- 01 Mar, 2016 4 commits
-
-
Alexandru Moșoi authored
* Decreases the generated code slightly. * Similar to phiopt pass from gcc, except it only handles booleans. Handling Eq/Neq had no impact on the generated code. name old time/op new time/op delta Template 453ms ± 4% 451ms ± 4% ~ (p=0.468 n=24+24) GoTypes 1.55s ± 1% 1.55s ± 2% ~ (p=0.287 n=24+25) Compiler 6.53s ± 2% 6.56s ± 1% +0.46% (p=0.050 n=23+23) MakeBash 45.8s ± 2% 45.7s ± 2% ~ (p=0.866 n=24+25) name old text-bytes new text-bytes delta HelloSize 676k ± 0% 676k ± 0% ~ (all samples are equal) CmdGoSize 8.07M ± 0% 8.07M ± 0% -0.03% (p=0.000 n=25+25) Change-Id: Ia62477b7554127958a14cb27f85849b095d63663 Reviewed-on: https://go-review.googlesource.com/20090Reviewed-by:
Keith Randall <khr@golang.org> Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro> TryBot-Result: Gobot Gobot <gobot@golang.org>
-
Alexandru Moșoi authored
* This is a very basic form of straight line strength reduction. * Removes one multiplication from a[b].c++; a[b+1].c++ * It increases pressure on the register allocator because CSE creates more copies of the multiplication sizeof(a[0])*b. Change-Id: I686a18e9c24cc6f8bdfa925713afed034f7d36d0 Reviewed-on: https://go-review.googlesource.com/20091 Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Ilya Tocar authored
Writing to low 8/16 bits of register creates false dependency Generate 32-bit operations when possible. Change-Id: I8eb6c1c43a66424eec6baa91a660bceb6b80d1d3 Reviewed-on: https://go-review.googlesource.com/19506Reviewed-by:
-
Keith Randall authored
Add writeback code to each return location which copies the final result back to the correct stack location. Cgo plays tricky games by taking the address of a in f(a int) (b int) and then using that address to modify b. So for cgo-generated Go code, disable the SSAing of output args. Update #14511 Change-Id: I95cba727d53699d31124eef41db0e03935862be9 Reviewed-on: https://go-review.googlesource.com/19988Reviewed-by:
Todd Neal <todd@tneal.org> Run-TryBot: Keith Randall <khr@golang.org> Reviewed-by:
Ian Lance Taylor <iant@golang.org>
-
- 29 Feb, 2016 6 commits
-
-
Gerrit Code Review authored
-
Josh Bleecher Snyder authored
The cache gets a 62% hit rate while compiling the standard library. name old time/op new time/op delta Template 449ms ± 2% 443ms ± 4% -1.40% (p=0.006 n=23+25) GoTypes 1.54s ± 1% 1.50s ± 2% -2.53% (p=0.000 n=22+22) Compiler 5.51s ± 1% 5.39s ± 1% -2.29% (p=0.000 n=23+25) name old alloc/op new alloc/op delta Template 90.4MB ± 0% 90.0MB ± 0% -0.45% (p=0.000 n=25+25) GoTypes 334MB ± 0% 331MB ± 0% -1.05% (p=0.000 n=25+25) Compiler 1.12GB ± 0% 1.10GB ± 0% -1.57% (p=0.000 n=25+24) name old allocs/op new allocs/op delta Template 681k ± 0% 682k ± 0% +0.26% (p=0.000 n=25+25) GoTypes 2.23M ± 0% 2.23M ± 0% +0.05% (p=0.000 n=23+24) Compiler 6.46M ± 0% 6.46M ± 0% +0.02% (p=0.000 n=24+25) Change-Id: I2629c291892827493d7b55ec4d83f6973a2ab133 Reviewed-on: https://go-review.googlesource.com/20026Reviewed-by:
Alexandru Moșoi <alexandru@mosoi.ro> TryBot-Result: Gobot Gobot <gobot@golang.org>
-
Keith Randall authored
(Last?) Semi-regular merge from tip to dev.ssa. Conflicts: src/cmd/compile/internal/gc/closure.go src/cmd/compile/internal/gc/gsubr.go src/cmd/compile/internal/gc/lex.go src/cmd/compile/internal/gc/pgen.go src/cmd/compile/internal/gc/syntax.go src/cmd/compile/internal/gc/walk.go src/cmd/internal/obj/pass.go Change-Id: Ib5ea8bf74d420f4902a9c6208761be9f22371ae7
-
Keith Randall authored
It gets rewritten to an xor by the linker also. Change-Id: Iae35130325d41bd1a09b7e971190cae6f4e17fac Reviewed-on: https://go-review.googlesource.com/20058Reviewed-by:
David Chase <drchase@google.com> Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
-
Ilya Tocar authored
This is a AMD64 version of CL19743. Saves additional 1574 bytes in go binary. This also speeds up bzip2 by 1-4% Change-Id: I031ba423663c4e83fdefe44e5296f24143e303da Reviewed-on: https://go-review.googlesource.com/19939 Run-TryBot: Ilya Tocar <ilya.tocar@intel.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Ilya Tocar authored
