Go 1.7

SSA backend for amd64: CSE, nil-check elimination, prove/BCE, a global register allocator. Mark-bit-based allocation (allocBits/freeindex). Smaller type descriptors (nameOff/typeOff). Binary export data by default. Linker speed and memory campaign. Frame pointers on by default on amd64.

Each area lists the stage every thread reached in this release (with what was left open afterwards), then changes that are not part of a thread. Thread names link to the full thread on its topic page.

Memory management

From Memory management: GC and allocator (Go 1.0–1.17)

dev.garbage bitmap allocator (allocBits, freeindex, ctz), allocate-black, O(dirty) stack rescan, concurrent stack shrink, 32-bit full address space.

• Decentralized GC state machine and mark termination simplification — Done: only per-P gcWork (0168c2676f), markrootDone generalised (c14d25c648); reset mark state before STW (9ab9053344). Left open: nothing noted.
• GC pacer: trigger controller, assists and background utilisation — Done: allocate black plus counting black allocations toward scan work (6002e01e34, 479501c14c, 64a26b79ac); floating garbage estimate removed (c8bd293e56); reduced assist extra credit (ccca9c9cc0); sweep ratio in gcpacertrace (17f6e5396b). Left open: next_gc still the trigger not the goal.
• Stack scanning: conservative, precise by frames, stack barriers, hybrid barrier, async safe points — Done: O(dirty stacks) rescan, no global rescan, concurrent shrink (2a889b9d93, b49b71ae19, f11e4eb5cc, e8337491aa, 269c969c81, 9f263c14ed, efb0c55407 span root marking out of mark termination). Left open: rescan still needed.
• Precise heap scanning: type information, the heap bitmap and GC programs — Done: conditional branch removed from heapBits.bits (4eb33f6b8d), clears reduced (0d26efb12a), first-word scan bit reclaimed (a20fd1f6ba), morePointers/isPointer without branches (a5d3f7ece9, d5e3d08b3a). Left open: noscan still checked through the bitmap.
• Allocating from mark bits: allocBits, freeindex and ctz — Done: the allocBits/freeindex/allocCache scheme (2ac8bdc52a, e1c4e9a754, dc65a82eff, 3479b065d4, 4093481523, 1354b32cd7, 2063d5d903, 2fb75ea6c6); stackfreelist added (aed861038f); mspan.start eliminated for base() (b7adc41fba, 3e2462387f); allocation-count fixes (d97625ae9e); publication barrier even for noscan objects (d8b08c3aa4). Left open: sweep and bitmap code needed further simplification (1.9).
• Size classes, tiny allocator and mcache/mcentral — Done: user-visible runtime.KeepAlive (6ab45c09f6), maxSliceCap for division-free slice caps (a4dd6ea152, MakeSlice -32%). Left open: nothing noted.
• Scavenger: returning memory to the OS — Done: physical-page aligned scavenging (2cdcb6f829, 1151473077, f407ca9288). Left open: nothing noted.
• Heap address space: contiguous arena to sparse arenas — Done: full 4 GB on 32-bit (e6ec82067a). Left open: nothing noted.
• Mark work distribution: parfor, work buffers, gcWork and root jobs — Done: only per-P gcWork (0168c2676f, 7b229001e7), gcWork tryGet/put fast paths (1354b32cd7), scanstack takes the gcWork (3be48b4dc8). Left open: nothing noted.

Other changes

• Misc internal — maxSliceCap, KeepAlive, priority elevation by GC worker + runnext (44497ebacb with 91740582c3 adding next flag to ready), atomic pointer stores (7c7081f514, d8e8fc292a, 8f6c35de2f), NOSPLIT address helpers (30172f1811).
• GC clean-up internal — workbuf logging removed (3b3d58e119), noescape hacks (98130b39f5), root-check debugging (b5481dd0a6, d7cedc4b74), start time merge (ff71ed86b6), pagesInUse accounting (61f56e925e), documentation (2e8b74b695, a1f7db88f8), stackfree on system stack (6a86dbe75f), poisonStack checks removed (9e8fa1e99c).

Scheduler and concurrency

From Scheduler and concurrency runtime (Go 1.0–1.17)

Concurrent stack shrinking support for channels, new Cond, better randomized stealing; threads: channels, Cond, scheduler.

• The G-M-P scheduler: from global-lock M:N to per-P run queues, spinning Ms and work stealing — Done: improved randomized stealing (the PRNG missed Ps with “frightening probability”: at GOMAXPROCS=3 one P missed 101,620 times per 1e6 tries) ea0386f85f; usleep(100) in runqgrab reduced to 3us, osyield on Windows 3b246fa863, CPU underutilization fcd7c02c70; spinning in scang/casgstatus replaced with yielding 475d113b53; MaxGomaxprocs clamp 31cf1c1779.
• Channels and select — Done: indexes for select lock order 26594c3dfd; g.waiting in lock order 005140a77e; hchan.lock protects sudog.elem db72b41bcd; lock order between G status and channel lock d45bf7228f; sudog documented 857d0b48db.
• sync.Mutex, RWMutex, WaitGroup, Cond, Once — Done: new Cond 8e7072ca83; RWMutex recursive RLock documented b3f98d7a42.
• sync.Map, atomic.Value and sync/atomic — Done: plan9 arm 5c613e9162; s390x 7da42d7597.
• Goroutine and M lifecycle: caches, ids, LockOSThread, Goexit, cgo extra Ms, deadlock detection — Done: gfree split 1a2cf91f5e; extra Ms 50048a4e8e; cgo SIGPROF traceback 84e808043f.
• Stop-the-world, safe points and GC–scheduler interaction — Done: concurrent-safe shrinkstack 276b177771, gopark never passed stack pointers 8fb182d020, O(# dirty) rescan 2a889b9d93, 5b765ce310.
• Scheduler diagnostics: schedtrace, GODEBUG, execution tracer, block/mutex profiles — Done: per-goroutine trace sequence a3703618ea; schedtrace blocked field 25a609556a.
• GOMAXPROCS, thread and resource limits — Done: clamp 31cf1c1779, 1024 fcb45b9c61^1.9, unlimited ee55000f6c^1.10.

Other changes

• gopark rule internal — 8fb182d020.
• Misc internal — 59fc42b230, e4f1d9cf2e (execution error panic values implement Error), 7e460e70d9, f135c32640, a2b615d527^1.8, 88518e7dd6^1.8, 61db2e4efa^1.8, c14050646f^1.8, 0c942e8f2c^1.8.
• G-state and sudog cleanups internal — 1e91e2a25a, cbe849fc38 (remove _Genqueue), bdc14698f8 (dropg always unwires), a4b833940d, ac74e5debc, 7043d2bb5e, 001e8e8070, 1f7a0d4b5e, 56e5e0b69c.
• Comment fixups internal — Applied review comments from CL 3742 and moved orphaned comments to proper homes. (02b8e6978a)

Data structures and type metadata

From Runtime data structures and type metadata (Go 1.0–1.17)

Descriptor shrinking (nameOff/typeOff, static name encoding, tflag, embedded rtype fields), StructOf, compiler itabs and pointer-shaped convT2I, growslice/makeslice division avoidance.

