| // Copyright 2015 The Go Authors. All rights reserved. |
| // Use of this source code is governed by a BSD-style |
| // license that can be found in the LICENSE file. |
| |
| package ssagen |
| |
| import ( |
| "bufio" |
| "bytes" |
| "fmt" |
| "go/constant" |
| "html" |
| "internal/buildcfg" |
| "os" |
| "path/filepath" |
| "sort" |
| "strings" |
| |
| "cmd/compile/internal/abi" |
| "cmd/compile/internal/base" |
| "cmd/compile/internal/ir" |
| "cmd/compile/internal/liveness" |
| "cmd/compile/internal/objw" |
| "cmd/compile/internal/reflectdata" |
| "cmd/compile/internal/ssa" |
| "cmd/compile/internal/staticdata" |
| "cmd/compile/internal/typecheck" |
| "cmd/compile/internal/types" |
| "cmd/internal/obj" |
| "cmd/internal/src" |
| "cmd/internal/sys" |
| |
| rtabi "internal/abi" |
| ) |
| |
| var ssaConfig *ssa.Config |
| var ssaCaches []ssa.Cache |
| |
| var ssaDump string // early copy of $GOSSAFUNC; the func name to dump output for |
| var ssaDir string // optional destination for ssa dump file |
| var ssaDumpStdout bool // whether to dump to stdout |
| var ssaDumpCFG string // generate CFGs for these phases |
| const ssaDumpFile = "ssa.html" |
| |
| // ssaDumpInlined holds all inlined functions when ssaDump contains a function name. |
| var ssaDumpInlined []*ir.Func |
| |
| func DumpInline(fn *ir.Func) { |
| if ssaDump != "" && ssaDump == ir.FuncName(fn) { |
| ssaDumpInlined = append(ssaDumpInlined, fn) |
| } |
| } |
| |
| func InitEnv() { |
| ssaDump = os.Getenv("GOSSAFUNC") |
| ssaDir = os.Getenv("GOSSADIR") |
| if ssaDump != "" { |
| if strings.HasSuffix(ssaDump, "+") { |
| ssaDump = ssaDump[:len(ssaDump)-1] |
| ssaDumpStdout = true |
| } |
| spl := strings.Split(ssaDump, ":") |
| if len(spl) > 1 { |
| ssaDump = spl[0] |
| ssaDumpCFG = spl[1] |
| } |
| } |
| } |
| |
| func InitConfig() { |
| types_ := ssa.NewTypes() |
| |
| if Arch.SoftFloat { |
| softfloatInit() |
| } |
| |
| // Generate a few pointer types that are uncommon in the frontend but common in the backend. |
| // Caching is disabled in the backend, so generating these here avoids allocations. |
| _ = types.NewPtr(types.Types[types.TINTER]) // *interface{} |
| _ = types.NewPtr(types.NewPtr(types.Types[types.TSTRING])) // **string |
| _ = types.NewPtr(types.NewSlice(types.Types[types.TINTER])) // *[]interface{} |
| _ = types.NewPtr(types.NewPtr(types.ByteType)) // **byte |
| _ = types.NewPtr(types.NewSlice(types.ByteType)) // *[]byte |
| _ = types.NewPtr(types.NewSlice(types.Types[types.TSTRING])) // *[]string |
| _ = types.NewPtr(types.NewPtr(types.NewPtr(types.Types[types.TUINT8]))) // ***uint8 |
| _ = types.NewPtr(types.Types[types.TINT16]) // *int16 |
| _ = types.NewPtr(types.Types[types.TINT64]) // *int64 |
| _ = types.NewPtr(types.ErrorType) // *error |
| types.NewPtrCacheEnabled = false |
| ssaConfig = ssa.NewConfig(base.Ctxt.Arch.Name, *types_, base.Ctxt, base.Flag.N == 0, Arch.SoftFloat) |
| ssaConfig.Race = base.Flag.Race |
| ssaCaches = make([]ssa.Cache, base.Flag.LowerC) |
| |
| // Set up some runtime functions we'll need to call. |
| ir.Syms.AssertE2I = typecheck.LookupRuntimeFunc("assertE2I") |
| ir.Syms.AssertE2I2 = typecheck.LookupRuntimeFunc("assertE2I2") |
| ir.Syms.AssertI2I = typecheck.LookupRuntimeFunc("assertI2I") |
| ir.Syms.AssertI2I2 = typecheck.LookupRuntimeFunc("assertI2I2") |
| ir.Syms.CgoCheckMemmove = typecheck.LookupRuntimeFunc("cgoCheckMemmove") |
| ir.Syms.CgoCheckPtrWrite = typecheck.LookupRuntimeFunc("cgoCheckPtrWrite") |
| ir.Syms.CheckPtrAlignment = typecheck.LookupRuntimeFunc("checkptrAlignment") |
| ir.Syms.Deferproc = typecheck.LookupRuntimeFunc("deferproc") |
| ir.Syms.Deferprocat = typecheck.LookupRuntimeFunc("deferprocat") |
| ir.Syms.DeferprocStack = typecheck.LookupRuntimeFunc("deferprocStack") |
| ir.Syms.Deferreturn = typecheck.LookupRuntimeFunc("deferreturn") |
| ir.Syms.Duffcopy = typecheck.LookupRuntimeFunc("duffcopy") |
| ir.Syms.Duffzero = typecheck.LookupRuntimeFunc("duffzero") |
| ir.Syms.GCWriteBarrier[0] = typecheck.LookupRuntimeFunc("gcWriteBarrier1") |
| ir.Syms.GCWriteBarrier[1] = typecheck.LookupRuntimeFunc("gcWriteBarrier2") |
| ir.Syms.GCWriteBarrier[2] = typecheck.LookupRuntimeFunc("gcWriteBarrier3") |
| ir.Syms.GCWriteBarrier[3] = typecheck.LookupRuntimeFunc("gcWriteBarrier4") |
| ir.Syms.GCWriteBarrier[4] = typecheck.LookupRuntimeFunc("gcWriteBarrier5") |
| ir.Syms.GCWriteBarrier[5] = typecheck.LookupRuntimeFunc("gcWriteBarrier6") |
| ir.Syms.GCWriteBarrier[6] = typecheck.LookupRuntimeFunc("gcWriteBarrier7") |
| ir.Syms.GCWriteBarrier[7] = typecheck.LookupRuntimeFunc("gcWriteBarrier8") |
| ir.Syms.Goschedguarded = typecheck.LookupRuntimeFunc("goschedguarded") |
| ir.Syms.Growslice = typecheck.LookupRuntimeFunc("growslice") |
| ir.Syms.Memmove = typecheck.LookupRuntimeFunc("memmove") |
| ir.Syms.Msanread = typecheck.LookupRuntimeFunc("msanread") |
| ir.Syms.Msanwrite = typecheck.LookupRuntimeFunc("msanwrite") |
| ir.Syms.Msanmove = typecheck.LookupRuntimeFunc("msanmove") |
| ir.Syms.Asanread = typecheck.LookupRuntimeFunc("asanread") |
| ir.Syms.Asanwrite = typecheck.LookupRuntimeFunc("asanwrite") |
| ir.Syms.Newobject = typecheck.LookupRuntimeFunc("newobject") |
| ir.Syms.Newproc = typecheck.LookupRuntimeFunc("newproc") |
| ir.Syms.Panicdivide = typecheck.LookupRuntimeFunc("panicdivide") |
| ir.Syms.PanicdottypeE = typecheck.LookupRuntimeFunc("panicdottypeE") |
| ir.Syms.PanicdottypeI = typecheck.LookupRuntimeFunc("panicdottypeI") |
| ir.Syms.Panicnildottype = typecheck.LookupRuntimeFunc("panicnildottype") |
| ir.Syms.Panicoverflow = typecheck.LookupRuntimeFunc("panicoverflow") |
| ir.Syms.Panicshift = typecheck.LookupRuntimeFunc("panicshift") |
| ir.Syms.Raceread = typecheck.LookupRuntimeFunc("raceread") |
| ir.Syms.Racereadrange = typecheck.LookupRuntimeFunc("racereadrange") |
| ir.Syms.Racewrite = typecheck.LookupRuntimeFunc("racewrite") |
| ir.Syms.Racewriterange = typecheck.LookupRuntimeFunc("racewriterange") |
| ir.Syms.WBZero = typecheck.LookupRuntimeFunc("wbZero") |
| ir.Syms.WBMove = typecheck.LookupRuntimeFunc("wbMove") |
| ir.Syms.X86HasPOPCNT = typecheck.LookupRuntimeVar("x86HasPOPCNT") // bool |
| ir.Syms.X86HasSSE41 = typecheck.LookupRuntimeVar("x86HasSSE41") // bool |
| ir.Syms.X86HasFMA = typecheck.LookupRuntimeVar("x86HasFMA") // bool |
| ir.Syms.ARMHasVFPv4 = typecheck.LookupRuntimeVar("armHasVFPv4") // bool |
| ir.Syms.ARM64HasATOMICS = typecheck.LookupRuntimeVar("arm64HasATOMICS") // bool |
| ir.Syms.Staticuint64s = typecheck.LookupRuntimeVar("staticuint64s") |
| ir.Syms.Typedmemmove = typecheck.LookupRuntimeFunc("typedmemmove") |
| ir.Syms.Udiv = typecheck.LookupRuntimeVar("udiv") // asm func with special ABI |
| ir.Syms.WriteBarrier = typecheck.LookupRuntimeVar("writeBarrier") // struct { bool; ... } |
| ir.Syms.Zerobase = typecheck.LookupRuntimeVar("zerobase") |
| |
| if Arch.LinkArch.Family == sys.Wasm { |
| BoundsCheckFunc[ssa.BoundsIndex] = typecheck.LookupRuntimeFunc("goPanicIndex") |
| BoundsCheckFunc[ssa.BoundsIndexU] = typecheck.LookupRuntimeFunc("goPanicIndexU") |
| BoundsCheckFunc[ssa.BoundsSliceAlen] = typecheck.LookupRuntimeFunc("goPanicSliceAlen") |
| BoundsCheckFunc[ssa.BoundsSliceAlenU] = typecheck.LookupRuntimeFunc("goPanicSliceAlenU") |
| BoundsCheckFunc[ssa.BoundsSliceAcap] = typecheck.LookupRuntimeFunc("goPanicSliceAcap") |
| BoundsCheckFunc[ssa.BoundsSliceAcapU] = typecheck.LookupRuntimeFunc("goPanicSliceAcapU") |
| BoundsCheckFunc[ssa.BoundsSliceB] = typecheck.LookupRuntimeFunc("goPanicSliceB") |
| BoundsCheckFunc[ssa.BoundsSliceBU] = typecheck.LookupRuntimeFunc("goPanicSliceBU") |
| BoundsCheckFunc[ssa.BoundsSlice3Alen] = typecheck.LookupRuntimeFunc("goPanicSlice3Alen") |
