://go-review.googlesource.com/20934
Run-TryBot: Rob Pike <r@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
linux/386 depends on modify_ldt system call, but recent Linux kernels
can disable this system call. Any Go programs built as linux/386
crash with the message 'Trace/breakpoint trap'.
The kernel config CONFIG_MODIFY_LDT_SYSCALL, which control
enable/disable modify_ldt, is disabled on Amazon Linux 2016.03.
This fixes this problem by using set_thread_area instead of modify_ldt
on linux/386.
Fixes#14795.
Change-Id: I0cc5139e40e9e5591945164156a77b6bdff2c7f1
Reviewed-on: https://go-review.googlesource.com/21190
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Minux Ma <minux@golang.org>
One intrinsic was needed to help get the very best
performance out of a future GC; as long as that one was
being added, I also added Bswap since that is sometimes
a handy thing to have. I had intended to fill out the
bit-scan intrinsic family, but the mismatch between the
"scan forward" instruction and "count leading zeroes"
was large enough to cause me to leave it out -- it poses
a dilemma that I'd rather dodge right now.
These intrinsics are not exposed for general use.
That's a separate issue requiring an API proposal change
( https://github.com/golang/proposal )
All intrinsics are tested, both that they are substituted
on the appropriate architecture, and that they produce the
expected result.
Change-Id: I5848037cfd97de4f75bdc33bdd89bba00af4a8ee
Reviewed-on: https://go-review.googlesource.com/20564
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Calling the read only Linkrlookup will now not cause the name
string to escape. So a lookup can be performed on a []byte
casted to a string without allocating. This will help a followup
cl and it is also much simpler and cleaner.
Performance not impacted by this.
name old s/op new s/op delta
LinkCmdGo 0.51 ± 6% 0.51 ± 5% ~ (p=0.192 n=98+98)
Change-Id: I7846ba3160eb845a3a29cbf0be703c47369ece16
Reviewed-on: https://go-review.googlesource.com/21187
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
I want to get rid of OTFUNC, which serves no useful purpose. However,
it turns out that the escape analysis pass looks at the node slices set
up for OTFUNC, even though by the time escape analysis runs the OTFUNC
has been converted to OTYPE. This CL converts the escape analysis code
to look at the function decls instead, and clears the OTFUNC info when
converting to OTYPE to ensure that nothing else looks at it.
Change-Id: I3f2f5997ea8ea7a127a858e94b20aabfab84a5bf
Reviewed-on: https://go-review.googlesource.com/21202
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