• Hash map core: from the first hashmap.c to the 8-entry bucket design — Done: no algs generated for map bucket types (a6abc1cd70), no write barrier on iteration (974c201f74). Left open: nothing noted.
• Map fast paths and compiler/runtime specialization — Done: append fast-path spill avoided (6b33b0e98e, slice side).
• Hash functions and algorithm tables: memhash, AES hash, fallback hash — Done: hash algs initialized before typemap so shared-library typemaps (themselves maps) work (f135c32640); eq/hash generation simplified (199cc194ac).
• Compiler-generated equality and hash functions — Done: simplified generation, no bucket algs (199cc194ac, e0032df81e, d3794f88e9, a6abc1cd70).
• Runtime type descriptors: layout, shrinking and name encoding — Done: the shrinking series and name encoding above (a858931200, 0231f5420f, 30f93f0994, a24b3ed753, 8df733bd22, f2772a4935, 24ce64d1a9, 7d469179e6, 95df0c6ab9, f120936dff, 1492e7db05, c165988360, 4d920410d2, e369490fb7, 797dc58457, af0fc83985, 43a274fd67 pkgPath on all struct types, c9638810df). Left open: “further shrink run time type information in Go 1.8” (e369490fb7).
• Interface values, itabs and type assertions — Done: compiler itabs (7427f2c4bd, 7043d2bb5e, 998c8e034c), getcallerpc argument fixes. Left open: shared-library duplicates.
• reflect dynamic types: ArrayOf, ChanOf, MapOf, SliceOf, FuncOf, StructOf — Done: user-visible StructOf (633ab7426a), unnamed directional channels (69285a8b46), SelectDir (368507bb6f), StructTag.Lookup (5c8674a497), unexported-methods hidden (be1b930653). Left open: embedded fields with methods unsupported (905ced0e6a).
• reflect performance, caches and API additions — Done: tflag/Name speedups (797dc58457, ed9362f769, e75c899a10, 3c6ed76da2), ptrToThis back (e369490fb7), unnamed types have no PkgPath (1f4e68d92b).
• memmove, memclr, duff routines and pointer-write atomicity — Done: MOVSB (6a33f7765f, 4b209dbf0b), ppc64 (c4807d4cc7).
• Slices and append: growslice, makeslice and string/slice conversions — Done: division avoidance (43ed65f869, 6b0688f742, a4dd6ea152, ad7448fe98, 0150f15a92, 44532f1a9d, 6f3f02f80d zeroing tmpbuf).
• Soft-float, 64-bit division and complex arithmetic helpers — Done: arm5 FP stores (9cb79e9536), MININT softfloat fix (6db98a3c51).

Other changes

• Compiler/linker metadata internal — argument count (7c546566c3), pruning methods prunes funcType (d06b0db5bd), KindSlice mapping (e05b9746dd), canonical import ordering in object files (d78c84c419), pedantic int-to-string conversions (e6066711a0), unnecessary conversions (0da4dbe232), race annotation for reflectOffs.lock (56e5e0b69c).
• Reflect semantics user-visible — s390x port (db9348b866), unnamed interface types have no name (217be5b35d), Field panics when out of bounds (9edb27e76f), method type string (0cc710dca6).
• Softfloat and type-name cleanups internal — Cleaned up softfloat code, alg.go and type names in comments, and updated dalgsym to use obj.LSym. (e6d6ad47f5, 903f096992)

Diagnostics and knobs

From Diagnostics and knobs (Go 1.0–1.17)

Per-goroutine sequence numbers cut tracing overhead (453% → 26% at 48 cores); file:line in traces; per-P TSan contexts.

• Execution tracer (trace v1): events, per-goroutine sequencing, annotations — Done: per-goroutine sequence numbers (a3703618ea); file:line in traces (0fb7b4cccd) and parsing of 1.5 traces (3fafe2e888); fixes for timestamp overflow and coarse timestamps (944a0859b9, c4c182140a); trace-event descriptions (2d342fba78); error when the binary is missing for an old trace (c3818e56d0). Left open: StartTrace cost with many goroutines (fixed in 1.8).
• Race detector integration — Done: per-P TSan contexts (caa2147532); racefini once only (733162fd6c); atomic ops on external memory (bb337372fb); OASWB rewriting (595426c0d9); a TSan memory leak (5.6GB after one day, 12.2GB after six) fixed by rebuilding the .syso files (d73ca5f4d8). Left open: nothing noted.
• CPU profiler and runtime/pprof writers (legacy format to compressed protobuf) — Done: counted profiles sorted by count (6969d9bf03); runtime.CallersFrames/Frames introduced for file and line (ad03af66eb); PIE profiling (4223294eab); ThreadCreateProfile stack copying simplified (756ea30eb0). Left open: CPU profiles are stored in memory in full.
• Heap profile accounting and MemStats — Done: scavenger memory summary documented (77026ef902). Left open: nothing noted.
• Traceback correctness and inlining (diagnostic output) — Done: traceback stops at foreign function (88d3db0a5b), cgo traceback PC and signal context (c08436d1c8, 3d037cfaf8, 03abde4971). Left open: nothing noted.

Other changes

• Misc internal — WriteHeapProfile EnableGC flag removed (8c838192b8), WriteHeapDump STW clarified (ed73efbb74), pprof package docs and a stray 0 after goexit (1cb3044c9f, b11fff3886, 29ed5da5f2).

OS layer, stacks and panics

From OS interface, stacks and panics (Go 1.5–1.17)

concurrent-safe stack shrinking during concurrent mark, s390x port, extra-M batching for cgo callbacks, plan9/arm, symbolic C backtraces, set_thread_area on 386, signal names in panics.

• New and ported OS/arch targets, 1.5–1.17 — Done: linux/s390x (cc17d1ba76, 0f08dd2183, e6f36f0cd5, 7cbe7b1e86, 78ecd61f62, d8e8fc292a, 0324a3f828), linux/mips64x cgo (b6687c8933, 5d002dbc21, 6e90432342, 2e32efc44a), plan9/arm signal handling and files (d145456b16, 0de0cafb9f), set_thread_area instead of modify_ldt on linux/386 (0f86d1edfb), int 0x80 for linux/386 syscalls (1439158120). Left open: nothing noted.
• Goroutine stack growth, shrinking, GC stack barriers and stack objects — Done: concurrent-safe shrinkstack (276b177771), shrink during concurrent mark (f11e4eb5cc), copy before adjusting (da153354b2), stack free-list (aed861038f), stackfree on the system stack (6a86dbe75f), duplicate channels in shrinkstack selects (84bb9e62f0), transient _Gwaiting states fixed in newstack (3c2a21ff13), poisonStack checks dropped (9e8fa1e99c), stack guard shrunk in [dev.ssa] (d3f15ff6bc).
• Tracebacks, symbol tables and runtime.Callers/CallersFrames — Done: print signal name in panic messages (1a7fc2357b), symbolic backtrace of C code during a cgo crash (ea306ae625), call user Error/String methods before freezing the world on crash (e960302410), off-by-one in module-for-PC (1716162a9a), cgo traceback gets signal context (3d037cfaf8), SIGPROF on non-Go thread collects a stack (84d8aff94c).
• Defer, panic, recover and Goexit (runtime side) — Done: simplified setPanicOnFault (6af4e996e2), user Error/String called before the world freezes (e960302410).
• Signals: masks, forwarding, sigaltstack and the unified Unix signal layer — Done: better last-ditch forwarding for Unix libraries (df2b2eb63d), don’t unblock all signals on BSDs (6f59ccb052), drop _SigUnblock for SIGSYS (be730b49ca), yield after raising a killing signal (c1b32acefb), openbsd/386 sigreturn (5bd37b8e78), darwin amd64 INT $3 after sigtramp (ff227b8a56), SIGPROF on non-Go thread traceback (84d8aff94c), Solaris signals (562e38c0af), Plan 9 signals (d145456b16).
• cgo runtime support: pointer checks, callbacks, extra Ms, C-to-Go call path — Done: cgo context function (5f9a870bf1, b30814bbd6, 88e0ec2979), TSAN support for malloc, acquire/release and locks (a5d1a72a40, 66736880ca, cf862478c8), mp.cgocallers (59fc42b230, 58394fd7d5), extra Ms as needed (50048a4e8e), cgocallback_gofunc fixes on arm64/ppc64x/arm (9fe572e509, 58f52cbb79, 690de51ffa), callee-saved FP registers (71916437fb, a7f7a9cca7, 72eb46c5a0), iOS exception handler signals (be10c51500).
• Shared-library and buildmode entry points (c-archive, c-shared, plugin, shared) — Done: ppc64le library startup (8a1dc32447, 32302d6289), Windows c-archive (ed8f0e5c33), defer in 386 shared (26849746c9), 16-byte stack alignment (26015b9563, 10c8b2374f).
• vDSO, time sources and process-wide randomness — Done: AT_RANDOM on arm64/mips64x (a2eded3421, 2e32efc44a), common auxv (c955bb2040, 90addd3d41), Sierra nanotime fixes (fad2bbdc6a, 2da5633eb9), iOS 10 gettimeofday ABI (9fde86b012).
• Windows runtime — Done: only search System32 for DLLs (683448a304), SwitchToThread osyield (1f5b1b2b66), c-archive on Windows (ed8f0e5c33), XMM callee-saved in cgo (71916437fb), framepointer mode safe on Windows (d9557523c2).
• Assembly-level runtime entry points: memequal/cmp/IndexByte, hashing, atomics and ABI wrappers — Done: no REP for IndexByte (687abca1ea), memeq unification (bd70bd9cb2), ppc64 unaligned loads and Equal (15ea61146e, baec148767, b75b0630fe), REP;MOVSB only when fast (4b209dbf0b), Ctz/Bswap asm on 386 (9c8809f82a), intrinsic fallbacks (98b6febcef), strings.Index 17–31 bytes (429bbf3312), aeshash seed (c83e6f50d9), 386 vsyscall PIC.
• Plan 9 and NaCl ports — Done: plan9/arm (d145456b16, 0de0cafb9f), GC-exit race (6b59d61822), fork/exit race (967b9940b4).