| BoundsCheckFunc[ssa.BoundsSlice3AlenU] = typecheck.LookupRuntimeFunc("goPanicSlice3AlenU") |
| BoundsCheckFunc[ssa.BoundsSlice3Acap] = typecheck.LookupRuntimeFunc("goPanicSlice3Acap") |
| BoundsCheckFunc[ssa.BoundsSlice3AcapU] = typecheck.LookupRuntimeFunc("goPanicSlice3AcapU") |
| BoundsCheckFunc[ssa.BoundsSlice3B] = typecheck.LookupRuntimeFunc("goPanicSlice3B") |
| BoundsCheckFunc[ssa.BoundsSlice3BU] = typecheck.LookupRuntimeFunc("goPanicSlice3BU") |
| BoundsCheckFunc[ssa.BoundsSlice3C] = typecheck.LookupRuntimeFunc("goPanicSlice3C") |
| BoundsCheckFunc[ssa.BoundsSlice3CU] = typecheck.LookupRuntimeFunc("goPanicSlice3CU") |
| BoundsCheckFunc[ssa.BoundsConvert] = typecheck.LookupRuntimeFunc("goPanicSliceConvert") |
| } else { |
| BoundsCheckFunc[ssa.BoundsIndex] = typecheck.LookupRuntimeFunc("panicIndex") |
| BoundsCheckFunc[ssa.BoundsIndexU] = typecheck.LookupRuntimeFunc("panicIndexU") |
| BoundsCheckFunc[ssa.BoundsSliceAlen] = typecheck.LookupRuntimeFunc("panicSliceAlen") |
| BoundsCheckFunc[ssa.BoundsSliceAlenU] = typecheck.LookupRuntimeFunc("panicSliceAlenU") |
| BoundsCheckFunc[ssa.BoundsSliceAcap] = typecheck.LookupRuntimeFunc("panicSliceAcap") |
| BoundsCheckFunc[ssa.BoundsSliceAcapU] = typecheck.LookupRuntimeFunc("panicSliceAcapU") |
| BoundsCheckFunc[ssa.BoundsSliceB] = typecheck.LookupRuntimeFunc("panicSliceB") |
| BoundsCheckFunc[ssa.BoundsSliceBU] = typecheck.LookupRuntimeFunc("panicSliceBU") |
| BoundsCheckFunc[ssa.BoundsSlice3Alen] = typecheck.LookupRuntimeFunc("panicSlice3Alen") |
| BoundsCheckFunc[ssa.BoundsSlice3AlenU] = typecheck.LookupRuntimeFunc("panicSlice3AlenU") |
| BoundsCheckFunc[ssa.BoundsSlice3Acap] = typecheck.LookupRuntimeFunc("panicSlice3Acap") |
| BoundsCheckFunc[ssa.BoundsSlice3AcapU] = typecheck.LookupRuntimeFunc("panicSlice3AcapU") |
| BoundsCheckFunc[ssa.BoundsSlice3B] = typecheck.LookupRuntimeFunc("panicSlice3B") |
| BoundsCheckFunc[ssa.BoundsSlice3BU] = typecheck.LookupRuntimeFunc("panicSlice3BU") |
| BoundsCheckFunc[ssa.BoundsSlice3C] = typecheck.LookupRuntimeFunc("panicSlice3C") |
| BoundsCheckFunc[ssa.BoundsSlice3CU] = typecheck.LookupRuntimeFunc("panicSlice3CU") |
| BoundsCheckFunc[ssa.BoundsConvert] = typecheck.LookupRuntimeFunc("panicSliceConvert") |
| } |
| if Arch.LinkArch.PtrSize == 4 { |
| ExtendCheckFunc[ssa.BoundsIndex] = typecheck.LookupRuntimeVar("panicExtendIndex") |
| ExtendCheckFunc[ssa.BoundsIndexU] = typecheck.LookupRuntimeVar("panicExtendIndexU") |
| ExtendCheckFunc[ssa.BoundsSliceAlen] = typecheck.LookupRuntimeVar("panicExtendSliceAlen") |
| ExtendCheckFunc[ssa.BoundsSliceAlenU] = typecheck.LookupRuntimeVar("panicExtendSliceAlenU") |
| ExtendCheckFunc[ssa.BoundsSliceAcap] = typecheck.LookupRuntimeVar("panicExtendSliceAcap") |
| ExtendCheckFunc[ssa.BoundsSliceAcapU] = typecheck.LookupRuntimeVar("panicExtendSliceAcapU") |
| ExtendCheckFunc[ssa.BoundsSliceB] = typecheck.LookupRuntimeVar("panicExtendSliceB") |
| ExtendCheckFunc[ssa.BoundsSliceBU] = typecheck.LookupRuntimeVar("panicExtendSliceBU") |
| ExtendCheckFunc[ssa.BoundsSlice3Alen] = typecheck.LookupRuntimeVar("panicExtendSlice3Alen") |
| ExtendCheckFunc[ssa.BoundsSlice3AlenU] = typecheck.LookupRuntimeVar("panicExtendSlice3AlenU") |
| ExtendCheckFunc[ssa.BoundsSlice3Acap] = typecheck.LookupRuntimeVar("panicExtendSlice3Acap") |
| ExtendCheckFunc[ssa.BoundsSlice3AcapU] = typecheck.LookupRuntimeVar("panicExtendSlice3AcapU") |
| ExtendCheckFunc[ssa.BoundsSlice3B] = typecheck.LookupRuntimeVar("panicExtendSlice3B") |
| ExtendCheckFunc[ssa.BoundsSlice3BU] = typecheck.LookupRuntimeVar("panicExtendSlice3BU") |
| ExtendCheckFunc[ssa.BoundsSlice3C] = typecheck.LookupRuntimeVar("panicExtendSlice3C") |
| ExtendCheckFunc[ssa.BoundsSlice3CU] = typecheck.LookupRuntimeVar("panicExtendSlice3CU") |
| } |
| |
| // Wasm (all asm funcs with special ABIs) |
| ir.Syms.WasmDiv = typecheck.LookupRuntimeVar("wasmDiv") |
| ir.Syms.WasmTruncS = typecheck.LookupRuntimeVar("wasmTruncS") |
| ir.Syms.WasmTruncU = typecheck.LookupRuntimeVar("wasmTruncU") |
| ir.Syms.SigPanic = typecheck.LookupRuntimeFunc("sigpanic") |
| } |
| |
| // AbiForBodylessFuncStackMap returns the ABI for a bodyless function's stack map. |
| // This is not necessarily the ABI used to call it. |
| // Currently (1.17 dev) such a stack map is always ABI0; |
| // any ABI wrapper that is present is nosplit, hence a precise |
| // stack map is not needed there (the parameters survive only long |
| // enough to call the wrapped assembly function). |
| // This always returns a freshly copied ABI. |
| func AbiForBodylessFuncStackMap(fn *ir.Func) *abi.ABIConfig { |
| return ssaConfig.ABI0.Copy() // No idea what races will result, be safe |
| } |
| |
| // abiForFunc implements ABI policy for a function, but does not return a copy of the ABI. |
| // Passing a nil function returns the default ABI based on experiment configuration. |
| func abiForFunc(fn *ir.Func, abi0, abi1 *abi.ABIConfig) *abi.ABIConfig { |
| if buildcfg.Experiment.RegabiArgs { |
| // Select the ABI based on the function's defining ABI. |
| if fn == nil { |
| return abi1 |
| } |
| switch fn.ABI { |
| case obj.ABI0: |
| return abi0 |
| case obj.ABIInternal: |
| // TODO(austin): Clean up the nomenclature here. |
| // It's not clear that "abi1" is ABIInternal. |
| return abi1 |
| } |
| base.Fatalf("function %v has unknown ABI %v", fn, fn.ABI) |
| panic("not reachable") |
| } |
| |
| a := abi0 |
| if fn != nil { |
| if fn.Pragma&ir.RegisterParams != 0 { // TODO(register args) remove after register abi is working |
| a = abi1 |
| } |
| } |
| return a |
| } |
| |
| // dvarint writes a varint v to the funcdata in symbol x and returns the new offset. |
| func dvarint(x *obj.LSym, off int, v int64) int { |
| if v < 0 || v > 1e9 { |
| panic(fmt.Sprintf("dvarint: bad offset for funcdata - %v", v)) |
| } |
| if v < 1<<7 { |
| return objw.Uint8(x, off, uint8(v)) |
| } |
| off = objw.Uint8(x, off, uint8((v&127)|128)) |
| if v < 1<<14 { |
| return objw.Uint8(x, off, uint8(v>>7)) |
| } |
| off = objw.Uint8(x, off, uint8(((v>>7)&127)|128)) |
| if v < 1<<21 { |
| return objw.Uint8(x, off, uint8(v>>14)) |
| } |
| off = objw.Uint8(x, off, uint8(((v>>14)&127)|128)) |
| if v < 1<<28 { |
| return objw.Uint8(x, off, uint8(v>>21)) |
| } |
| off = objw.Uint8(x, off, uint8(((v>>21)&127)|128)) |
| return objw.Uint8(x, off, uint8(v>>28)) |
| } |
| |
| // emitOpenDeferInfo emits FUNCDATA information about the defers in a function |
| // that is using open-coded defers. This funcdata is used to determine the active |
| // defers in a function and execute those defers during panic processing. |
| // |
| // The funcdata is all encoded in varints (since values will almost always be less than |
| // 128, but stack offsets could potentially be up to 2Gbyte). All "locations" (offsets) |
| // for stack variables are specified as the number of bytes below varp (pointer to the |
| // top of the local variables) for their starting address. The format is: |
| // |
| // - Offset of the deferBits variable |
| // - Number of defers in the function |
| // - Information about each defer call, in reverse order of appearance in the function: |
| // - Offset of the closure value to call |
| func (s *state) emitOpenDeferInfo() { |
| x := base.Ctxt.Lookup(s.curfn.LSym.Name + ".opendefer") |
| x.Set(obj.AttrContentAddressable, true) |
| s.curfn.LSym.Func().OpenCodedDeferInfo = x |
| off := 0 |
| off = dvarint(x, off, -s.deferBitsTemp.FrameOffset()) |
| off = dvarint(x, off, int64(len(s.openDefers))) |
| |
| // Write in reverse-order, for ease of running in that order at runtime |
| for i := len(s.openDefers) - 1; i >= 0; i-- { |
| r := s.openDefers[i] |
| off = dvarint(x, off, -r.closureNode.FrameOffset()) |
| } |
| } |