Movups is 1 byte smaller than movapd that we currently use. Change-Id: I22f771f066529352722a28543535ec43497cb9c5 Reviewed-on: https://go-review.googlesource.com/19938Reviewed-by:
-
- 28 Feb, 2016 4 commits
-
-
David Chase authored
Avoid targeting a partial register with load; ensure source of load (writebarrier) is aligned. Better yet would be "CMPB $1,writebarrier" but that requires wrestling with flagalloc (mem operand complicates moving instruction around). Didn't see a change in time for benchcmd -n 10 Build go build net/http Verified that we clean the code up properly: 0x20a8 <main.main+104>: mov 0xc30a2(%rip),%eax # 0xc5150 <runtime.writeBarrier> 0x20ae <main.main+110>: test %al,%al Change-Id: Id5fb8c260eaec27bd727cb0ae1476c60343b0986 Reviewed-on: https://go-review.googlesource.com/19998Reviewed-by: -
David Chase authored
Exposed data already in sdom to avoid recreating it in prove. Change-Id: I834c9c03ed8faeaee013e5a1b3f955908f0e0915 Reviewed-on: https://go-review.googlesource.com/19999 Run-TryBot: David Chase <drchase@google.com> Reviewed-by:
-
Alexandru Moșoi authored
* It does very simple bounds checking elimination. E.g. removes the second check in for i := range a { a[i]++; a[i++]; } * Improves on the following redundant expression: return a6 || (a6 || (a6 || a4)) || (a6 || (a4 || a6 || (false || a6))) * Linear in the number of block edges. I patched in CL 12960 that does bounds, nil and constant propagation to make sure this CL is not just redundant. Size of pkg/tool/linux_amd64/* (excluding compile which is affected by this change): With IsInBounds and IsSliceInBounds -this -12960 92285080 +this -12960 91947416 -this +12960 91978976 +this +12960 91923088 Gain is ~110% of 12960. Without IsInBounds and IsSliceInBounds (older run) -this -12960 95515512 +this -12960 95492536 -this +12960 95216920 +this +12960 95204440 Shaves 22k on its own. * Can we handle IsInBounds better with this? In for i := range a { a[i]++; } the bounds checking at a[i] is not eliminated. Change-Id: I98957427399145fb33693173fd4d5a8d71c7cc20 Reviewed-on: https://go-review.googlesource.com/19710Reviewed-by: -
Todd Neal authored
Add the max arg length to opcodes and use it in zcse. Doesn't affect speed, but allows better checking in checkFunc and removes the need to keep a list of zero arg opcodes up to date. Change-Id: I157c6587154604119720ec6228b767b6e52bb5c7 Reviewed-on: https://go-review.googlesource.com/19994Reviewed-by:
-
- 27 Feb, 2016 12 commits
-
-
Ian Lance Taylor authored
Fixes #14483. Change-Id: I0cddfe27fd8d00ba85659d0b618410e39ebf45cb Reviewed-on: https://go-review.googlesource.com/19860Reviewed-by:
Brad Fitzpatrick <bradfitz@golang.org>
-
Ian Lance Taylor authored
Update #14473. Change-Id: Iba1ecf42d9ab5a93144941439d5cc6b0b4f4a3ac Reviewed-on: https://go-review.googlesource.com/19992Reviewed-by:
-
Martin Möhrmann authored
Once upon a time fmt did use bytes.Buffer for its buffer. The buffer write methods still mimic the bytes.Buffer signatures. The current code depends on manipulating the buffer []bytes array directly which makes going back to bytes.Buffer by only changing the type of buffer impossible. Since type buffer is not exported the methods can be simplified to the needs of fmt. This saves space and avoids unnecessary overhead. Use WriteString instead of Write for known inputs since WriteString is faster than Write to append the same data. This also saves space in the binary. Remove the add method from Printer and depending on the data to be written use WriteRune or WriteByte directly instead. In total makes the go binary around 4 kilobyte smaller. name old time/op new time/op delta SprintfEmpty-2 24.1ns ± 3% 23.8ns ± 1% -1.14% (p=0.000 n=20+20) SprintfString-2 114ns ± 2% 114ns ± 4% ~ (p=0.558 n=20+19) SprintfInt-2 116ns ± 9% 118ns ± 7% ~ (p=0.086 n=20+20) SprintfIntInt-2 195ns ± 6% 193ns ± 5% ~ (p=0.345 n=20+19) SprintfPrefixedInt-2 251ns ±16% 241ns ± 9% -3.69% (p=0.024 n=20+19) SprintfFloat-2 203ns ± 4% 205ns ± 5% ~ (p=0.153 n=20+20) SprintfBoolean-2 101ns ± 7% 96ns ±11% -5.23% (p=0.005 n=19+20) ManyArgs-2 651ns ± 7% 628ns ± 7% -3.44% (p=0.002 n=20+20) FprintInt-2 164ns ± 2% 158ns ± 2% -3.62% (p=0.000 n=20+18) FprintfBytes-2 215ns ± 1% 216ns ± 1% +0.58% (p=0.000 n=20+20) FprintIntNoAlloc-2 115ns ± 0% 112ns ± 0% -2.61% (p=0.000 n=20+20) ScanInts-2 700µs ± 0% 702µs ± 1% +0.38% (p=0.000 n=18+20) ScanRecursiveInt-2 82.7ms ± 0% 82.7ms ± 0% ~ (p=0.820 n=20+20) Change-Id: I0409eb170b8a26d9f4eb271f6292e5d39faf2d8b Reviewed-on: https://go-review.googlesource.com/19955Reviewed-by:
Rob Pike <r@golang.org>
-
Justin Nuß authored
The current implementations of the AppendQuote functions use quoteWith (through Quote) for quoting the given value and appends the returned string to the dst byte slice. quoteWith internally creates a byte slice on each call which gets converted to a string in Quote. This means the AppendQuote functions always allocates a new byte slice and a string only to append them to an existing byte slice. In the case of (Append)QuoteRune the string passed to quoteWith will also needs to be allocated from a rune first. Split quoteWith into two functions (quoteWith and appendQuotedWith) and replace the call to Quote inside AppendQuote with appendQuotedWith, which appends directly to the byte slice passed to AppendQuote and also avoids the []byte->string conversion. Also introduce the 2 functions quoteRuneWith and appendQuotedRuneWith that work the same way as quoteWith and appendQuotedWith, but take a single rune instead of a string, to avoid allocating a new string when appending a single rune, and use them in (Append)QuoteRune. Also update the ToASCII and ToGraphic variants to use the new functions. Benchmark results: benchmark old ns/op new ns/op delta BenchmarkQuote-8 428 503 +17.52% BenchmarkQuoteRune-8 148 105 -29.05% BenchmarkAppendQuote-8 435 307 -29.43% BenchmarkAppendQuoteRune-8 158 23.5 -85.13% benchmark old allocs new allocs delta BenchmarkQuote-8 3 3 +0.00% BenchmarkQuoteRune-8 3 2 -33.33% BenchmarkAppendQuote-8 3 0 -100.00% BenchmarkAppendQuoteRune-8 3 0 -100.00% benchmark old bytes new bytes delta BenchmarkQuote-8 144 144 +0.00% BenchmarkQuoteRune-8 16 16 +0.00% BenchmarkAppendQuote-8 144 0 -100.00% BenchmarkAppendQuoteRune-8 16 0 -100.00% Change-Id: I77c148d5c7242f1b0edbbeeea184878abb51a522 Reviewed-on: https://go-review.googlesource.com/18962 Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Mikio Hara authored
This is a followup change to #13111 for filtering out IPv6 literals and absolute FQDNs from being as the SNI values. Updates #13111. Fixes #14404. Change-Id: I09ab8d2a9153d9a92147e57ca141f2e97ddcef6e Reviewed-on: https://go-review.googlesource.com/19704Reviewed-by:
-
Martin Möhrmann authored
Move the decision if zero padding is allowed to doPrintf where the other formatting decisions are made. Removes some dead code for negative f.wid that was never used due to f.wid always being positive and f.minus deciding if left or right padding should be used. New padding code writes directly into the buffer and is as fast as the old version but avoids the cost of needing package init. name old time/op new time/op delta SprintfPadding-2 246ns ± 5% 245ns ± 4% ~ (p=0.345 n=50+47) Change-Id: I7dfddbac8e328f4ef0cdee8fafc0d06c784b2711 Reviewed-on: https://go-review.googlesource.com/19957 Run-TryBot: Rob Pike <r@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Matthew Dempsky authored
Accidentally added in https://golang.org/cl/19968. Change-Id: Id70917c4d1f69db149688d797c90a19557d16f72 Reviewed-on: https://go-review.googlesource.com/19985Reviewed-by:
Minux Ma <minux@golang.org>
-
David du Colombier authored
Plan 9 doesn't define main, so the INITENTRY symbol remains with the SXREF type, which leads Entryvalue to fail on "entry not text: main". Fixes #14536. Change-Id: Id9b7d61e5c2202aba3ec9cd52f5b56e0a38f7c47 Reviewed-on: https://go-review.googlesource.com/19973 Run-TryBot: David du Colombier <0intro@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Keith Randall authored