Other changes

• Windows internal — user-visible runtime: fix windows build (594c3aa063); user-visible runtime: revert “do not call timeBeginPeriod on windows” (0435e88a11); user-visible runtime: merge os{,2}_windows.go into os1_windows.go (e095f53e9b)
• other runtime/OS plumbing internal — all: use cannot instead of can not (e43c74a0d8); cmd/go: ignore C files when CGO_ENABLED=0 (0c7ccbf601); all: single space after period (5fea2ccc77); cmd/compile: note escape of parts of closured-capture vars (d8c815d8b5); user-visible os: consolidate os{1,2}_*.go files (8455f3a3d5)
• Darwin/iOS internal — user-visible runtime: use correct psABI SP alignment before calling libc mmap (315f4c70f1); user-visible runtime: move machport into darwin’s mOS (a4b833940d); user-visible runtime: merge the darwin os*.go files together (d9712aa82a)
• Linux, vDSO, time and auxv internal — user-visible runtime: eliminate unnecessary type conversions (a03bdc3e6b); user-visible runtime: don’t clear gcscanvalid in casfrom_Gscanstatus (5b765ce310); user-visible runtime: merge os_linux.go into os1_linux.go (5103fbfdb2); user-visible runtime: remove remaining references to TheChar (ba09d06e16); user-visible runtime: remove empty 386 archauxv (d6b177d1eb); user-visible runtime: use s.base() everywhere it makes sense (b7adc41fba); user-visible runtime: use morePointers and isPointer in more places (d5e3d08b3a); user-visible runtime: merge the last four os-vs-os1 files together (2dc680007e); user-visible runtime: delete empty files (61602b0e9e); user-visible runtime: check only up to ptrdata bytes for pointers (538537a28d); user-visible runtime: better error message for newosproc failure (c7ae41e577); runtime/internal/atomic: fix assembly arg sizes (4054769a31)
• cgo and C interop internal — runtime/cgo: make cgo work with openbsd ABI changes (0dcd330bc8)
• stacks, defer and unwinding internal — user-visible runtime: unwind BP in jmpdefer to match SP unwind (b92f423879)
• cgo check skips non-pointer slice elements internal — The cgo pointer check skips slice elements that cannot hold pointers. (c8e7b34b59)

Compiler frontend

From Compiler frontend (Go 1.1–1.9)

Binary export data on by default (inlined bodies, positions, -linkobj), NodeList removed in favour of Nodes, Type encapsulation and shrinking, itabs emitted by compiler, [dev.ssa] backend lands with frontend support, lexer rewrite.

• The C-to-Go translation of the compiler (cmd/gc → cmd/internal/gc → cmd/compile/internal/gc) — Done: NodeList eliminated in favour of Nodes (see data-structure thread).
• Parser: from yacc to a hand-written recursive-descent parser — Done: lexer/parser simplifications listed above; 3a11c8d319, e4571d33ac, 9f301643bd/bb3b10214d (parameter parsing without OKEY nodes), 135109d241. Left open: new AST parser exists only as an unused package.
• Binary export data and the object-file split — Done: 4598382724, a9ea36afbb, 5c593a3227, 7538b1db8e (default), feb6131b1a (-linkobj), a5cd53a9fd invalid types/constants, d78c84c419 sorted imports for canonical obj files, 6e4a8615f6 deterministic output via AbsFileLine, eb4d1be285 keep variable numbering for inlinable exported funcs; 80e9a7f079 all-or-no named params, 8650c23034 verify imported types after full import, 87a2ae1fa2, ef62f641c3, 21d781070c one format for builtin calls, 054a721dca safemode bit from package header, 30282b091d labels/goto/fallthrough import, bbd1dcdf7d, 575a871662 //go:nointerface, 15f7a66f36 blank names, 1402e522c6 load builtin export data once, d930d69fd9, 113c4d2581/7d3a40978a refactor import/export parsing, 7d80291c4c, f8e41f6f59, 699a2ba137, 338a891e79, 3e40f13cf3, e7524d51fd removal of lexer-IO-state globals (the “Io” type). Left open: position export disabled “for now” (889c0a66fc) and parameter numbering kept (0b8c0767d0).
• Inlining — Done: fc6bcdee79, e41f527f4d, ac8d97b679, a9ea36afbb makes imported bodies actually available; 17db07f9b5 don’t discard empty inlinable functions.
• Escape analysis and stack allocation — Done: 20803b845f, b6dc3e6f66, fb2af2b35b, 7e88826a69 (escape reads function decls instead of OTFUNC node slices).
• Closures, method values and variable capture — Done: 20803b845f, 93369001c7, b64f549ba9, e0f986bf26 (function literal name collision with “glob”), c7917de0bb.
• Language changes and rules as implemented in the compiler — Done: user-visible 49dad0f571, 75b886ab79, 3eaa304629, f5423a63df, e48a2958d1.
• Constant arithmetic, Mpint/Mpflt and floating-point constants — Done: d3253876f2, 07749aef98, 97360096e5, 35dd2ed58d (rename mparith files), f32161daf8 (one new Node per defaultlit: Unicode compile alloc −5.8%).
• Static initialization, composite literals and init — Done: 934c359964, 55ab07c224, b024ed0d94, 3c6e60c0e4, f973e79d09/7971864267/4e75932cf7 sinit cleanups, f3575a9561.
• Walk-level lowering and runtime-call optimisations — Done: f00bbd5f81, 859b63cc09, 4cef0e980a/bc8458ab02/36694064e5 (no algs for [0]T and [1]T, && in generated eq), d0c11577b9, 0f5d78f678, 92bf58c238, b30e1d728c alg-generation cleanups, 8ffe496ae7 fewer algorithm kinds, d1341d6cf3.
• Write barriers, precise stacks and liveness (compiler side) — Done: b6dc3e6f66, dec1bae916, d25c3eadea.
• Compiler data structures: Node, Nodes, Type, and the types package — Done: 466c948b55..72655afb4e, 6c4e90a99e, 52d9479e3b, 2e9369067b, 517b6131b2, f38f43d029, 5f525ca60d, 40f1d0ca9f, eb98e51563, 09c672d50a, e3c7497327 (typWrapper), fdf6761e01, 76e72691a0, 8640b51df8 (Type.Elem), 1624a9c9e7, 3efefd9395, e76fc1b921, 007b12977a (Maplineno out of Type), b83618f964 (Type.Nname), b2b5e779f5 (ignore receiver in Eqtype; moves receiver check into addmethod), 62e26e7511, b594b8b039, 1765863e57 (TypeList→slice), 34699bc7a8 (reduce **Node params: Template allocs −15.7%, Compiler −17.5%), 7e8e9abe0a, a4be24cbe6/edca4cda88/removing Nod(OXXX,...), 3621600c92 magic.go, 9d7b2b7b10 universe.go, 4720f49e18 relocate Type code, 11d916b157 (Node.Nointerface), e3a9dca7cc, 3da1a26fba (‘any’ no longer leaks into user package); 1220ac27ce, 68c86a0096. Left open: Type still lives in package gc.
• Compiler performance, formatting and concurrency — Done: ad4c55c076, 65b4020403, 135109d241, d447613875, 0373128318, 1a9373bc57, ea5091fcd7, f32161daf8.
• Pragmas, directives and compiler flags — Done: feb6131b1a, 3c0fae5306.
• Race detector and sanitizer instrumentation in the frontend — Done: c3b3e7b4ef.