| |
| func okOffset(offset int64) int64 { |
| if offset == types.BOGUS_FUNARG_OFFSET { |
| panic(fmt.Errorf("Bogus offset %d", offset)) |
| } |
| return offset |
| } |
| |
| // buildssa builds an SSA function for fn. |
| // worker indicates which of the backend workers is doing the processing. |
| func buildssa(fn *ir.Func, worker int) *ssa.Func { |
| name := ir.FuncName(fn) |
| printssa := false |
| if ssaDump != "" { // match either a simple name e.g. "(*Reader).Reset", package.name e.g. "compress/gzip.(*Reader).Reset", or subpackage name "gzip.(*Reader).Reset" |
| pkgDotName := base.Ctxt.Pkgpath + "." + name |
| printssa = name == ssaDump || |
| strings.HasSuffix(pkgDotName, ssaDump) && (pkgDotName == ssaDump || strings.HasSuffix(pkgDotName, "/"+ssaDump)) |
| } |
| var astBuf *bytes.Buffer |
| if printssa { |
| astBuf = &bytes.Buffer{} |
| ir.FDumpList(astBuf, "buildssa-enter", fn.Enter) |
| ir.FDumpList(astBuf, "buildssa-body", fn.Body) |
| ir.FDumpList(astBuf, "buildssa-exit", fn.Exit) |
| if ssaDumpStdout { |
| fmt.Println("generating SSA for", name) |
| fmt.Print(astBuf.String()) |
| } |
| } |
| |
| var s state |
| s.pushLine(fn.Pos()) |
| defer s.popLine() |
| |
| s.hasdefer = fn.HasDefer() |
| if fn.Pragma&ir.CgoUnsafeArgs != 0 { |
| s.cgoUnsafeArgs = true |
| } |
| s.checkPtrEnabled = ir.ShouldCheckPtr(fn, 1) |
| |
| fe := ssafn{ |
| curfn: fn, |
| log: printssa && ssaDumpStdout, |
| } |
| s.curfn = fn |
| |
| s.f = ssa.NewFunc(&fe) |
| s.config = ssaConfig |
| s.f.Type = fn.Type() |
| s.f.Config = ssaConfig |
| s.f.Cache = &ssaCaches[worker] |
| s.f.Cache.Reset() |
| s.f.Name = name |
| s.f.PrintOrHtmlSSA = printssa |
| if fn.Pragma&ir.Nosplit != 0 { |
| s.f.NoSplit = true |
| } |
| s.f.ABI0 = ssaConfig.ABI0.Copy() // Make a copy to avoid racy map operations in type-register-width cache. |
| s.f.ABI1 = ssaConfig.ABI1.Copy() |
| s.f.ABIDefault = abiForFunc(nil, s.f.ABI0, s.f.ABI1) |
| s.f.ABISelf = abiForFunc(fn, s.f.ABI0, s.f.ABI1) |
| |
| s.panics = map[funcLine]*ssa.Block{} |
| s.softFloat = s.config.SoftFloat |
| |
| // Allocate starting block |
| s.f.Entry = s.f.NewBlock(ssa.BlockPlain) |
| s.f.Entry.Pos = fn.Pos() |
| |
| if printssa { |
| ssaDF := ssaDumpFile |
| if ssaDir != "" { |
| ssaDF = filepath.Join(ssaDir, base.Ctxt.Pkgpath+"."+name+".html") |
| ssaD := filepath.Dir(ssaDF) |
| os.MkdirAll(ssaD, 0755) |
| } |
| s.f.HTMLWriter = ssa.NewHTMLWriter(ssaDF, s.f, ssaDumpCFG) |
| // TODO: generate and print a mapping from nodes to values and blocks |
| dumpSourcesColumn(s.f.HTMLWriter, fn) |
| s.f.HTMLWriter.WriteAST("AST", astBuf) |
| } |
| |
| // Allocate starting values |
| s.labels = map[string]*ssaLabel{} |
| s.fwdVars = map[ir.Node]*ssa.Value{} |
| s.startmem = s.entryNewValue0(ssa.OpInitMem, types.TypeMem) |
| |
| s.hasOpenDefers = base.Flag.N == 0 && s.hasdefer && !s.curfn.OpenCodedDeferDisallowed() |
| switch { |
| case base.Debug.NoOpenDefer != 0: |
| s.hasOpenDefers = false |
| case s.hasOpenDefers && (base.Ctxt.Flag_shared || base.Ctxt.Flag_dynlink) && base.Ctxt.Arch.Name == "386": |
| // Don't support open-coded defers for 386 ONLY when using shared |
| // libraries, because there is extra code (added by rewriteToUseGot()) |
| // preceding the deferreturn/ret code that we don't track correctly. |
| s.hasOpenDefers = false |
| } |
| if s.hasOpenDefers && len(s.curfn.Exit) > 0 { |
| // Skip doing open defers if there is any extra exit code (likely |
| // race detection), since we will not generate that code in the |
| // case of the extra deferreturn/ret segment. |
| s.hasOpenDefers = false |
| } |
| if s.hasOpenDefers { |
| // Similarly, skip if there are any heap-allocated result |
| // parameters that need to be copied back to their stack slots. |
| for _, f := range s.curfn.Type().Results().FieldSlice() { |
| if !f.Nname.(*ir.Name).OnStack() { |
| s.hasOpenDefers = false |
| break |
| } |
| } |
| } |
| if s.hasOpenDefers && |
| s.curfn.NumReturns*s.curfn.NumDefers > 15 { |
| // Since we are generating defer calls at every exit for |
| // open-coded defers, skip doing open-coded defers if there are |
| // too many returns (especially if there are multiple defers). |
| // Open-coded defers are most important for improving performance |
| // for smaller functions (which don't have many returns). |
| s.hasOpenDefers = false |
| } |
| |
| s.sp = s.entryNewValue0(ssa.OpSP, types.Types[types.TUINTPTR]) // TODO: use generic pointer type (unsafe.Pointer?) instead |
| s.sb = s.entryNewValue0(ssa.OpSB, types.Types[types.TUINTPTR]) |
| |
| s.startBlock(s.f.Entry) |
| s.vars[memVar] = s.startmem |
| if s.hasOpenDefers { |
| // Create the deferBits variable and stack slot. deferBits is a |
| // bitmask showing which of the open-coded defers in this function |
| // have been activated. |
| deferBitsTemp := typecheck.TempAt(src.NoXPos, s.curfn, types.Types[types.TUINT8]) |
| deferBitsTemp.SetAddrtaken(true) |
| s.deferBitsTemp = deferBitsTemp |
| // For this value, AuxInt is initialized to zero by default |
| startDeferBits := s.entryNewValue0(ssa.OpConst8, types.Types[types.TUINT8]) |
| s.vars[deferBitsVar] = startDeferBits |
| s.deferBitsAddr = s.addr(deferBitsTemp) |
| s.store(types.Types[types.TUINT8], s.deferBitsAddr, startDeferBits) |
| // Make sure that the deferBits stack slot is kept alive (for use |
| // by panics) and stores to deferBits are not eliminated, even if |
| // all checking code on deferBits in the function exit can be |
| // eliminated, because the defer statements were all |
| // unconditional. |
| s.vars[memVar] = s.newValue1Apos(ssa.OpVarLive, types.TypeMem, deferBitsTemp, s.mem(), false) |
| } |
| |
| var params *abi.ABIParamResultInfo |
| params = s.f.ABISelf.ABIAnalyze(fn.Type(), true) |
| |
| // The backend's stackframe pass prunes away entries from the fn's |
| // Dcl list, including PARAMOUT nodes that correspond to output |
| // params passed in registers. Walk the Dcl list and capture these |
| // nodes to a side list, so that we'll have them available during |
| // DWARF-gen later on. See issue 48573 for more details. |
| var debugInfo ssa.FuncDebug |
| for _, n := range fn.Dcl { |
| if n.Class == ir.PPARAMOUT && n.IsOutputParamInRegisters() { |
| debugInfo.RegOutputParams = append(debugInfo.RegOutputParams, n) |
| } |
| } |
| fn.DebugInfo = &debugInfo |
| |
| // Generate addresses of local declarations |
| s.decladdrs = map[*ir.Name]*ssa.Value{} |
| for _, n := range fn.Dcl { |
| switch n.Class { |
| case ir.PPARAM: |
| // Be aware that blank and unnamed input parameters will not appear here, but do appear in the type |
| s.decladdrs[n] = s.entryNewValue2A(ssa.OpLocalAddr, types.NewPtr(n.Type()), n, s.sp, s.startmem) |
| case ir.PPARAMOUT: |
| s.decladdrs[n] = s.entryNewValue2A(ssa.OpLocalAddr, types.NewPtr(n.Type()), n, s.sp, s.startmem) |
| case ir.PAUTO: |
| // processed at each use, to prevent Addr coming |
| // before the decl. |
| default: |
| s.Fatalf("local variable with class %v unimplemented", n.Class) |
| } |
| } |
| |
| s.f.OwnAux = ssa.OwnAuxCall(fn.LSym, params) |
| |
| // Populate SSAable arguments. |
| for _, n := range fn.Dcl { |
| if n.Class == ir.PPARAM { |
| if s.canSSA(n) { |
| v := s.newValue0A(ssa.OpArg, n.Type(), n) |
| s.vars[n] = v |
| s.addNamedValue(n, v) // This helps with debugging information, not needed for compilation itself. |
| } else { // address was taken AND/OR too large for SSA |
| paramAssignment := ssa.ParamAssignmentForArgName(s.f, n) |
| if len(paramAssignment.Registers) > 0 { |
| if TypeOK(n.Type()) { // SSA-able type, so address was taken -- receive value in OpArg, DO NOT bind to var, store immediately to memory. |
| v := s.newValue0A(ssa.OpArg, n.Type(), n) |
