- 06 Aug, 2015 6 commits
-
-
Josh Bleecher Snyder authored
Change-Id: I684440dc316625d5572cc12179adbc206e306429 Reviewed-on: https://go-review.googlesource.com/13263Reviewed-by:
Keith Randall <khr@golang.org>
-
Josh Bleecher Snyder authored
No functional changes. The intent is just to make this easier to read and maintain. Change-Id: Iec207546482cd62bcb22eaae8efe5be6c4f15378 Reviewed-on: https://go-review.googlesource.com/13284Reviewed-by:
-
Josh Bleecher Snyder authored
Change-Id: I402841743fcdc287631646039eb149f5cfdf886c Reviewed-on: https://go-review.googlesource.com/13269Reviewed-by:
-
Josh Bleecher Snyder authored
regalloc expects to find all OpSP and OpSB values in the entry block. There is no value to moving them; don't. Change-Id: I775198f03ce7420348721ffc5e7d2bab065465b1 Reviewed-on: https://go-review.googlesource.com/13266Reviewed-by:
-
Josh Bleecher Snyder authored
Change-Id: I2667b0923e17df7cbf08e34ebec1b69a0f2f02b2 Reviewed-on: https://go-review.googlesource.com/13265Reviewed-by:
-
Josh Bleecher Snyder authored
Given (say) b1: <- b2 b3 v1 = Phi <t> v2 v3 b2: v2 = ... b3: ... tighten will move v2 to b1, since it is only used in b1. This is wrong; v2 needs to be evaluated before entering b1. Fix it. Change-Id: I2cc3b30e3ffd221cf594e36cec534dfd9cf3c6a7 Reviewed-on: https://go-review.googlesource.com/13264Reviewed-by:
-
- 05 Aug, 2015 1 commit
-
-
Josh Bleecher Snyder authored
Even this very simple, restricted initial implementation helps. While running make.bash, it moves 84437 values to new, closer homes. As a concrete example: func f_ssa(i, j int, b bool) int { if !b { return 0 } return i + j } It cuts off one stack slot and two instructions: Before: "".f_ssa t=1 size=96 value=0 args=0x20 locals=0x18 0x0000 00000 (x.go:3) TEXT "".f_ssa(SB), $24-32 0x0000 00000 (x.go:3) SUBQ $24, SP 0x0004 00004 (x.go:3) FUNCDATA $0, "".gcargs·0(SB) 0x0004 00004 (x.go:3) FUNCDATA $1, "".gclocals·1(SB) 0x0004 00004 (x.go:5) MOVQ $0, AX 0x0006 00006 (x.go:3) MOVQ 32(SP), CX 0x000b 00011 (x.go:3) MOVQ 40(SP), DX 0x0010 00016 (x.go:3) LEAQ 48(SP), BX 0x0015 00021 (x.go:3) MOVB (BX), BPB 0x0018 00024 (x.go:3) MOVQ $0, SI 0x001a 00026 (x.go:3) MOVQ SI, 56(SP) 0x001f 00031 (x.go:3) TESTB BPB, BPB 0x0022 00034 (x.go:5) MOVQ AX, (SP) 0x0026 00038 (x.go:3) MOVQ CX, 8(SP) 0x002b 00043 (x.go:3) MOVQ DX, 16(SP) 0x0030 00048 (x.go:4) JEQ 74 0x0032 00050 (x.go:3) MOVQ 8(SP), AX 0x0037 00055 (x.go:3) MOVQ 16(SP), CX 0x003c 00060 (x.go:7) LEAQ (AX)(CX*1), DX 0x0040 00064 (x.go:7) MOVQ DX, 56(SP) 0x0045 00069 (x.go:3) ADDQ $24, SP 0x0049 00073 (x.go:3) RET 0x004a 00074 (x.go:5) MOVQ (SP), AX 0x004e 00078 (x.go:5) MOVQ AX, 56(SP) 0x0053 00083 (x.go:3) JMP 69 After: "".f_ssa t=1 size=80 value=0 args=0x20 locals=0x10 0x0000 00000 (x.go:3) TEXT "".f_ssa(SB), $16-32 0x0000 00000 (x.go:3) SUBQ $16, SP 0x0004 00004 (x.go:3) FUNCDATA $0, "".gcargs·0(SB) 0x0004 00004 (x.go:3) FUNCDATA $1, "".gclocals·1(SB) 0x0004 00004 (x.go:3) MOVQ 32(SP), AX 0x0009 00009 (x.go:3) MOVQ 24(SP), CX 0x000e 00014 (x.go:3) LEAQ 40(SP), DX 0x0013 00019 (x.go:3) MOVB (DX), BL 0x0015 00021 (x.go:3) MOVQ $0, BP 0x0017 00023 (x.go:3) MOVQ BP, 48(SP) 0x001c 00028 (x.go:3) TESTB BL, BL 0x001e 00030 (x.go:3) MOVQ AX, (SP) 0x0022 00034 (x.go:3) MOVQ CX, 8(SP) 0x0027 00039 (x.go:4) JEQ 64 0x0029 00041 (x.go:3) MOVQ 8(SP), AX 0x002e 00046 (x.go:3) MOVQ (SP), CX 0x0032 00050 (x.go:7) LEAQ (AX)(CX*1), DX 0x0036 00054 (x.go:7) MOVQ DX, 48(SP) 0x003b 00059 (x.go:3) ADDQ $16, SP 0x003f 00063 (x.go:3) RET 0x0040 00064 (x.go:5) MOVQ $0, AX 0x0042 00066 (x.go:5) MOVQ AX, 48(SP) 0x0047 00071 (x.go:3) JMP 59 Of course, the old backend is still well ahead: "".f_ssa t=1 size=48 value=0 args=0x20 locals=0x0 0x0000 00000 (x.go:3) TEXT "".f_ssa(SB), $0-32 0x0000 00000 (x.go:3) NOP 0x0000 00000 (x.go:3) NOP 0x0000 00000 (x.go:3) FUNCDATA $0, gclocals·a8eabfc4a4514ed6b3b0c61e9680e440(SB) 0x0000 00000 (x.go:3) FUNCDATA $1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB) 0x0000 00000 (x.go:4) CMPB "".b+24(FP), $0 0x0005 00005 (x.go:4) JNE 17 0x0007 00007 (x.go:5) MOVQ $0, "".