Other changes

• SSA-era frontend adjustments ([dev.ssa] commits shipped in 1.7) internal — the SSA backend commits that touch gc as the frontend: 23df95b9b5, 7af53d98cf (OCAP), b06961b4f0, 5254be3a9c, 212a1763fc, 0ffd9c8cad, e3869a6b65 (write barriers in SSA), 8c5bfcc52b, 57670ad8b2, fbfc18c522, 18559e2da7, 5ba31940dc, 6a8a9da572, 846bc6c5ab, b64c7fc683 (“never CSE two memories”); these belong to the SSA backend report and are only noted here.
• Smaller lexer/Type refactors internal — f28bbb776a (hiter, hmap, scase definitions into builtin.go), 9a184b22ee, a689f6b8af, d481ffc1af, c1a4fe8d57, 3a0783c504, e504055e78, 5dd129bcff, e775b8df7a, 00289c296c, 7c4d53c2c8, 331f962508, 268c31870a, 390d1ce686, bf5f24b98e, 093a9a1f56, 2592e0999e, cabf73ffb8, a6c95ae1df, 55e7636f9c, 1e4e09612c, d2e1dae3fe, d1cc7f70cd, 772cea817d, 211cc40b11, ac006ad103, 37c2972772, b83f3972fe, 2faf5bca2e, c41608f4ac, 533f30b924, 868fb9ff1d, 2350ca5a41, 075d66646c, bd83cc6dae, f6bca3f32d, fe5b4a6503, ed4a27a8dd, 9c8f549abb, 1c2bdfb7d9, 88a21ebb21, 157f069873, def9c0b06d, d9bb693960, a2a48069fe, 11a80860ad, f6fab93a46, 63142027de, fc4358951a, d53287d0c3, fcca9d265c, 7e88826a69, 621aa713d4, e6066711a0, ca397bb68e, 93368be61e, 012557b376, 24967ec122, db5338f879, 6531fab06f, 0e01db4b8d, f4f1b30749, f12bd8a5a8, 1da62afeef, d3c79d324a, 733f835f30, 2da642a917, f027241445, fca0f331c8, 707aed0363, 8d075beeef, ae98045958, 980ab12ade, d57a118afa, a5386f3c7d, ae98045958, 5ec87ba554, 3747c00923, 448246adff, 85e3c9e6b8, 2f088884ae, 1f446432dd, 6e5027a37a, 0b8c0767d0, b04f3b06ec, 7d80291c4c, a4dce12803, 34699bc7a8, bd83cc6dae.
• Large map values stored indirectly internal — 60fd32a47f changes how large map values are handled (compiler decides indirect storage; companion to the mapassign work). 394ac818b0 adds an internal option, enabled, to track only named types in the type cache.
• List/Type/lexer micro-refactors internal — 6bd63ca346, 6abc8c9a88, 188e3d2515, 52f099c426, 7e92c86df2 (listsort), 4ab1793a78, a81283d5d9, e28a890d5e, c4012b6bbb, 7758a9405f, db506fe98c, 38921b36ba, 33fd4535a4, aa3650f019, b014b55b82, 0cff505871, c63dbd87c1, de4317cbd7, ec7c494535, 060a6915d4, 29267c2131, 386c0e6598, 5ba797bd18, fda831ed3f, 49e07f2b7e, 2244ae4173, 86c93c989e (no position info for builtin packages), 5621b09dad, e34295e647, 11e51ed4bc, e78ff3823e, af558acc47.
• Vendored math/big updates internal — Updated the compiler’s vendored copy of math/big. (c12e1b0b2e)
• Comment updates in gc internal — Updated old C-style comments, go.y references and added comments on how declarations and scopes work. (b330987711, cca4ddb497)
• Lexer and sinit cleanups internal — Moved the lexer into its own file, moved lexn and lexname to lex.go, cleaned up maplit, sinit.go, Isfoo type helpers and made minor cleanups. (1b417e20ff, 998b6dd75c, 76461c93a1, d33e37a7e3, b9feb91f32, 00e5a68c3e)

Middle end and SSA optimizer

From Middle end and SSA optimizer (Go 1.0–1.9)

SSA arrives (dev.ssa merged): IR, rulegen, CSE, nilcheckelim, prove/BCE and induction-variable BCE, regalloc, decompose, phi building, dominators; amd64 default (threads: SSA arrival, rewrite rules, CSE, nilcheck, prove, phi construction, dominators/CFG, write barriers).