| s.store(n.Type(), s.decladdrs[n], v) |
| } else { // Too big for SSA. |
| // Brute force, and early, do a bunch of stores from registers |
| // TODO fix the nasty storeArgOrLoad recursion in ssa/expand_calls.go so this Just Works with store of a big Arg. |
| s.storeParameterRegsToStack(s.f.ABISelf, paramAssignment, n, s.decladdrs[n], false) |
| } |
| } |
| } |
| } |
| } |
| |
| // Populate closure variables. |
| if fn.Needctxt() { |
| clo := s.entryNewValue0(ssa.OpGetClosurePtr, s.f.Config.Types.BytePtr) |
| offset := int64(types.PtrSize) // PtrSize to skip past function entry PC field |
| for _, n := range fn.ClosureVars { |
| typ := n.Type() |
| if !n.Byval() { |
| typ = types.NewPtr(typ) |
| } |
| |
| offset = types.RoundUp(offset, typ.Alignment()) |
| ptr := s.newValue1I(ssa.OpOffPtr, types.NewPtr(typ), offset, clo) |
| offset += typ.Size() |
| |
| // If n is a small variable captured by value, promote |
| // it to PAUTO so it can be converted to SSA. |
| // |
| // Note: While we never capture a variable by value if |
| // the user took its address, we may have generated |
| // runtime calls that did (#43701). Since we don't |
| // convert Addrtaken variables to SSA anyway, no point |
| // in promoting them either. |
| if n.Byval() && !n.Addrtaken() && TypeOK(n.Type()) { |
| n.Class = ir.PAUTO |
| fn.Dcl = append(fn.Dcl, n) |
| s.assign(n, s.load(n.Type(), ptr), false, 0) |
| continue |
| } |
| |
| if !n.Byval() { |
| ptr = s.load(typ, ptr) |
| } |
| s.setHeapaddr(fn.Pos(), n, ptr) |
| } |
| } |
| |
| // Convert the AST-based IR to the SSA-based IR |
| s.stmtList(fn.Enter) |
| s.zeroResults() |
| s.paramsToHeap() |
| s.stmtList(fn.Body) |
| |
| // fallthrough to exit |
| if s.curBlock != nil { |
| s.pushLine(fn.Endlineno) |
| s.exit() |
| s.popLine() |
| } |
| |
| for _, b := range s.f.Blocks { |
| if b.Pos != src.NoXPos { |
| s.updateUnsetPredPos(b) |
| } |
| } |
| |
| s.f.HTMLWriter.WritePhase("before insert phis", "before insert phis") |
| |
| s.insertPhis() |
| |
| // Main call to ssa package to compile function |
| ssa.Compile(s.f) |
| |
| if s.hasOpenDefers { |
| s.emitOpenDeferInfo() |
| } |
| |
| // Record incoming parameter spill information for morestack calls emitted in the assembler. |
| // This is done here, using all the parameters (used, partially used, and unused) because |
| // it mimics the behavior of the former ABI (everything stored) and because it's not 100% |
| // clear if naming conventions are respected in autogenerated code. |
| // TODO figure out exactly what's unused, don't spill it. Make liveness fine-grained, also. |
| for _, p := range params.InParams() { |
| typs, offs := p.RegisterTypesAndOffsets() |
| for i, t := range typs { |
| o := offs[i] // offset within parameter |
| fo := p.FrameOffset(params) // offset of parameter in frame |
| reg := ssa.ObjRegForAbiReg(p.Registers[i], s.f.Config) |
| s.f.RegArgs = append(s.f.RegArgs, ssa.Spill{Reg: reg, Offset: fo + o, Type: t}) |
| } |
| } |
| |
| return s.f |
| } |
| |
| func (s *state) storeParameterRegsToStack(abi *abi.ABIConfig, paramAssignment *abi.ABIParamAssignment, n *ir.Name, addr *ssa.Value, pointersOnly bool) { |
| typs, offs := paramAssignment.RegisterTypesAndOffsets() |
| for i, t := range typs { |
| if pointersOnly && !t.IsPtrShaped() { |
| continue |
| } |
| r := paramAssignment.Registers[i] |
| o := offs[i] |
| op, reg := ssa.ArgOpAndRegisterFor(r, abi) |
| aux := &ssa.AuxNameOffset{Name: n, Offset: o} |
| v := s.newValue0I(op, t, reg) |
| v.Aux = aux |
| p := s.newValue1I(ssa.OpOffPtr, types.NewPtr(t), o, addr) |
| s.store(t, p, v) |
| } |
| } |
| |
| // zeroResults zeros the return values at the start of the function. |
| // We need to do this very early in the function. Defer might stop a |
| // panic and show the return values as they exist at the time of |
| // panic. For precise stacks, the garbage collector assumes results |
| // are always live, so we need to zero them before any allocations, |
| // even allocations to move params/results to the heap. |
| func (s *state) zeroResults() { |
| for _, f := range s.curfn.Type().Results().FieldSlice() { |
| n := f.Nname.(*ir.Name) |
| if !n.OnStack() { |
| // The local which points to the return value is the |
| // thing that needs zeroing. This is already handled |
| // by a Needzero annotation in plive.go:(*liveness).epilogue. |
| continue |
| } |
| // Zero the stack location containing f. |
| if typ := n.Type(); TypeOK(typ) { |
| s.assign(n, s.zeroVal(typ), false, 0) |
| } else { |
| if typ.HasPointers() { |
| s.vars[memVar] = s.newValue1A(ssa.OpVarDef, types.TypeMem, n, s.mem()) |
| } |
| s.zero(n.Type(), s.decladdrs[n]) |
| } |
| } |
| } |
| |
| // paramsToHeap produces code to allocate memory for heap-escaped parameters |
| // and to copy non-result parameters' values from the stack. |
| func (s *state) paramsToHeap() { |
| do := func(params *types.Type) { |
| for _, f := range params.FieldSlice() { |
| if f.Nname == nil { |
| continue // anonymous or blank parameter |
| } |
| n := f.Nname.(*ir.Name) |
| if ir.IsBlank(n) || n.OnStack() { |
| continue |
| } |
| s.newHeapaddr(n) |
| if n.Class == ir.PPARAM { |
| s.move(n.Type(), s.expr(n.Heapaddr), s.decladdrs[n]) |
| } |
| } |
| } |
| |
| typ := s.curfn.Type() |
| do(typ.Recvs()) |
| do(typ.Params()) |
| do(typ.Results()) |
| } |
| |
| // newHeapaddr allocates heap memory for n and sets its heap address. |
| func (s *state) newHeapaddr(n *ir.Name) { |
| s.setHeapaddr(n.Pos(), n, s.newObject(n.Type(), nil)) |
| } |
| |
| // setHeapaddr allocates a new PAUTO variable to store ptr (which must be non-nil) |
| // and then sets it as n's heap address. |
| func (s *state) setHeapaddr(pos src.XPos, n *ir.Name, ptr *ssa.Value) { |
| if !ptr.Type.IsPtr() || !types.Identical(n.Type(), ptr.Type.Elem()) { |
| base.FatalfAt(n.Pos(), "setHeapaddr %L with type %v", n, ptr.Type) |
| } |
| |
| // Declare variable to hold address. |
| addr := ir.NewNameAt(pos, &types.Sym{Name: "&" + n.Sym().Name, Pkg: types.LocalPkg}) |
| addr.SetType(types.NewPtr(n.Type())) |
| addr.Class = ir.PAUTO |
| addr.SetUsed(true) |
| addr.Curfn = s.curfn |
| s.curfn.Dcl = append(s.curfn.Dcl, addr) |
| types.CalcSize(addr.Type()) |
| |
| if n.Class == ir.PPARAMOUT { |
| addr.SetIsOutputParamHeapAddr(true) |
| } |
| |
| n.Heapaddr = addr |
| s.assign(addr, ptr, false, 0) |
| } |
| |
| // newObject returns an SSA value denoting new(typ). |
| func (s *state) newObject(typ *types.Type, rtype *ssa.Value) *ssa.Value { |
| if typ.Size() == 0 { |
| return s.newValue1A(ssa.OpAddr, types.NewPtr(typ), ir.Syms.Zerobase, s.sb) |
| } |
| if rtype == nil { |
| rtype = s.reflectType(typ) |
| } |
| return s.rtcall(ir.Syms.Newobject, true, []*types.Type{types.NewPtr(typ)}, rtype)[0] |
| } |
| |
| func (s *state) checkPtrAlignment(n *ir.ConvExpr, v *ssa.Value, count *ssa.Value) { |
| if !n.Type().IsPtr() { |
| s.Fatalf("expected pointer type: %v", n.Type()) |
| } |
| elem, rtypeExpr := n.Type().Elem(), n.ElemRType |
| if count != nil { |
| if !elem.IsArray() { |
| s.Fatalf("expected array type: %v", elem) |
| } |
| elem, rtypeExpr = elem.Elem(), n.ElemElemRType |
| } |
| size := elem.Size() |
| // Casting from larger type to smaller one is ok, so for smallest type, do nothing. |
| if elem.Alignment() == 1 && (size == 0 || size == 1 || count == nil) { |
| return |
| } |
| if count == nil { |
| count = s.constInt(types.Types[types.TUINTPTR], 1) |
| } |
| if count.Type.Size() != s.config.PtrSize { |
| s.Fatalf("expected count fit to a uintptr size, have: %d, want: %d", count.Type.Size(), s.config.PtrSize) |
| } |
| var rtype *ssa.Value |
| if rtypeExpr != nil { |
| rtype = s.expr(rtypeExpr) |
| } else { |
| rtype = s.reflectType(elem) |