plive doesn't like the fact that we put JMPs right after CALL ops to select{send,recv}. Fixes SSA -N build. Change-Id: I9b3c9e5293196094fd5a6206dd2f99784951f7a9 Reviewed-on: https://go-review.googlesource.com/19982 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: -
Brad Fitzpatrick authored
Change-Id: I6ac3d11ee8896f636235fe10c2556f93584019b5 Reviewed-on: https://go-review.googlesource.com/19984Reviewed-by:
-
Matthew Dempsky authored
Split the syms array into separate basicTypes and builtinFuncs arrays. Also, in lexfini, instead of duplicating the code from lexinit to declare the builtin identifiers in the user package, just import them from builtinpkg like how importdot works. Change-Id: Ic3b3b454627a46f7bd5f290d0e31443e659d431f Reviewed-on: https://go-review.googlesource.com/19936 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
Robert Griesemer <gri@golang.org>
-
Ian Lance Taylor authored
Introduces a new types Nodes that can be used to replace NodeList. Update #14473. Change-Id: Id77c5dcae0cbeb898ba12dd46bd400aad408871c Reviewed-on: https://go-review.googlesource.com/19969Reviewed-by:
David Crawshaw <crawshaw@golang.org> Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by:
-
- 26 Feb, 2016 14 commits
-
-
Martin Möhrmann authored
All io.Reader that are passed to newScanState in all the standard library tests that implement io.RuneReader also implement io.RuneScanner. Do not check on each call ScanState's UnreadRune that the used RuneReader also implements the UnreadRune method by using a private interface. Instead require the used Reader to implement the public RuneScanner interface. The extra implementation logic for UnreadRune is removed from ScanState. Instead the readRune wrapper is extended to implement UnreadRune for the RuneScanner interface. If the Reader passed to newScanstate does not implement RuneScanner the readRune wrapper is used to implement the missing functionality. Note that a RuneReader that does not implement RuneScanner will also be wrapped by runeRead which was not the case before. Performance with the readRune wrapper is better than without before. Add benchmark to compare performance with and without using the readRune wrapper. name old time/op new time/op delta ScanInts-2 704µs ± 0% 615µs ± 1% -12.73% (p=0.000 n=20+20) ScanRecursiveInt-2 82.6ms ± 0% 51.4ms ± 0% -37.71% (p=0.000 n=20+20) ScanRecursiveIntReaderWrapper-2 85.1ms ± 0% 52.4ms ± 0% -38.36% (p=0.000 n=20+20) Change-Id: I8c6e85db9b87a8171caab12f020b6e256b498e81 Reviewed-on: https://go-review.googlesource.com/19895 Run-TryBot: Rob Pike <r@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Josh Bleecher Snyder authored
Pull all alg-related code into its own file. subr.go is a Hobbesian Leviathan. 100% code movement. Cleanup and improvements to follow. Change-Id: Ib9c8f66563fdda90c6e8cf646d366a9487a4648d Reviewed-on: https://go-review.googlesource.com/19980Reviewed-by:
Dave Cheney <dave@cheney.net> Run-TryBot: Dave Cheney <dave@cheney.net> Reviewed-by:
-
Matthew Dempsky authored
While here, merge LINC and LDEC into LINCOP. Fixes #13244. Change-Id: I8ea426f986d60d35c3b1a80c056a7aa49d22d802 Reviewed-on: https://go-review.googlesource.com/19928Reviewed-by:
-
Ian Lance Taylor authored
Save a few bytes in Func. Passes toolstash -cmp. Update #14473. Change-Id: I824fa7d5cb2d93f6f59938ccd86114abcbea0043 Reviewed-on: https://go-review.googlesource.com/19968Reviewed-by:
-
Ian Lance Taylor authored
Passes toolstash -cmp. Update #14473. Change-Id: I7285175b1992a29033fdc9e81d6f30545e5cc30d Reviewed-on: https://go-review.googlesource.com/19967Reviewed-by:
-
Josh Bleecher Snyder authored
The old implementation assumed that all memory runs were terminated by non-memory fields. This isn't necessarily so. They might be terminated by padding or blank fields. For example, given type T struct { a int64 b byte c, d, e int64 } the old implementation did a memory comparison on a+b, on c, and on d+e. Instead, check for memory runs at the beginning of every round. This now generates a memory comparison on a+b and on c+d+e. Also, delete some now-dead code. Change-Id: I66bffb111420adf6919bd708e4fb3a1e1f07fadd Reviewed-on: https://go-review.googlesource.com/19841Reviewed-by: -