~r3+32(FP) 0x0010 00016 (x.go:5) RET 0x0011 00017 (x.go:7) MOVQ "".i+8(FP), BX 0x0016 00022 (x.go:7) MOVQ "".j+16(FP), BP 0x001b 00027 (x.go:7) ADDQ BP, BX 0x001e 00030 (x.go:7) MOVQ BX, "".~r3+32(FP) 0x0023 00035 (x.go:7) RET Some regalloc improvements should help considerably. Change-Id: I95bb5dd83e56afd70ae4e983f1d32dffd0c3d46a Reviewed-on: https://go-review.googlesource.com/13142Reviewed-by:
-
- 04 Aug, 2015 9 commits
-
-
Keith Randall authored
Implement ITAB, selecting the itable field of an interface. Soften the lowering check to allow lowerings that leave generic but dead ops behind. (The ITAB lowering does this.) Change-Id: Icc84961dd4060d143602f001311aa1d8be0d7fc0 Reviewed-on: https://go-review.googlesource.com/13144Reviewed-by:
Josh Bleecher Snyder <josharian@gmail.com>
-
Josh Bleecher Snyder authored
I find myself always adding this in temporarily. Make it permanent. Change-Id: I1646b3930a07d0ea01840736ccd449b7fd24f06e Reviewed-on: https://go-review.googlesource.com/13141Reviewed-by:
-
Josh Bleecher Snyder authored
Failure to treat control ops as live can lead to them being eliminated when they live in other blocks. Change-Id: I604a1977a3d3884b1f4516bea4e15885ce38272d Reviewed-on: https://go-review.googlesource.com/13138Reviewed-by:
-
Josh Bleecher Snyder authored
They were being omitted after scheduling. Change-Id: Ia20e2dcb61fde9ec854918b958c3897bafd282a6 Reviewed-on: https://go-review.googlesource.com/13140Reviewed-by:
-
Keith Randall authored
Don't put them in the control value's block. That may be many blocks up the dominator tree. Change-Id: Iab3ea36a890ffe0e355dadec7aeb676901c4f070 Reviewed-on: https://go-review.googlesource.com/13134Reviewed-by:
-
Josh Bleecher Snyder authored
This is helpful when debugging generated code. Change-Id: I268efa3593a03bb2c4e9f07d9034c004cd40df41 Reviewed-on: https://go-review.googlesource.com/13099Reviewed-by:
-
Todd Neal authored
Rewrite ^{n}x to be ^{n % 2}x. This will eventually resolve a fuzz issue that breaks v1.5. Updates #11352 Change-Id: I1b3f93872d06222f9ff5f6fd5580178ebaf4c003 Reviewed-on: https://go-review.googlesource.com/13110Reviewed-by: -
Todd Neal authored
Change-Id: Ibb5169aade15190773ff7dd11b303c1f1345a0c2 Reviewed-on: https://go-review.googlesource.com/13100Reviewed-by:
-
Keith Randall authored
The DFS scheduler doesn't do the right thing. If a Value x is used by more than one other Value, then x is put into the DFS queue when its first user (call it y) is visited. It is not removed and reinserted when the second user of x (call it z) is visited, so the dependency between x and z is not respected. There is no easy way to fix this with the DFS queue because we'd have to rip values out of the middle of the DFS queue. The new scheduler works from the end of the block backwards, scheduling instructions which have had all of their uses already scheduled. A simple priority scheme breaks ties between multiple instructions that are ready to schedule simultaneously. Keep track of whether we've scheduled or not, and make print() use the scheduled order if we have. Fix some shift tests that this change tickles. Add unsigned right shift tests. Change-Id: I44164c10bb92ae8ab8f76d7a5180cbafab826ea1 Reviewed-on: https://go-review.googlesource.com/13069Reviewed-by:
Todd Neal <todd@tneal.org>
-
- 03 Aug, 2015 3 commits
-
-
Todd Neal authored
Reworks nilcheck to be performed by a depth first traversal of the dominator tree, keeping an updated map of the values that have been nil-checked during the traversal. benchmark old ns/op new ns/op delta BenchmarkNilCheckDeep1-8 1242 1825 +46.94% BenchmarkNilCheckDeep10-8 2397 3942 +64.46% BenchmarkNilCheckDeep100-8 29105 24873 -14.54% BenchmarkNilCheckDeep1000-8 2742563 265760 -90.31% BenchmarkNilCheckDeep10000-8 335690119 3157995 -99.06% benchmark old MB/s new MB/s speedup BenchmarkNilCheckDeep1-8 0.81 0.55 0.68x BenchmarkNilCheckDeep10-8 4.17 2.54 0.61x BenchmarkNilCheckDeep100-8 3.44 4.02 1.17x BenchmarkNilCheckDeep1000-8 0.36 3.76 10.44x BenchmarkNilCheckDeep10000-8 0.03 3.17 105.67x benchmark old allocs new allocs delta BenchmarkNilCheckDeep1-8 9 14 +55.56% BenchmarkNilCheckDeep10-8 9 23 +155.56% BenchmarkNilCheckDeep100-8 9 113 +1155.56% BenchmarkNilCheckDeep1000-8 9 1015 +11177.78% BenchmarkNilCheckDeep10000-8 9 10024 +111277.78% benchmark old bytes new bytes delta BenchmarkNilCheckDeep1-8 432 608 +40.74% BenchmarkNilCheckDeep10-8 1008 1496 +48.41% BenchmarkNilCheckDeep100-8 8064 11656 +44.54% BenchmarkNilCheckDeep1000-8 73728 145240 +96.99% BenchmarkNilCheckDeep10000-8 737280 2144411 +190.85% Change-Id: I0f86010e9823aec04aac744fdb589b65ec8acefc Reviewed-on: https://go-review.googlesource.com/12332Reviewed-by:
David Chase <drchase@google.com>
-
Todd Neal authored
Modify tests to use a known value instead of comparing the backends directly. Change-Id: I32e804e12515885bd94c4f83644cbca03b018fea Reviewed-on: https://go-review.googlesource.com/13042Reviewed-by:
-
Alexandru Moșoi authored
This is a follow up on https://go-review.googlesource.com/#/c/12420/ with some rules moved to AMD64 closer to the existing rules. Change-Id: Id346bb0fc4459b3c49b826a59cc74308a590310e Reviewed-on: https://go-review.googlesource.com/12906Reviewed-by:
-
- 31 Jul, 2015 1 commit
-
-
Josh Bleecher Snyder authored
This fixes the crypto/subtle tests. Change-Id: Ie6e721eec3481f67f13de1bfbd7988e227793148 Reviewed-on: https://go-review.googlesource.com/13000Reviewed-by:
-
- 30 Jul, 2015 20 commits
-
-
Josh Bleecher Snyder authored
The only types that remain in the ssa package are special compiler-only types. Change-Id: If957abf128ec0778910d67666c297f97f183b7ee Reviewed-on: https://go-review.googlesource.com/12933Reviewed-by:
-
Josh Bleecher Snyder authored
Change-Id: I9cbb6d53a8c2302222b13d2f33b081b704208b8a Reviewed-on: https://go-review.googlesource.com/12932Reviewed-by:
-
Alexandru Moșoi authored
From compiling go there were 260 functions where XOR was needed. Much of the required changes for implementing XOR were already done in 12813. Change-Id: I5a68aa028f5ed597bc1d62cedbef3620753dfe82 Reviewed-on: https://go-review.googlesource.com/12901Reviewed-by:
-
Josh Bleecher Snyder authored
The existing backend simply elides OCONVNOP. There's no reason for us to do any differently. Rather than insert ConvNops and then rewrite them away, stop creating them in the first place. Change-Id: I4bcbe2229fcebd189ae18df24f2c612feb6e215e Reviewed-on: https://go-review.googlesource.com/12810Reviewed-by:
-
Todd Neal authored