| } |
| s.rtcall(ir.Syms.CheckPtrAlignment, true, nil, v, rtype, count) |
| } |
| |
| // reflectType returns an SSA value representing a pointer to typ's |
| // reflection type descriptor. |
| func (s *state) reflectType(typ *types.Type) *ssa.Value { |
| // TODO(mdempsky): Make this Fatalf under Unified IR; frontend needs |
| // to supply RType expressions. |
| lsym := reflectdata.TypeLinksym(typ) |
| return s.entryNewValue1A(ssa.OpAddr, types.NewPtr(types.Types[types.TUINT8]), lsym, s.sb) |
| } |
| |
| func dumpSourcesColumn(writer *ssa.HTMLWriter, fn *ir.Func) { |
| // Read sources of target function fn. |
| fname := base.Ctxt.PosTable.Pos(fn.Pos()).Filename() |
| targetFn, err := readFuncLines(fname, fn.Pos().Line(), fn.Endlineno.Line()) |
| if err != nil { |
| writer.Logf("cannot read sources for function %v: %v", fn, err) |
| } |
| |
| // Read sources of inlined functions. |
| var inlFns []*ssa.FuncLines |
| for _, fi := range ssaDumpInlined { |
| elno := fi.Endlineno |
| fname := base.Ctxt.PosTable.Pos(fi.Pos()).Filename() |
| fnLines, err := readFuncLines(fname, fi.Pos().Line(), elno.Line()) |
| if err != nil { |
| writer.Logf("cannot read sources for inlined function %v: %v", fi, err) |
| continue |
| } |
| inlFns = append(inlFns, fnLines) |
| } |
| |
| sort.Sort(ssa.ByTopo(inlFns)) |
| if targetFn != nil { |
| inlFns = append([]*ssa.FuncLines{targetFn}, inlFns...) |
| } |
| |
| writer.WriteSources("sources", inlFns) |
| } |
| |
| func readFuncLines(file string, start, end uint) (*ssa.FuncLines, error) { |
| f, err := os.Open(os.ExpandEnv(file)) |
| if err != nil { |
| return nil, err |
| } |
| defer f.Close() |
| var lines []string |
| ln := uint(1) |
| scanner := bufio.NewScanner(f) |
| for scanner.Scan() && ln <= end { |
| if ln >= start { |
| lines = append(lines, scanner.Text()) |
| } |
| ln++ |
| } |
| return &ssa.FuncLines{Filename: file, StartLineno: start, Lines: lines}, nil |
| } |
| |
| // updateUnsetPredPos propagates the earliest-value position information for b |
| // towards all of b's predecessors that need a position, and recurs on that |
| // predecessor if its position is updated. B should have a non-empty position. |
| func (s *state) updateUnsetPredPos(b *ssa.Block) { |
| if b.Pos == src.NoXPos { |
| s.Fatalf("Block %s should have a position", b) |
| } |
| bestPos := src.NoXPos |
| for _, e := range b.Preds { |
| p := e.Block() |
| if !p.LackingPos() { |
| continue |
| } |
| if bestPos == src.NoXPos { |
| bestPos = b.Pos |
| for _, v := range b.Values { |
| if v.LackingPos() { |
| continue |
| } |
| if v.Pos != src.NoXPos { |
| // Assume values are still in roughly textual order; |
| // TODO: could also seek minimum position? |
| bestPos = v.Pos |
| break |
| } |
| } |
| } |
| p.Pos = bestPos |
| s.updateUnsetPredPos(p) // We do not expect long chains of these, thus recursion is okay. |
| } |
| } |
| |
| // Information about each open-coded defer. |
| type openDeferInfo struct { |
| // The node representing the call of the defer |
| n *ir.CallExpr |
| // If defer call is closure call, the address of the argtmp where the |
| // closure is stored. |
| closure *ssa.Value |
| // The node representing the argtmp where the closure is stored - used for |
| // function, method, or interface call, to store a closure that panic |
| // processing can use for this defer. |
| closureNode *ir.Name |
| } |
| |
| type state struct { |
| // configuration (arch) information |
| config *ssa.Config |
| |
| // function we're building |
| f *ssa.Func |
| |
| // Node for function |
| curfn *ir.Func |
| |
| // labels in f |
| labels map[string]*ssaLabel |
| |
| // unlabeled break and continue statement tracking |
| breakTo *ssa.Block // current target for plain break statement |
| continueTo *ssa.Block // current target for plain continue statement |
| |
| // current location where we're interpreting the AST |
| curBlock *ssa.Block |
| |
| // variable assignments in the current block (map from variable symbol to ssa value) |
| // *Node is the unique identifier (an ONAME Node) for the variable. |
| // TODO: keep a single varnum map, then make all of these maps slices instead? |
| vars map[ir.Node]*ssa.Value |
| |
| // fwdVars are variables that are used before they are defined in the current block. |
| // This map exists just to coalesce multiple references into a single FwdRef op. |
| // *Node is the unique identifier (an ONAME Node) for the variable. |
| fwdVars map[ir.Node]*ssa.Value |
| |
| // all defined variables at the end of each block. Indexed by block ID. |
| defvars []map[ir.Node]*ssa.Value |
| |
| // addresses of PPARAM and PPARAMOUT variables on the stack. |
| decladdrs map[*ir.Name]*ssa.Value |
| |
| // starting values. Memory, stack pointer, and globals pointer |
| startmem *ssa.Value |
| sp *ssa.Value |
| sb *ssa.Value |
| // value representing address of where deferBits autotmp is stored |
| deferBitsAddr *ssa.Value |
| deferBitsTemp *ir.Name |
| |
| // line number stack. The current line number is top of stack |
| line []src.XPos |
| // the last line number processed; it may have been popped |
| lastPos src.XPos |
| |
| // list of panic calls by function name and line number. |
| // Used to deduplicate panic calls. |
| panics map[funcLine]*ssa.Block |
| |
| cgoUnsafeArgs bool |
| hasdefer bool // whether the function contains a defer statement |
| softFloat bool |
| hasOpenDefers bool // whether we are doing open-coded defers |
| checkPtrEnabled bool // whether to insert checkptr instrumentation |
| |
| // If doing open-coded defers, list of info about the defer calls in |
| // scanning order. Hence, at exit we should run these defers in reverse |
| // order of this list |
| openDefers []*openDeferInfo |
| // For open-coded defers, this is the beginning and end blocks of the last |
| // defer exit code that we have generated so far. We use these to share |
| // code between exits if the shareDeferExits option (disabled by default) |
| // is on. |
| lastDeferExit *ssa.Block // Entry block of last defer exit code we generated |
| lastDeferFinalBlock *ssa.Block // Final block of last defer exit code we generated |
| lastDeferCount int // Number of defers encountered at that point |
| |
| prevCall *ssa.Value // the previous call; use this to tie results to the call op. |
| } |
| |
| type funcLine struct { |
| f *obj.LSym |
| base *src.PosBase |
| line uint |
| } |
| |
| type ssaLabel struct { |
| target *ssa.Block // block identified by this label |
| breakTarget *ssa.Block // block to break to in control flow node identified by this label |
| continueTarget *ssa.Block // block to continue to in control flow node identified by this label |
| } |
| |
| // label returns the label associated with sym, creating it if necessary. |
| func (s *state) label(sym *types.Sym) *ssaLabel { |
| lab := s.labels[sym.Name] |
| if lab == nil { |
| lab = new(ssaLabel) |
| s.labels[sym.Name] = lab |
| } |
| return lab |
| } |
| |
| func (s *state) Logf(msg string, args ...interface{}) { s.f.Logf(msg, args...) } |
| func (s *state) Log() bool { return s.f.Log() } |
| func (s *state) Fatalf(msg string, args ...interface{}) { |
| s.f.Frontend().Fatalf(s.peekPos(), msg, args...) |
| } |
| func (s *state) Warnl(pos src.XPos, msg string, args ...interface{}) { s.f.Warnl(pos, msg, args...) } |
| func (s *state) Debug_checknil() bool { return s.f.Frontend().Debug_checknil() } |
| |
| func ssaMarker(name string) *ir.Name { |
| return typecheck.NewName(&types.Sym{Name: name}) |
| } |
| |
| var ( |
| // marker node for the memory variable |
| memVar = ssaMarker("mem") |
| |
| // marker nodes for temporary variables |
| ptrVar = ssaMarker("ptr") |
| lenVar = ssaMarker("len") |