Josh Bleecher Snyder authored
Passes toolstash -cmp. Change-Id: Ifae69e5ba673f01da3dfc1fd30cdc51873481623 Reviewed-on: https://go-review.googlesource.com/19840Reviewed-by:
-
Robert Griesemer authored
By using a Pragma bit set (8 bits) rather than 8 booleans, also reduce Func type size by 8 bytes (208B -> 200B on 64bit platforms, 116B -> 108B on 32bit platforms). Change-Id: Ibb7e1f8c418a0b5bc6ff813cbdde7bc6f0013b5a Reviewed-on: https://go-review.googlesource.com/19966Reviewed-by:
-
Matthew Dempsky authored
Instead add a dedicated keywords map for use in lexer.ident and drop Sym's Lexical field. Change-Id: Ia668e65499035ff7167fabbbd0cd027102b21231 Reviewed-on: https://go-review.googlesource.com/19935Reviewed-by:
-
Matthew Dempsky authored
If a general comment contains multiple newline characters, we can't simply unread one and then re-lex it via the general whitespace lexing phase, because then we'll reset lineno to the line before the "*/" marker, rather than keeping it where we found the "/*" marker. Also, for processing imports, call importfile before advancing the lexer with p.next(), so that lineno reflects the line where we found the import path, and not the token afterwards. Fixes #14520. Change-Id: I785a2d83d632280113d4b757de0d57c88ba2caf4 Reviewed-on: https://go-review.googlesource.com/19934Reviewed-by:
-
Ian Lance Taylor authored
The new TestDashS was leaving a dreg "test" file in cmd/compile/internal/gc. Create it in the temporary directory instead. Also change path.Join to filepath.Join throughout global_test.go. Change-Id: Ib7707fada2b3ab5e8abc2ba74e4c402821c1408b Reviewed-on: https://go-review.googlesource.com/19965 Run-TryBot: Ian Lance Taylor <iant@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Ian Lance Taylor authored
A slice uses less memory than a NodeList, and has better memory locality when walking the list. This uncovered a tricky case involving closures: the escape analysis pass when run on a closure was appending to the Dcl list of the OCLOSURE rather than the ODCLFUNC. This happened to work because they shared the same NodeList. Fixed with a change to addrescapes, and a check to Tempname to catch any recurrences. This removes the last use of the listsort function outside of tests. I'll send a separate CL to remove it. Unfortunately, while this passes all tests, it does not pass toolstash -cmp. The problem is that cmpstackvarlt does not fully determine the sort order, and the change from listsort to sort.Sort, while generally desirable, produces a different ordering. I could stage this by first making cmpstackvarlt fully determined, but no matter what toolstash -cmp is going to break at some point. In my casual testing the compiler is 2.2% faster. Update #14473. Change-Id: I367d66daa4ec73ed95c14c66ccda3a2133ad95d5 Reviewed-on: https://go-review.googlesource.com/19919Reviewed-by:
-
Matthew Dempsky authored
The Go toolchain stopped creating them before Go 1.3, so no point in worrying about them today. History: - Git commit 250a0919 added cmd/ar, which wrote Plan 9 __.SYMDEF entries into archive files. - golang.org/cl/6500117 renamed __.SYMDEF to __.GOSYMDEF. (Notably, the commit message suggests users need to use Go nm to read symbols, but even back then the toolchain did nothing with __.(GO)?SYMDEF files except skip over them.) - golang.org/cl/42880043 added the -pack flag to cmd/gc to directly produce archives by the Go compiler, and did not write __.GOSYMDEF entries. - golang.org/cl/52310044 rewrote cmd/pack in Go, and removed support for producing __.GOSYMDEF entries. Change-Id: I255edf40d0d3690e3447e488039fcdef73c6d6b1 Reviewed-on: https://go-review.googlesource.com/19924 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
-
Richard Miller authored
Atomic load/store/add/swap routines, as for other ARM platforms, but with DMB inserted for load/store (assuming that "atomic" also implies acquire/release memory ordering). Change-Id: I70a283d8f0ae61a66432998ce59eac76fd940c67 Reviewed-on: https://go-review.googlesource.com/18965Reviewed-by:
-