• SSA backend arrival (dev.ssa): bring-up, coverage, default-on — Done: The whole dev.ssa history is listed under 1.7. 2015-05: global variables (23df95b9b5), code generation from SSA (083a646f63), reorganised opcode tables (b3137966db), dead-store pass (8d32360bdd), string ops (f7f604e284), block rewriting rules (a9cec30fdc), removal of the cgen pass (7bdecbf840). Summer 2015: Auxint field (8f22b5292f), per-Value line numbers (81ccf508aa), nilcheckelim (2a846d2bd3), call opcodes (290d8fc14a), Lengauer-Tarjan dominators (078ba138d3), constant bools, loops, OANDAND/OOROR (e81671115c), width/sign-specific opcodes (67fdb0de86), Const split into Const8..64 (9cb332efd4), the HTML printer (35fb514596), per-pass timing (24dcede1c0), initial likely-direction (bbf8c5ce2f), the first global register allocator replacing the per-block one (0b46b42943), first floating point (997a9f32b0), division and high-multiply (a45f2d8f28, 67cbd5b51d), decompose pass (9f954db170), complex arithmetic (3a9d0ac3c8), explicit exit blocks vs implicit ones (f5c53e0deb). Autumn: append (9d22c101f5), type assertions (269baa981e), struct/array literals (d3886906b1), stack-based pointer arithmetic without uintptr (582baae22a), arguments start spilled (02f4d0a130), better register allocation (75102afce7), panic deduplication (74e568f43a). Dec 2015–Feb 2016: compilation trigger (5b355a7907), structs SSA-able (a734bbc953), phi optimisation (b5c5efd5de), empty-block short-circuiting (3c26c0db39), &&/|| optimisation (d8a65672f8), PPARAMOUT SSA-able (6a8a9da572), constant cache (194c79c163), inlinable bool ops (8b92397bcd). The remaining 1.7-cycle work on master is in the other threads. Left open: 5b355a7907 leaves “get rid of all the UnimplementedF stuff” as a TODO; the in-tree ssa/TODO file kept the list of unfinished work (9787ba43ee, 7730880f7c, ad5ceafa2c update it).
• Generic rewrite rules and rulegen — Done: block-rewriting rules (a9cec30fdc), rulegen sanity checks and logging as above, first folding rules (9b25da732b, 198fc9a9eb, fd458ba499), constant-comparison rules (cdc36252fe, d4a95e78fa), constant-shift rules (93a0b0f315), Neg folding (964dda9bf1), strength-reduction of 64-bit constant divides copied from cgen_div (a3055af45e: TimeFormat from ~40% slower to ~10% slower than the old backend), rewrite of Phis (d0d04d2d6c), opt run twice after lowering builtins later (dfcb853d9d, which also makes SliceLen/SliceCap visible to bounds-check passes and saves ~45K in tool binaries), Phi→Or8/And8 (8b20fd000d, caef4496fc), Trunc/AND, mod and bounds folds (97b2295f06, 060501dc0e, 68325b56e7), unsafe.Pointer conversion cleanup (a3180d8b1d). Left open: 882dd1c3ba^1.8 in 1.8 shows arch rules still blocked by old backends.
• Common subexpression elimination (cse) and its scalability — Done: type-refined CSE (d9a704cd40), the phi bug (00437ebe73), key/alloc improvements (317226e61c, 851ceebceb, 8c954d5780, 6a96a2fe5a), N^2 dominator fix (2a29576562), zero-arg CSE pass (94f0245114), commutative ops (88c1ef5b45), faster initial partitions (b96189d1a0), shared dominator tree (e0611b1664), dominance sort (3ea7cfabbb, 77d374940e). Left open: 3ea7cfabbb fixes a pathological case but CSE was still super-linear (e6051de035^1.8 in 1.8).
• Nil check elimination and the redesign of nil checks — Done: the pass (2a846d2bd3), benchmarks (1746e711ad), DFS rewrite (4dcf8ea1a4), non-nil sources (b14b8d6bc7, e0e556620a, 3c1a4c1902), faulting-load checks (31115a5c98, 4a346e7489), optimise OCHECKNIL use (e22ae879d3), opt-time simplification (fb2f99d5fd), no simplification in loops (c1892b9c4b), nilcheck for interface go/defer calls (f3689d1382). Left open: intra-loop use of facts was disabled.
• Prove pass and bounds-check elimination — Done: bdea1d58cf, 34f048c9d9, b91cc53033, cd798dcb88, 47c9e139ae, 27ebc84716, 6c6089b3fd, 1747788c56, 9549c06ce6 (IsInBounds with small index), bbd3ffbd83, 68325b56e7. Left open: derived induction variables; pointer-increment form of range loops.
• Escape analysis (esc.c → esc.go) — Done: 2d56dee61b, d8c815d8b5, d603c27c6b, fb2af2b35b (don’t walk field-name syntax), 7e88826a69 (look at function decls instead of OTFUNC). Left open: data structure for reasons “might be extendable to allow a cleanup of the underlying algorithms”.
• Inlining (inl.c → inl.go) — Done: e41f527f4d, 060038bdd0, e6ea01685f, ed4a27a8dd. Left open: break/type switch.
• Write-barrier insertion: from gc to the SSA writebarrier pass — Done: aebf6611df, 3e54ca9a46 (omit barrier on global function), 15ed37d7b7 (enforce nowritebarrier in SSA), 42da35c699, 6b33b0e98e, a4650a2111, 03e216f30d, 934c359964, 8107b0012f. Left open: barrier decisions still in the front end.
• Switch, type-switch and select lowering — Done: e41f527f4d makes switches inlinable. Left open: nothing noted.
• Interface conversions, type assertions and devirtualization — Done: 7427f2c4bd, 5c5e8d4105. Left open: nothing noted.
• Phi construction and the SSA builder’s scalability — Done: b5c5efd5de, 6b99fb5bea, bc1989f115, 8b20fd000d. Left open: old algorithm remains for small functions.
• Dominators, loops, CFG editing and block-level passes — Done: 078ba138d3, d32229b3b1, 0b6332eb54, 4fa050024f, 6ed79fbd1a, 9743e4b031, d610d304f8, f91ff1a509, c67cac0703, 3c26c0db39, 6b3462c784, c03ed491fe, 6b85a45edc, 815c9a7f28, 305a0ac123, e197f467d5, bbf8c5ce2f, 186cf1b9ba. Left open: nothing noted.
• Compile-time performance and determinism of the SSA package — Done: the items above plus pass stats (a3f72956f1), 24dcede1c0. Left open: nothing noted.
• Lowering of copies, strings and slices; intrinsics — Done: 995fb0319e (stringtoslicebytetmp fix), b024ed0d94. Left open: nothing noted.