| capVar = ssaMarker("cap") |
| typVar = ssaMarker("typ") |
| okVar = ssaMarker("ok") |
| deferBitsVar = ssaMarker("deferBits") |
| ) |
| |
| // startBlock sets the current block we're generating code in to b. |
| func (s *state) startBlock(b *ssa.Block) { |
| if s.curBlock != nil { |
| s.Fatalf("starting block %v when block %v has not ended", b, s.curBlock) |
| } |
| s.curBlock = b |
| s.vars = map[ir.Node]*ssa.Value{} |
| for n := range s.fwdVars { |
| delete(s.fwdVars, n) |
| } |
| } |
| |
| // endBlock marks the end of generating code for the current block. |
| // Returns the (former) current block. Returns nil if there is no current |
| // block, i.e. if no code flows to the current execution point. |
| func (s *state) endBlock() *ssa.Block { |
| b := s.curBlock |
| if b == nil { |
| return nil |
| } |
| for len(s.defvars) <= int(b.ID) { |
| s.defvars = append(s.defvars, nil) |
| } |
| s.defvars[b.ID] = s.vars |
| s.curBlock = nil |
| s.vars = nil |
| if b.LackingPos() { |
| // Empty plain blocks get the line of their successor (handled after all blocks created), |
| // except for increment blocks in For statements (handled in ssa conversion of OFOR), |
| // and for blocks ending in GOTO/BREAK/CONTINUE. |
| b.Pos = src.NoXPos |
| } else { |
| b.Pos = s.lastPos |
| } |
| return b |
| } |
| |
| // pushLine pushes a line number on the line number stack. |
| func (s *state) pushLine(line src.XPos) { |
| if !line.IsKnown() { |
| // the frontend may emit node with line number missing, |
| // use the parent line number in this case. |
| line = s.peekPos() |
| if base.Flag.K != 0 { |
| base.Warn("buildssa: unknown position (line 0)") |
| } |
| } else { |
| s.lastPos = line |
| } |
| |
| s.line = append(s.line, line) |
| } |
| |
| // popLine pops the top of the line number stack. |
| func (s *state) popLine() { |
| s.line = s.line[:len(s.line)-1] |
| } |
| |
| // peekPos peeks the top of the line number stack. |
| func (s *state) peekPos() src.XPos { |
| return s.line[len(s.line)-1] |
| } |
| |
| // newValue0 adds a new value with no arguments to the current block. |
| func (s *state) newValue0(op ssa.Op, t *types.Type) *ssa.Value { |
| return s.curBlock.NewValue0(s.peekPos(), op, t) |
| } |
| |
| // newValue0A adds a new value with no arguments and an aux value to the current block. |
| func (s *state) newValue0A(op ssa.Op, t *types.Type, aux ssa.Aux) *ssa.Value { |
| return s.curBlock.NewValue0A(s.peekPos(), op, t, aux) |
| } |
| |
| // newValue0I adds a new value with no arguments and an auxint value to the current block. |
| func (s *state) newValue0I(op ssa.Op, t *types.Type, auxint int64) *ssa.Value { |
| return s.curBlock.NewValue0I(s.peekPos(), op, t, auxint) |
| } |
| |
| // newValue1 adds a new value with one argument to the current block. |
| func (s *state) newValue1(op ssa.Op, t *types.Type, arg *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue1(s.peekPos(), op, t, arg) |
| } |
| |
| // newValue1A adds a new value with one argument and an aux value to the current block. |
| func (s *state) newValue1A(op ssa.Op, t *types.Type, aux ssa.Aux, arg *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue1A(s.peekPos(), op, t, aux, arg) |
| } |
| |
| // newValue1Apos adds a new value with one argument and an aux value to the current block. |
| // isStmt determines whether the created values may be a statement or not |
| // (i.e., false means never, yes means maybe). |
| func (s *state) newValue1Apos(op ssa.Op, t *types.Type, aux ssa.Aux, arg *ssa.Value, isStmt bool) *ssa.Value { |
| if isStmt { |
| return s.curBlock.NewValue1A(s.peekPos(), op, t, aux, arg) |
| } |
| return s.curBlock.NewValue1A(s.peekPos().WithNotStmt(), op, t, aux, arg) |
| } |
| |
| // newValue1I adds a new value with one argument and an auxint value to the current block. |
| func (s *state) newValue1I(op ssa.Op, t *types.Type, aux int64, arg *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue1I(s.peekPos(), op, t, aux, arg) |
| } |
| |
| // newValue2 adds a new value with two arguments to the current block. |
| func (s *state) newValue2(op ssa.Op, t *types.Type, arg0, arg1 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue2(s.peekPos(), op, t, arg0, arg1) |
| } |
| |
| // newValue2A adds a new value with two arguments and an aux value to the current block. |
| func (s *state) newValue2A(op ssa.Op, t *types.Type, aux ssa.Aux, arg0, arg1 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue2A(s.peekPos(), op, t, aux, arg0, arg1) |
| } |
| |
| // newValue2Apos adds a new value with two arguments and an aux value to the current block. |
| // isStmt determines whether the created values may be a statement or not |
| // (i.e., false means never, yes means maybe). |
| func (s *state) newValue2Apos(op ssa.Op, t *types.Type, aux ssa.Aux, arg0, arg1 *ssa.Value, isStmt bool) *ssa.Value { |
| if isStmt { |
| return s.curBlock.NewValue2A(s.peekPos(), op, t, aux, arg0, arg1) |
| } |
| return s.curBlock.NewValue2A(s.peekPos().WithNotStmt(), op, t, aux, arg0, arg1) |
| } |
| |
| // newValue2I adds a new value with two arguments and an auxint value to the current block. |
| func (s *state) newValue2I(op ssa.Op, t *types.Type, aux int64, arg0, arg1 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue2I(s.peekPos(), op, t, aux, arg0, arg1) |
| } |
| |
| // newValue3 adds a new value with three arguments to the current block. |
| func (s *state) newValue3(op ssa.Op, t *types.Type, arg0, arg1, arg2 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue3(s.peekPos(), op, t, arg0, arg1, arg2) |
| } |
| |
| // newValue3I adds a new value with three arguments and an auxint value to the current block. |
| func (s *state) newValue3I(op ssa.Op, t *types.Type, aux int64, arg0, arg1, arg2 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue3I(s.peekPos(), op, t, aux, arg0, arg1, arg2) |
| } |
| |
| // newValue3A adds a new value with three arguments and an aux value to the current block. |
| func (s *state) newValue3A(op ssa.Op, t *types.Type, aux ssa.Aux, arg0, arg1, arg2 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue3A(s.peekPos(), op, t, aux, arg0, arg1, arg2) |
| } |
| |
| // newValue3Apos adds a new value with three arguments and an aux value to the current block. |
| // isStmt determines whether the created values may be a statement or not |
| // (i.e., false means never, yes means maybe). |
| func (s *state) newValue3Apos(op ssa.Op, t *types.Type, aux ssa.Aux, arg0, arg1, arg2 *ssa.Value, isStmt bool) *ssa.Value { |
| if isStmt { |
| return s.curBlock.NewValue3A(s.peekPos(), op, t, aux, arg0, arg1, arg2) |
| } |
| return s.curBlock.NewValue3A(s.peekPos().WithNotStmt(), op, t, aux, arg0, arg1, arg2) |
| } |
| |
| // newValue4 adds a new value with four arguments to the current block. |
| func (s *state) newValue4(op ssa.Op, t *types.Type, arg0, arg1, arg2, arg3 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue4(s.peekPos(), op, t, arg0, arg1, arg2, arg3) |
| } |
| |
| // newValue4I adds a new value with four arguments and an auxint value to the current block. |
| func (s *state) newValue4I(op ssa.Op, t *types.Type, aux int64, arg0, arg1, arg2, arg3 *ssa.Value) *ssa.Value { |
| return s.curBlock.NewValue4I(s.peekPos(), op, t, aux, arg0, arg1, arg2, arg3) |
| } |
| |
| func (s *state) entryBlock() *ssa.Block { |
| b := s.f.Entry |
| if base.Flag.N > 0 && s.curBlock != nil { |
| // If optimizations are off, allocate in current block instead. Since with -N |
| // we're not doing the CSE or tighten passes, putting lots of stuff in the |
| // entry block leads to O(n^2) entries in the live value map during regalloc. |
| // See issue 45897. |
| b = s.curBlock |
| } |