Fix code generation error that resulted in a multi-argument NEGQ doasm: notfound ft=13 tt=13 00134 NEGQ AX, AX 13 13 Change-Id: I8b712d21a5523eccbae1f33ccea417844c27073e Reviewed-on: https://go-review.googlesource.com/12869Reviewed-by:
-
Keith Randall authored
Semi-regular merge from tip into ssa branch Change-Id: Ida553b5c504058347c0bdcb1a987727bdcea456b
-
Russ Cox authored
Was not allocating space for the frame above sigpanic, nor was it pushing the LR into the right place. Because traceback past sigpanic only needs the LR for faulting leaves, this was not noticed too much. But it did break the sync/atomic nil deref tests. Change-Id: Icba53fffa193423aab744c37f21ee893ce2ee3ac Reviewed-on: https://go-review.googlesource.com/12926Reviewed-by:
David Crawshaw <crawshaw@golang.org>
-
Keith Randall authored
Convert shift ops to also encode the size of the shift amount. Change signed right shift from using CMOV to using bit twiddles. It is a little bit better (5 instructions instead of 4, but fewer bytes and slightly faster code). It's also a bit faster than the 4-instruction branch version, even with a very predictable branch. As tested on my machine, YMMV. Implement OCOM while we are here. Change-Id: I8ca12dd62fae5d626dc0e6da5d4bbd34fd9640d2 Reviewed-on: https://go-review.googlesource.com/12867Reviewed-by:
-
David Chase authored
ODOTTYPE should be treated a whole lot like ODOT, but it was missing completely from the switch in escwalk and thus escape status did not propagate to fields. Since interfaces are required to trigger this bug, the test was added to escape_iface.go. Fixes #11931. Change-Id: Id0383981cc4b1a160f6ad447192a112eed084538 Reviewed-on: https://go-review.googlesource.com/12921 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by:
Dmitry Vyukov <dvyukov@google.com> Reviewed-by:
Russ Cox <rsc@golang.org>
-
Keith Randall authored
Missed somehow in #12813 Change-Id: I28f2789e33822a4ff884d8a3f474522747f61c73 Reviewed-on: https://go-review.googlesource.com/12868Reviewed-by:
-
Russ Cox authored
There is absolutely no information about how this was failing. If we reenable the test then at least we can get a build log from darwin/arm. There are not even freebsd/arm or netbsd/arm builders, so not too worried about those. (That is another problem.) Change-Id: I0e739a4dd2897adbe110aa400d720d8fa02ae65f Reviewed-on: https://go-review.googlesource.com/12920Reviewed-by:
-
Russ Cox authored
Instead of pushing the denominator argument on the stack, the denominator is now passed in m. This fixes a variety of bugs related to trying to take stack traces backwards from the middle of the software div/mod routines. Some of those bugs have been kludged around in the past, but others have not. Instead of trying to patch up after breaking the stack, this CL stops breaking the stack. This is an update of https://golang.org/cl/19810043, which was rolled back in https://golang.org/cl/20350043. The problem in the original CL was that there were divisions at bad times, when m was not available. These were divisions by constant denominators, either in C code or in assembly. The Go compiler knows how to generate division by multiplication for constant denominators, but the C compiler did not. There is no longer any C code, so that's taken care of. There was one problematic DIV in runtime.usleep (assembly) but https://golang.org/cl/12898 took care of that one. So now this approach is safe. Reject DIV/MOD in NOSPLIT functions to keep them from coming back. Fixes #6681. Fixes #6699. Fixes #10486. Change-Id: I09a13c76ad08ba75b3bd5d46a3eb78e66a84ab38 Reviewed-on: https://go-review.googlesource.com/12899Reviewed-by:
Ian Lance Taylor <iant@golang.org>
-
Ian Lance Taylor authored
In order to fix issue #9401 the compiler was changed to add a padding byte to any non-empty Go struct that ends in a zero-sized field. That causes the Go version of such a C struct to have a different size than the C struct, which can considerable confusion. Change cgo so that it discards any such zero-sized fields, so that the Go and C structs are the same size. This is a change from previous releases, in that it used to be possible to refer to a zero-sized trailing field (by taking its address), and with this change it no longer is. That is unfortunate, but something has to change. It seems better to visibly break programs that do this rather than to silently break programs that rely on the struct sizes being the same. Update #9401. Fixes #11925. Change-Id: I3fba3f02f11265b3c41d68616f79dedb05b81225 Reviewed-on: https://go-review.googlesource.com/12864Reviewed-by:
-
Russ Cox authored
We want to adjust the DIV calling convention to use m, and usleep can be called without an m, so switch to a multiplication by the reciprocal (and test). Step toward a fix for #6699 and #10486. Change-Id: Iccf76a18432d835e48ec64a2fa34a0e4d6d4b955 Reviewed-on: https://go-review.googlesource.com/12898Reviewed-by:
-
Russ Cox authored
If a function is large enough to need to flush the constant pool mid-function, the line number assignment code was forcing the line numbers not just for the constant pool but for all the instructions that follow it. This made the line number information completely wrong for all but the beginning of large functions on arm. Same problem in code copied into arm64. This broke runtime/trace's TestTraceSymbolize. Fixes arm build. Change-Id: I84d9fb2c798c4085f69b68dc766ab4800c7a6ca4 Reviewed-on: https://go-review.googlesource.com/12894Reviewed-by:
-
Russ Cox authored
This allows running a cross-compile like GOOS=darwin GOARCH=arm go build std to check that everything builds. Otherwise there is a redefinition error because both root_nocgo_darwin.go and root_darwin_armx.go supply initSystemRoots. Change-Id: Ic95976b2b698d28c629bfc93d8dac0048b023578 Reviewed-on: https://go-review.googlesource.com/12897Reviewed-by:
-
Russ Cox authored
The test expects the dial to take 1.0 seconds on Windows and allows it to go to 1.095 seconds. That's far too optimistic. Recent failures are reporting roughly 1.2 seconds. Let it have 1.5. Change-Id: Id69811ccb65bf4b4c159301a2b4767deb6ee8d28 Reviewed-on: https://go-review.googlesource.com/12895Reviewed-by:
-
Andrey Petrov authored
Urge users of math/rand to consider using crypto/rand when doing security-sensitive work. Related to issue #11871. While we haven't reached consensus on how to make the package inherently safer, everyone agrees that the docs for math/rand can be improved. Change-Id: I576a312e51b2a3445691da6b277c7b4717173197 Reviewed-on: https://go-review.googlesource.com/12900Reviewed-by:
Rob Pike <r@golang.org>
-
David Crawshaw authored
For the android/arm builder. Change-Id: Iad4881689223cd6479870da9541524a8cc458cce Reviewed-on: https://go-review.googlesource.com/12859Reviewed-by:
Andrew Gerrand <adg@golang.org> Run-TryBot: David Crawshaw <crawshaw@golang.org>
-
Russ Cox authored
Fixes arm64 builder crash. The bug is possible on all architectures; you just have to get lucky and hit a preemption or a stack growth on entry to assertE2I2. The test stacks the deck. Change-Id: I8419da909b06249b1ad15830cbb64e386b6aa5f6 Reviewed-on: https://go-review.googlesource.com/12890Reviewed-by:
-