Other changes

• Front-end containers refactor (NodeList → Nodes / nodeSeq) internal — 466c948b55, 7047742f12, 132ebeac3f, eb6ee75add, cd6619d2e6, ed4a27a8dd, 89cfdda44d, e28a890d5e, ec7c494535.
• Algs for tiny types internal — 4cef0e980a, b6b144bf97.
• SSA bring-up fixes (dev.ssa) internal — long tail of per-construct implementation commits (strings, complex numbers, OLEN, ODOT, OCALLMETH, defer, etc.), e.g. f7f604e284, cc3f031a31, 15dcdfba0f, a3f72956f1, 3a9d0ac3c8, 1792b36310, fa5fe19137, 9aba7e76f6, ddc6b64444.
• GOSSA debugging internal — d298209b1c, 378a863682 (command-line option processing for SSA), c0740fed37, 09a9ce60c7, 13f74db304.
• DWARF naming of compound AUTOs internal — 4a7aba775d, 259b7edf5c, b04e145248: strings/slices/interfaces named by splitting into one-word autos so liveness can handle them.
• Slice zero-cap detection internal — 69a7c152a7.
• dev.ssa per-construct implementation, debugging aids and small SSA fixes (175 commits) internal — the long tail of the SSA bring-up: implementing individual node kinds (OSLICESTR, OCOMPLEX, OFALL, OCFUNC, ORETJMP, OGETG, OSPTR, PPARAMREF, len/cap of maps and channels), floating-point and complex comparison fixes, defer/return handling, -N fixes, logging and HTML printer tweaks, TODO updates (bd95412d23, e00d60901a, 6f1884757f, ba8a146af4, d5ad9ced1c, 0ad9c8c720, 47791c1a18, ea5cd68646, 3b817ef8f8, 703ef06039, a5c3b6642b, 2aabacdb5a, 8c46aa5481, 37ddc270ca, 7b858018b3, 66b47812eb, dbd83c4464, 596ddf4368, c9372619f2, be2a3e2ac9, 1edf4897df, 0a133cddd5, d465f049cd, 6c14059e65, 7d10a2c04a, 5173868325, 06f329220f, 050ce4390a, 4c521ac8f2, 7e4c06dad0, c3c84a2544, 337b7e7e3b, b383de2ef9, 3dcc424be7, cd7e059496, 766bcc92a5, 2574e4ac1c, 21bd483c4a, 26f135d7c1, 61aa0953e5, 8fb6358153, e61e7c96f7, be1eb57a8b, 5254be3a9c, d90e0481bf, 52d76f7a6a, 8d31df18af, 5c5f2a731c, 0bb2a50a55, b61f8c8df3, 9ca24fcda8, 67df7934cc, 25d1916816, 1bab5b9b41, 20550cbaf1, 9271ecc8c6, 6d9362a1f7, 165c1c16d1, 93c354b6cc, 38ed6c10eb, b8efee0d8a, 683f5d7bdb, 54dca047dd, 573c791e81, 9495e45c80, d1c15a0e3e, cfd8dfaa10, e13954981e, 5584523224, dee1f2750b, 2af0648014, d56d2fa206, a2d1580fdb, 3d23afb913, 867662da6a, 3e7904b648, d4cc51d411, 759b9c3b80, d9dc2f2230, 8e601b23cd, 7393c24877, 0ffd9c8cad, 3526cf586b, a2f8b0d0e4, 752fe4dcb5, e0e4068c6e, 73151067bc, 35ad1fcb11, 44d87035b6, 525785885e, f8093b8f24, 67ac8a3b22, a5cffb6182, 707af252d9, 67e43c1e3b, 47d6799b0f, 9654873a76, 0726931c10, 5fdd4fea90, ec8a597cd2, 65677cabfd, 19285efaed, 10f38f51ef, d9f2cafb50, 991036aef3, 19447a66d6, 4178f20776, c3eb1a7e8a, 3fa0a75a02, def7c65b70, ca9e450bed, c684d4d26c, a7cfc759f2, fd8c71be86, 8a1f6217c5, cea441427e, 2f51807157, db380bf44b, c244ce097c, c7081409bb, 1e4ebfdda5, 7e390724d2, 46ffb026b4, fda72e0375, 5505e8ccc7, 5f10573e60, b32217a273, 37590bddc4, d29e92be52, 579ccd831f, 808d7c70d5, 8824dccc28, 32ffbf7e0f, 9703564c9a, d076ef749b, d694f83c21, 10462eb30f, cd01c0be26, 129261ae83, 74180dd7da, 991873116e, d19bfc3b68, 170589ee1c, 7807bda91d, 9094e3ada2, 035fcc0c4d, da8af47710, 90065eaba4, 733bf6ef67, 8a961aee28, 16b1fce921, 955749c45f, 2df4b9c265, a0da2d242c, e93410d3e5, faf1bdb42b, 58cfa40419, 7f7f7cddec, 65855cf640, 9dc1334cc7, bc1fb32e9d, e4bee4be92, ae276d8c23, c17b6b488c, 40f2b57e0b, 8906d2a171, d337e55672, e173ab1434, ed737fd8cd, a5325761cd, fb54e0305f, 91f69c675d, 1b1d0a9a80, 4e95dfed01, f1f366c1e7)
• Compiler fixes and refactors (56 commits) internal — small fixes, line-number repairs, node-API tidy-ups and diagnostics in gc and the ssa package that do not carry separate design (bcce5bdabd, 26594c3dfd, 59681802ad, 88e18032bd, 8b4deb448e, 3a880bae91, c41608f4ac, 71d13a8c01, 16e2e95dd9, e3a9dca7cc, dafbcf6df7, 4c8589c328, df035a8d86, 0321cabdfa, 31d13f479a, d20b92e07a, 40bfec0022, 8ec80176d4, 98b88de56f, 1b9f168f73, 8edb72587f, f6fab93a46, ac74e5debc, a9407b5797, d4663e1353, a4be24cbe6, 7e8e9abe0a, babc73547e, 688840593b, d7f7ea6ea8, 92bb694a49, 75a22d0d10, 309144b7f1, 5e1b7bdecf, 7208a2cd78, 68ac1f7746, 7f53391f6b, 811ebb6ac9, 6b0b3f86d6, 644493f109, 508a424eed, 3c6e60c0e4, bfe0cbdc50, c65647d620, 4938d7b5fc, e48434887e, 7879e9193b, 3b0efa689e, a19e60b2c3, 5325fbc7db, b13b249f43, 24744f6561, 9d7c9b4384, 0e13dbc1a9, 29f0984a35, 846bc6c5ab)
• Early dev.ssa cleanups internal — Handled more instructions in the early SSA package and reorganized it. (cfc2aa56b0, 247786c174)
• SSA and inlining cleanups internal — Some SSA cleanup, fixed misaligned comments, removed unused code and simplified inlnode. (62ac107a34)
• uintptrescapes directive user-visible — Added the //go:uintptrescapes comment directive and used it in syscall. (bbe5da4260)

Backends and assemblers

From Backends and assemblers (Go 1.7–1.11)

The dev.ssa SSA backend for amd64 lands (IR to SSA, generic and AMD64 rules, global register allocator, tighten, flag recomputation, scheduler), together with intrinsics, load combining, and a slew of cmd/internal/obj cleanups (As type, ADATA removal, cmd/internal/sys); s390x object/assembler support arrives and the ARM SSA port is started. Threads: SSA bring-up, register allocation, scheduling, obj refactoring.