| return b |
| } |
| |
| // entryNewValue0 adds a new value with no arguments to the entry block. |
| func (s *state) entryNewValue0(op ssa.Op, t *types.Type) *ssa.Value { |
| return s.entryBlock().NewValue0(src.NoXPos, op, t) |
| } |
| |
| // entryNewValue0A adds a new value with no arguments and an aux value to the entry block. |
| func (s *state) entryNewValue0A(op ssa.Op, t *types.Type, aux ssa.Aux) *ssa.Value { |
| return s.entryBlock().NewValue0A(src.NoXPos, op, t, aux) |
| } |
| |
| // entryNewValue1 adds a new value with one argument to the entry block. |
| func (s *state) entryNewValue1(op ssa.Op, t *types.Type, arg *ssa.Value) *ssa.Value { |
| return s.entryBlock().NewValue1(src.NoXPos, op, t, arg) |
| } |
| |
| // entryNewValue1I adds a new value with one argument and an auxint value to the entry block. |
| func (s *state) entryNewValue1I(op ssa.Op, t *types.Type, auxint int64, arg *ssa.Value) *ssa.Value { |
| return s.entryBlock().NewValue1I(src.NoXPos, op, t, auxint, arg) |
| } |
| |
| // entryNewValue1A adds a new value with one argument and an aux value to the entry block. |
| func (s *state) entryNewValue1A(op ssa.Op, t *types.Type, aux ssa.Aux, arg *ssa.Value) *ssa.Value { |
| return s.entryBlock().NewValue1A(src.NoXPos, op, t, aux, arg) |
| } |
| |
| // entryNewValue2 adds a new value with two arguments to the entry block. |
| func (s *state) entryNewValue2(op ssa.Op, t *types.Type, arg0, arg1 *ssa.Value) *ssa.Value { |
| return s.entryBlock().NewValue2(src.NoXPos, op, t, arg0, arg1) |
| } |
| |
| // entryNewValue2A adds a new value with two arguments and an aux value to the entry block. |
| func (s *state) entryNewValue2A(op ssa.Op, t *types.Type, aux ssa.Aux, arg0, arg1 *ssa.Value) *ssa.Value { |
| return s.entryBlock().NewValue2A(src.NoXPos, op, t, aux, arg0, arg1) |
| } |
| |
| // const* routines add a new const value to the entry block. |
| func (s *state) constSlice(t *types.Type) *ssa.Value { |
| return s.f.ConstSlice(t) |
| } |
| func (s *state) constInterface(t *types.Type) *ssa.Value { |
| return s.f.ConstInterface(t) |
| } |
| func (s *state) constNil(t *types.Type) *ssa.Value { return s.f.ConstNil(t) } |
| func (s *state) constEmptyString(t *types.Type) *ssa.Value { |
| return s.f.ConstEmptyString(t) |
| } |
| func (s *state) constBool(c bool) *ssa.Value { |
| return s.f.ConstBool(types.Types[types.TBOOL], c) |
| } |
| func (s *state) constInt8(t *types.Type, c int8) *ssa.Value { |
| return s.f.ConstInt8(t, c) |
| } |
| func (s *state) constInt16(t *types.Type, c int16) *ssa.Value { |
| return s.f.ConstInt16(t, c) |
| } |
| func (s *state) constInt32(t *types.Type, c int32) *ssa.Value { |
| return s.f.ConstInt32(t, c) |
| } |
| func (s *state) constInt64(t *types.Type, c int64) *ssa.Value { |
| return s.f.ConstInt64(t, c) |
| } |
| func (s *state) constFloat32(t *types.Type, c float64) *ssa.Value { |
| return s.f.ConstFloat32(t, c) |
| } |
| func (s *state) constFloat64(t *types.Type, c float64) *ssa.Value { |
| return s.f.ConstFloat64(t, c) |
| } |
| func (s *state) constInt(t *types.Type, c int64) *ssa.Value { |
| if s.config.PtrSize == 8 { |
| return s.constInt64(t, c) |
| } |
| if int64(int32(c)) != c { |
| s.Fatalf("integer constant too big %d", c) |
| } |
| return s.constInt32(t, int32(c)) |
| } |
| func (s *state) constOffPtrSP(t *types.Type, c int64) *ssa.Value { |
| return s.f.ConstOffPtrSP(t, c, s.sp) |
| } |
| |
| // newValueOrSfCall* are wrappers around newValue*, which may create a call to a |
| // soft-float runtime function instead (when emitting soft-float code). |
| func (s *state) newValueOrSfCall1(op ssa.Op, t *types.Type, arg *ssa.Value) *ssa.Value { |
| if s.softFloat { |
| if c, ok := s.sfcall(op, arg); ok { |
| return c |
| } |
| } |
| return s.newValue1(op, t, arg) |
| } |
| func (s *state) newValueOrSfCall2(op ssa.Op, t *types.Type, arg0, arg1 *ssa.Value) *ssa.Value { |
| if s.softFloat { |
| if c, ok := s.sfcall(op, arg0, arg1); ok { |
| return c |
| } |
| } |
| return s.newValue2(op, t, arg0, arg1) |
| } |
| |
| type instrumentKind uint8 |
| |
| const ( |
| instrumentRead = iota |
| instrumentWrite |
| instrumentMove |
| ) |
| |
| func (s *state) instrument(t *types.Type, addr *ssa.Value, kind instrumentKind) { |
| s.instrument2(t, addr, nil, kind) |
| } |
| |
| // instrumentFields instruments a read/write operation on addr. |
| // If it is instrumenting for MSAN or ASAN and t is a struct type, it instruments |
| // operation for each field, instead of for the whole struct. |
| func (s *state) instrumentFields(t *types.Type, addr *ssa.Value, kind instrumentKind) { |
| if !(base.Flag.MSan || base.Flag.ASan) || !t.IsStruct() { |
| s.instrument(t, addr, kind) |
| return |
| } |
| for _, f := range t.Fields().Slice() { |
| if f.Sym.IsBlank() { |
| continue |
| } |
| offptr := s.newValue1I(ssa.OpOffPtr, types.NewPtr(f.Type), f.Offset, addr) |
| s.instrumentFields(f.Type, offptr, kind) |
| } |
| } |
| |
| func (s *state) instrumentMove(t *types.Type, dst, src *ssa.Value) { |
| if base.Flag.MSan { |
| s.instrument2(t, dst, src, instrumentMove) |
| } else { |
| s.instrument(t, src, instrumentRead) |
| s.instrument(t, dst, instrumentWrite) |
| } |
| } |
| |
| func (s *state) instrument2(t *types.Type, addr, addr2 *ssa.Value, kind instrumentKind) { |
| if !s.curfn.InstrumentBody() { |
| return |
| } |
| |
| w := t.Size() |
| if w == 0 { |
| return // can't race on zero-sized things |
| } |
| |
| if ssa.IsSanitizerSafeAddr(addr) { |
| return |
| } |
| |
| var fn *obj.LSym |
| needWidth := false |
| |
| if addr2 != nil && kind != instrumentMove { |
| panic("instrument2: non-nil addr2 for non-move instrumentation") |
| } |
| |
| if base.Flag.MSan { |
| switch kind { |
| case instrumentRead: |
| fn = ir.Syms.Msanread |
| case instrumentWrite: |
| fn = ir.Syms.Msanwrite |
| case instrumentMove: |
| fn = ir.Syms.Msanmove |
| default: |
| panic("unreachable") |
| } |
| needWidth = true |
| } else if base.Flag.Race && t.NumComponents(types.CountBlankFields) > 1 { |
| // for composite objects we have to write every address |
| // because a write might happen to any subobject. |
| // composites with only one element don't have subobjects, though. |
| switch kind { |
| case instrumentRead: |
| fn = ir.Syms.Racereadrange |
| case instrumentWrite: |
| fn = ir.Syms.Racewriterange |
| default: |
| panic("unreachable") |
| } |
| needWidth = true |
| } else if base.Flag.Race { |
| // for non-composite objects we can write just the start |
| // address, as any write must write the first byte. |
| switch kind { |
| case instrumentRead: |
| fn = ir.Syms.Raceread |
| case instrumentWrite: |
| fn = ir.Syms.Racewrite |
| default: |
| panic("unreachable") |
| } |
| } else if base.Flag.ASan { |
| switch kind { |
| case instrumentRead: |
| fn = ir.Syms.Asanread |
| case instrumentWrite: |
| fn = ir.Syms.Asanwrite |
| default: |
| panic("unreachable") |
| } |
| needWidth = true |
| } else { |
| panic("unreachable") |
| } |
| |
| args := []*ssa.Value{addr} |
| if addr2 != nil { |
| args = append(args, addr2) |
| } |
| if needWidth { |
| args = append(args, s.constInt(types.Types[types.TUINTPTR], w)) |
| } |
| s.rtcall(fn, true, nil, args...) |
| } |
| |
| func (s *state) load(t *types.Type, src *ssa.Value) *ssa.Value { |
| s.instrumentFields(t, src, instrumentRead) |
| return s.rawLoad(t, src) |
| } |
| |
| func (s *state) rawLoad(t *types.Type, src *ssa.Value) *ssa.Value { |
| return s.newValue2(ssa.OpLoad, t, src, s.mem()) |
| } |
| |
| func (s *state) store(t *types.Type, dst, val *ssa.Value) { |
| s.vars[memVar] = s.newValue3A(ssa.OpStore, types.TypeMem, t, dst, val, s.mem()) |
| } |
| |
| func (s *state) zero(t *types.Type, dst *ssa.Value) { |