• SSA backend bring-up on amd64 (dev.ssa merge) — Done: user-visible the SSA path exists and generates amd64 code: conversion from IR (d2fd43aa77), codegen from SSA (083a646f63), memory zeroing (66b47812eb), static data (6b41665039, 7c18f8cdc3 writes static data straight into the linker symbol instead of ADATA Progs), strength reduction of constant multiplies/divides (3e7e519c36, d8f1f8d856, a3055af45e for 64-bit constant divides, 77b527e283 for 24), const folding for floats (f6ceed2cab), bool ops (8b92397bcd), intrinsics (8eec2bbfbc). The compilation trigger changed from “compile everything with SSA and decide whether to use it” to deciding up front which functions go through SSA (5b355a7907). Compile-time work made the pass pipeline cheap: sparse sets cached on Func and reused across passes (f962f33035, 3297a4f5f3), preallocated values/blocks and backing store for Preds/Succs/Values (2f57d0fe02, 056c09bb88), faster CSE (6a96a2fe5a, 88c1ef5b45 handling commutative ops), value use counts kept in SSA (56e0ecc5ea), critical-edge pass tweaks (50bc546d43, b2dc1f82a5). Bugs of the period (flags clobbered by SETcc/REPSTOSQ/LEA, e.g. 8f51ae8ba5, f3171994e9, 4219aba5db, 2511cf03b9) show the flag handling being worked out empirically. Also: ba8a146af4 prints register names, 288817b05a reduces line-number churn, 686fbdb3b0 makes compilation deterministic. Left open:* only amd64 is supported (4c9a470d46 “start on ARM port” at the end of the cycle); the old cgen-based backend still exists and is the fallback, and UnimplementedF stubs remain (5b355a7907 TODO: “get rid of all the UnimplementedF stuff”).
• Register allocation, flag allocation and stack slots — Done: global allocator (0b46b42943), reserved registers (7d61246972), rematerialization (c64a6f6362), stack-slot reuse (366dcc4529), separate flag allocation (c140df0326) and spilling flags by recomputation (9f8f8c27dc, d04f38e3ee), allocation-order change (7d9f1067d1), two-address instruction allocation (369f4f5de5, 0004f34cef), loop-aware and forward-looking tweaks (815c9a7f28, 9d6e605cf7, b57ac33331), moving constant phi arguments closer to the phi (305a0ac123), 3572c6418b keeping pointer input arguments live. Left open: spills always placed right after the defining op; stack-slot assignment still coarse.
• Scheduling, tighten, block layout and CFG passes — Done: scheduler fixes (4e204b42f5, a678a5c7a5, e98edc88c9), tighten (f91ff1a509), priority-queue scheduler (b1785a5065), critical edges (a2f8b0d0e4), constant-time CFG editing (4fa050024f), exit vs return blocks (10f38f51ef), load-of-store replaced with copy (5c5fa3628c), prove pass extended to constant comparisons (47c9e139ae). Left open: tighten ran only before lowering.
• Rewrite-rule machinery and compile-time performance of the SSA compiler — Done: rule sanity checks (e4355aeedf), strength-reduction rule generalizations (3e7e519c36, d8f1f8d856), 1cc5789df9 many small rewrites, 0fb818a424 copy values during rewrites, 4989337192 allow control values to be CSEd, 16b1fce921 aux typing. Left open: both operand orders written out by hand.
• Multi-architecture SSA ports: ARM, 386, ARM64 (dev.ssa) — Done: 4c9a470d46 starts ARM port; 7c6b48ffba fixes comparison and conditional branch for SSA on ARM; 74a9bad638 enables const division on arm64 (-67% for DivconstI64); 52fe472472 zeroes unaligned memory a byte at a time on ARM; f3d54789f7 power5 instructions for uint64 to float on ppc64; 1f4f2d0d39, bf151cc2aa unsigned to float fixes. Left open: all ports other than amd64 still on the old backend.
• Multi-architecture SSA ports: PPC64, MIPS64, MIPS32, s390x — Done: s390x assembler/object/asm/ssa/gc support (2a7e85f162, 13f97ea456, 85bbabd9c4, 354e9778a3, a223ccae38, 613ba6cda8, cd6b2b7451, 3c8ef0e0c9 64-bit multiplication with immediates); MIPS64 REGTMP/SB/TLS work (a409fb80b0, 8dc0444a04, 981395103e, 3b0b30727c); ppc64x atomics (eeca3ba92f, with sync moved outside the lwarx/stwcx loop). Left open: s390x and MIPS64 only reach SSA in the next release.
• Removal of the legacy backend and decoupling gc from the code generators — Done: amd64 code removed from gc (a4e31d42ee); a40b27e304 eliminates arch-specific typedefs; cea5d26aa2 removes Thearch.Linkarchinit. Left open: old backend still compiles everything not in SSA.
• Liveness, stack maps, safepoints and write barriers — Done: d5297f7261 fixes quadratic regalloc liveness (slice3.go >10m to ~20s), d2107fc987 GC bitmaps for SSA code, 4a7aba775d splitting compound autos, 3572c6418b pointer args live throughout, 6effdd28de, 519b469795, f4d38a8792 de-dups gclocals symbols. Left open: liveness still operates on Progs.
• Intrinsics for atomics, math/bits, math and runtime helpers — Done: first intrinsics (8eec2bbfbc). Left open: bit-scan family incomplete (8eec2bbfbc).
• Memory-op combining, byte-order idioms and load/store folding — Done: load combining (12e60452e9, 7294ccb54e, b81f2f106f, 481fe59012), store combining (217c284995), idx1 loads/stores (af517da2f9), ec5083e49b SHLQ into loads/stores, 3dd4f74e06 shifts into LEAs, 036d09d5bf/dc5a7682f0/9e3c68f1e0 32-bit operands for byte ops, drop most byte/word insns. Left open: bounds checks of combined loads separate issue (12e60452e9).
• amd64 and 386 instruction selection — Done: the items above; 6b02a19247 SSE4.2 strings.Index; 009c002c92 AVX2 for sha. Left open: indexed forms incomplete.
• Memory clearing and copying: Zero, Move and Duff’s device — Done: 66b47812eb, f8fc3710fd, 039d455f9d, d55a099e22, 111d590f86. Left open: Duff’s device unsafe interplay with the allocator.
• cmd/internal/obj: Prog/LSym refactoring, concurrent backend and object-writing — Done: As type (0d9258a830, e853131699, 7893442011, 4fc02d175b), ADATA removal (2dcbbbd193, 4f753e77f1, 7c18f8cdc3), object file symbol refs (3648d2d4cd), c6e11fe037 new cmd/internal/sys common architecture representation (b17b95301a ArchFamily iotas) replacing the ‘8’,‘6’ constants, remove LSym fields (61b9315d37, dd2ba0c7a7, 31a9e50524), Prog cache (f388b58540, 45c2e38b37). Left open: assemblers use global Link state.
• x86 assembler: AVX, AVX2, AVX-512 and the instruction table — Done: 76f272d717, 0b4f578266, 009c002c92, 8262564971, 7de8cfdf9c dynlink rewrite no longer clobbers flags. Left open: AVX coverage partial.
• Prologues, frame pointers, NOFRAME and tail calls — Done: 7fdec6216c, d9557523c2, 26849746c9 fixes for defer in 386 shared libraries.

Other changes

• Compiler data-structure and memory work internal — reuse of []Flow (028247d2cd, -14.8% alloc for Template), Prog cache (f388b58540), Dcl/Inldcl as slices (b66a892358), shrink tables (0ac0e22fa7), int32 line numbers (b83f3972fe), LSym.RefIdx packing (826831acf7), deterministic compilation (686fbdb3b0).
• Sqrt on Power user-visible — math.Sqrt uses the sqrt instruction on ppc64/ppc64le (b4b2ddb867).
• strings.Index with SSE4.2 user-visible — (6b02a19247).
• GOSSAFUNC HTML internal — improved readability (b13b249f43).
• s390x load/atomic instructions internal — MULHD (cd6b2b7451).
• cmd/asm internal — removal of nonexistent instructions (0b4f578266), EOF message (45c4ebec5b), arm64 SIMD register name (c9389a101b).
• Stack guard internal — SSA shrinks the stack guard (d3f15ff6bc).
• Inlined IR bug fixes internal — pointless gotos while inlining (d1faf3879e^1.8, 1.8).
• dev.ssa feature-by-feature lowering internal — the incremental steps that took SSA from a prototype to a compiler: line numbers on Values (81ccf508aa, 3a0df1f82a, 0991ec454b), call and comparison ops (290d8fc14a, 46815b9f62, 71b570774d, 9f8a677eab, 9ca24fcda8), 64-bit shifts and small-int ops (6f1884757f, 73054f5729, 20550cbaf1, 4b803151ce, c55870a541, d2150c8328), bool and logical ops (d9c72d739c, e81671115c, 8adc905a10, edff881ce5, 7402416a8b, 954d5ada29), multiply/divide/mod/hmul (be1eb57a8b, a45f2d8f28, 67cbd5b51d, 57d9e7e3c4), sqrt, itab, getg, slicestr (a329e21ccd, d1c15a0e3e, 3d23afb913, 3526cf586b), SETcc/branch/idx genValue (a794074da0, a402b58e51, 3e3d162f50), AMD64 SETEQ/InvertFlags rewrites (71a4c4bb8d, c15b0e83d2), PAUTO/PHEAP variables and variable handling (c31b6dd0be, 8c46aa5481), zero type (a92bd66282), constant-fold fixes (3b705824ce, 7a6de6d5f2, 198fc9a9eb, 9b25da732b, c1593da817, db5232620a, 7a982e3c49), d4cc51d411 explicit size for store ops, 95aff4db54 Copy instead of ConvNop, 85e0329fbc moving most types out of SSA, 65677cabfd default op types, 19447a66d6 floats stored in AuxInt, 634b50c6e1 truncating float to int, e99dd52066/10462eb30f OpConvert and better copying, d43f2e37ed/a6fb514bf8 storeconst ops, 582baae22a/7807bda91d pointer arithmetic with int and uintptr/unsafe.Pointer safety, 463858e6ff/8d23681cc8 nil-check and panic support.
• dev.ssa regalloc/liveness debugging and correctness internal — cff0c6ad0f, a0022d9b8c, 763afe13b9, 6b3d4a5353, 32efa16c3d (instrumentation and stats), 67bfd69564, 0bb2a50a55, c9a38ce758, 9094e3ada2, cd0cb0a9d6, 2dc88eead8, d694f83c21, 7cadf23afb, 6b0b3f86d6, 8ad8d7d87e, 3c09001917, 7b773946c0 (xor clearing disabled when flags are live), 3a17fdaba0, 8dc04cbedc, ca088cf4e5, 2efdaefdb0, 47d6799b0f, 6411533ebf, 5cb352edeb, 220e705404, e514021153, c24681ae2e, 01490eb96c, 9f66636c93, ac8127d7e6, ade0eb2f06, 3dcc424be7, 8fb6358153, bc1f5768ce, cea441427e, cde977c23c, 6d40c62732, 789c5255a4^1.9.
• -N and tooling behavior internal — keep JMPs with -N (8906d2a171, e360f7c4db), line numbers for Unimplemented/Fatal (da8af47710), c17b6b488c truncate AuxInt in Prog, e49c910900, 88b230eaa6 do-log boolean, 9787ba43ee TODO list, 270e2152c4, 50ceef26e6, fb54e0305f, 8c6abfeacb separates logging, work-in-progress and fatal errors so SSA can be attempted for every function (turning the standard library into a compilation test), 37ddc270ca, 2df4b9c265, 43a2676ff0, 703ef06039.
• Allocation and cleanup in the compiler internal — f3a29f1f81 preallocated storage for three Value args, ca5417b8e0, b86cafc7dc, 6cb2e1d015, e6beec1fc8, 8edb72587f, ddc6b64444, beabd872f4, cc158403d6 (track reflect.Type.Method in deadcode), afad74ec30, 7e40627a0e, 5609a48931, 2a4158207e, 613ba6cda8 (s390x in gc), bf3909824c, def1e7276a, c3dfad5df9, d20b92e07a, def9c0b06d, e59c1729ba, a4dce12803, 361b334cbd, e76fc1b921, 38e11d05b9, 786a2ef380, 8f2e780e8a, 1441f76938, e07a4459a1, d0fb649713, 70090654de, 41f9f6f471, 68e86e6dfa, 7177cb9fa4, 080e2d4320, 7735dfb67c.
• obj minor cleanups internal — 14113b3a89, c747fce242, d05773ce73, edde955d7f (arbitrarily sized string data), af1c29c1c1, 04f23b6428, 63142027de, fb950cd778, 3504945081, 39af1eb96f, ca0f5c9740, b07a214d39, 2326c24cc7, 71ab3c1ccf, 045411e6f2, babfb4ec3b, 060a2bacb7 (eliminates ~700 conversions on ppc64), b4c66b176e, 9c269e6a46.
• Backend cleanups internal — Style cleanups in obj and link, post-CL cleanups, peep.go and bv.go cleanups. (106b9d3915, a162d11540, dbe54d23fe, 5edcff0134)

Linker and object files

From Linker and object files (Go 1.0–1.9)

Linker speed/memory campaign, new object format with reference indexes and contiguous data, string merging and method pruning, typelink offsets, cmd/newlink removed, deterministic builds, s390x.

• Go symbol table, pc/line table and function metadata (gosymtab, pclntab, findfunc) — Done: typelinks stored as offsets (f028b9f9e2), padding between functions rather than inside them (5701174c52), short type symbol names (f81ae3b22c).
• Dead code and data elimination — Done: method pruning (862b9ddda7, e2836935bb, d06b0db5bd), unused go.string.hdr symbols (1.8, 80a034642e^1.8), -dumpdep flag for the dependency graph (d224e98d9a) and Allsym no longer contains .dup symbols (956e9e6c83). Left open: method pruning is all-or-nothing on reflect.Value.Call.
• DWARF generation in the linker — Done: DWARF from symbols (a037c73ccf), own hashmap replaced (c0f7195d62), PC table without seeking (3601a1e77d), framepointer offsets fixed (adff422779), hide go.dwarf symbols (9658b7ef83), DWARF disabled without symtab (e6181eb9e1), .def suffix collision (6901b08482).
• External (host) linking and cgo — Done: deterministic output: pass -gno-record-gcc-switches and -fdebug-prefix-map=$WORK=/tmp/go-build, empty ELF NAME symbol (5bbb98df09); -znocopyreloc (3c8d6af8e0), gold on arm64 (a3c92c9db1), -no-pie for race (3a72d626a8, cb97fd7741, 2a6544b604), no PIE with sanitizers (2c1791b13b), -z relro for PIE (45f39fb467), mips64 (073d292c45).
• Build modes: -shared, c-archive, c-shared, shared, PIE, plugin — Done: c-archive on windows/darwin text relocation fix (ed8f0e5c33, aea4222f67), mips64 external (073d292c45), -linkobj (feb6131b1a, compiler flag splitting export data from link data), PIE/sanitizer interplay (2c1791b13b).
• liblink, the object file format and the move into the compilers — Done: the size/speed changes above; Biobuf code extracted to cmd/internal/bio (4b7e36cdfe, 8f2edf1199).
• From C to Go: cmd/link — Done: delete cmd/newlink (fb880b8a03); cleanups and concurrency: alignment concurrently (5a0881a1d1), data symbol processing with slices (ed41054b6d).
• Linker speed and memory (Go 1.7 focus) — Done: the micro-optimizations above. Left open: nothing noted.
• Runtime type data, GC metadata and sections laid out by the linker — Done: offsets (f028b9f9e2), go.string handling, itablinks slice (79688ca58f).
• Container writers: Windows PE, Mach-O and Plan 9 — Done: PE symbol table rewrite (bb48b864b1).
• Architecture and OS ports in the linker (arm, power64, arm64, mips, s390x, nacl, BSDs, Solaris) and trampolines — Done: s390x (163ff700b8, 29a6149e67), mips64 external, Minalign (90cb1fdaf2), sys.Arch common representation (c6e11fe037, d481ffc1af, 22ef687da8), gdb backtrace on link register architectures (55154cf0b2), dynamic symbol size for 386 (0b6e5e3d73).
• Build IDs, ABI hashes, flags and determinism — Done: user-visible deterministic cgo builds (5bbb98df09), -dumpdep (d224e98d9a), -H elf removed (71ab3c1ccf).

Other changes

• Misc internal — 1012892f1e remove C-calling check, d0fb649713 operator &^, acc757f678 SeekStart, fe5be5aced, ec3c5b9d17 elf64/rela switches, 1934a77f5a, 79048df2cc long symbol names, 96b8f70e22, b0cbe158da, 66afbf1010, 9568d54fb8 plt/rel ordering, 4b175fd23b, 25d95ee918, 1dad218da1, 854ab14b7e, 79c527f4a7, d7cdf66978 elfhash garbage, 9629f55fbb darwin/arm c-archive absolute address, 3c4ebd2023 mips64 extld argument, a0abecf102 and f2f3b6cd8f gold detection, e79fef8e55 Solaris, 3375974e0d short buildid, 2f41edf120 dynimport type data, 438ce713a1, cb97fd7741, 6fee4aa5c7, 9d4efdfd12, de7ee57c7e, 2ac8555d57, 1441f76938, 90a59d448e.
• Remaining 1.7 linker cleanups internal — dd0a128a02, 021e0c5f1f, 1661493c73 (queue as slice), 4c69e92f51, 74a3b205eb, 2e90192b0e, 1b6402ea9d, 8677cad199, bc94282306.
• Linker allocation and main internal — Skipped allocation reading symbol names, skipped arch-specific main, and removed unnecessary comments. (07ccc21295)