| s.instrument(t, dst, instrumentWrite) |
| store := s.newValue2I(ssa.OpZero, types.TypeMem, t.Size(), dst, s.mem()) |
| store.Aux = t |
| s.vars[memVar] = store |
| } |
| |
| func (s *state) move(t *types.Type, dst, src *ssa.Value) { |
| s.moveWhichMayOverlap(t, dst, src, false) |
| } |
| func (s *state) moveWhichMayOverlap(t *types.Type, dst, src *ssa.Value, mayOverlap bool) { |
| s.instrumentMove(t, dst, src) |
| if mayOverlap && t.IsArray() && t.NumElem() > 1 && !ssa.IsInlinableMemmove(dst, src, t.Size(), s.f.Config) { |
| // Normally, when moving Go values of type T from one location to another, |
| // we don't need to worry about partial overlaps. The two Ts must either be |
| // in disjoint (nonoverlapping) memory or in exactly the same location. |
| // There are 2 cases where this isn't true: |
| // 1) Using unsafe you can arrange partial overlaps. |
| // 2) Since Go 1.17, you can use a cast from a slice to a ptr-to-array. |
| // https://go.dev/ref/spec#Conversions_from_slice_to_array_pointer |
| // This feature can be used to construct partial overlaps of array types. |
| // var a [3]int |
| // p := (*[2]int)(a[:]) |
| // q := (*[2]int)(a[1:]) |
| // *p = *q |
| // We don't care about solving 1. Or at least, we haven't historically |
| // and no one has complained. |
| // For 2, we need to ensure that if there might be partial overlap, |
| // then we can't use OpMove; we must use memmove instead. |
| // (memmove handles partial overlap by copying in the correct |
| // direction. OpMove does not.) |
| // |
| // Note that we have to be careful here not to introduce a call when |
| // we're marshaling arguments to a call or unmarshaling results from a call. |
| // Cases where this is happening must pass mayOverlap to false. |
| // (Currently this only happens when unmarshaling results of a call.) |
| if t.HasPointers() { |
| s.rtcall(ir.Syms.Typedmemmove, true, nil, s.reflectType(t), dst, src) |
| // We would have otherwise implemented this move with straightline code, |
| // including a write barrier. Pretend we issue a write barrier here, |
| // so that the write barrier tests work. (Otherwise they'd need to know |
| // the details of IsInlineableMemmove.) |
| s.curfn.SetWBPos(s.peekPos()) |
| } else { |
| s.rtcall(ir.Syms.Memmove, true, nil, dst, src, s.constInt(types.Types[types.TUINTPTR], t.Size())) |
| } |
| ssa.LogLargeCopy(s.f.Name, s.peekPos(), t.Size()) |
| return |
| } |
| store := s.newValue3I(ssa.OpMove, types.TypeMem, t.Size(), dst, src, s.mem()) |
| store.Aux = t |
| s.vars[memVar] = store |
| } |
| |
| // stmtList converts the statement list n to SSA and adds it to s. |
| func (s *state) stmtList(l ir.Nodes) { |
| for _, n := range l { |
| s.stmt(n) |
| } |
| } |
| |
| // stmt converts the statement n to SSA and adds it to s. |
| func (s *state) stmt(n ir.Node) { |
| s.pushLine(n.Pos()) |
| defer s.popLine() |
| |
| // If s.curBlock is nil, and n isn't a label (which might have an associated goto somewhere), |
| // then this code is dead. Stop here. |
| if s.curBlock == nil && n.Op() != ir.OLABEL { |
| return |
| } |
| |
| s.stmtList(n.Init()) |
| switch n.Op() { |
| |
| case ir.OBLOCK: |
| n := n.(*ir.BlockStmt) |
| s.stmtList(n.List) |
| |
| // No-ops |
| case ir.ODCLCONST, ir.ODCLTYPE, ir.OFALL: |
| |
| // Expression statements |
| case ir.OCALLFUNC: |
| n := n.(*ir.CallExpr) |
| if ir.IsIntrinsicCall(n) { |
| s.intrinsicCall(n) |
| return |
| } |
| fallthrough |
| |
| case ir.OCALLINTER: |
| n := n.(*ir.CallExpr) |
| s.callResult(n, callNormal) |
| if n.Op() == ir.OCALLFUNC && n.X.Op() == ir.ONAME && n.X.(*ir.Name).Class == ir.PFUNC { |
| if fn := n.X.Sym().Name; base.Flag.CompilingRuntime && fn == "throw" || |
| n.X.Sym().Pkg == ir.Pkgs.Runtime && (fn == "throwinit" || fn == "gopanic" || fn == "panicwrap" || fn == "block" || fn == "panicmakeslicelen" || fn == "panicmakeslicecap" || fn == "panicunsafeslicelen" || fn == "panicunsafeslicenilptr" || fn == "panicunsafestringlen" || fn == "panicunsafestringnilptr") { |
| m := s.mem() |
| b := s.endBlock() |
| b.Kind = ssa.BlockExit |
| b.SetControl(m) |
| // TODO: never rewrite OPANIC to OCALLFUNC in the |
| // first place. Need to wait until all backends |
| // go through SSA. |
| } |
| } |
| case ir.ODEFER: |
| n := n.(*ir.GoDeferStmt) |
| if base.Debug.Defer > 0 { |
| var defertype string |
| if s.hasOpenDefers { |
| defertype = "open-coded" |
| } else if n.Esc() == ir.EscNever { |
| defertype = "stack-allocated" |
| } else { |
| defertype = "heap-allocated" |
| } |
| base.WarnfAt(n.Pos(), "%s defer", defertype) |
| } |
| if s.hasOpenDefers { |
| s.openDeferRecord(n.Call.(*ir.CallExpr)) |
| } else { |
| d := callDefer |
| if n.Esc() == ir.EscNever && n.DeferAt == nil { |
| d = callDeferStack |
| } |
| s.call(n.Call.(*ir.CallExpr), d, false, n.DeferAt) |
| } |
| case ir.OGO: |
| n := n.(*ir.GoDeferStmt) |
| s.callResult(n.Call.(*ir.CallExpr), callGo) |
| |
| case ir.OAS2DOTTYPE: |
| n := n.(*ir.AssignListStmt) |
| var res, resok *ssa.Value |
| if n.Rhs[0].Op() == ir.ODOTTYPE2 { |
| res, resok = s.dottype(n.Rhs[0].(*ir.TypeAssertExpr), true) |
| } else { |
| res, resok = s.dynamicDottype(n.Rhs[0].(*ir.DynamicTypeAssertExpr), true) |
| } |
| deref := false |
| if !TypeOK(n.Rhs[0].Type()) { |
| if res.Op != ssa.OpLoad { |
| s.Fatalf("dottype of non-load") |
| } |
| mem := s.mem() |
| if res.Args[1] != mem { |
| s.Fatalf("memory no longer live from 2-result dottype load") |
| } |
| deref = true |
| res = res.Args[0] |
| } |
| s.assign(n.Lhs[0], res, deref, 0) |
| s.assign(n.Lhs[1], resok, false, 0) |
| return |
| |
| case ir.OAS2FUNC: |
| // We come here only when it is an intrinsic call returning two values. |
| n := n.(*ir.AssignListStmt) |
| call := n.Rhs[0].(*ir.CallExpr) |
| if !ir.IsIntrinsicCall(call) { |
| s.Fatalf("non-intrinsic AS2FUNC not expanded %v", call) |
| } |
| v := s.intrinsicCall(call) |
| v1 := s.newValue1(ssa.OpSelect0, n.Lhs[0].Type(), v) |
| v2 := s.newValue1(ssa.OpSelect1, n.Lhs[1].Type(), v) |
| s.assign(n.Lhs[0], v1, false, 0) |
| s.assign(n.Lhs[1], v2, false, 0) |
| return |
| |
| case ir.ODCL: |
| n := n.(*ir.Decl) |
| if v := n.X; v.Esc() == ir.EscHeap { |
| s.newHeapaddr(v) |
| } |
| |
| case ir.OLABEL: |
| n := n.(*ir.LabelStmt) |
| sym := n.Label |
| if sym.IsBlank() { |
| // Nothing to do because the label isn't targetable. See issue 52278. |
| break |
| } |
| lab := s.label(sym) |
| |
| // The label might already have a target block via a goto. |
| if lab.target == nil { |
| lab.target = s.f.NewBlock(ssa.BlockPlain) |
| } |
| |
| // Go to that label. |
| // (We pretend "label:" is preceded by "goto label", unless the predecessor is unreachable.) |
| if s.curBlock != nil { |
| b := s.endBlock() |
| b.AddEdgeTo(lab.target) |
| } |
| s.startBlock(lab.target) |
| |
| case ir.OGOTO: |
| n := n.(*ir.BranchStmt) |
| sym := n.Label |
| |
| lab := s.label(sym) |
| if lab.target == nil { |
| lab.target = s.f.NewBlock(ssa.BlockPlain) |
| } |
| |
| b := s.endBlock() |
| b.Pos = s.lastPos.WithIsStmt() // Do this even if b is an empty block. |
| b.AddEdgeTo(lab.target) |
| |
| case ir.OAS: |
| n := n.(*ir.AssignStmt) |
| if n.X == n.Y && n.X.Op() == ir.ONAME { |
| // An x=x assignment. No point in doing anything |
| // here. In addition, skipping this assignment |
| // prevents generating: |
| // VARDEF x |
| // COPY x -> x |
| // which is bad because x is incorrectly considered |
| // dead before the vardef. See issue #14904. |
| return |
| } |
| |
| // mayOverlap keeps track of whether the LHS and RHS might |
| // refer to partially overlapping memory. Partial overlapping can |
| // only happen for arrays, see the comment in moveWhichMayOverlap. |
| // |
| // If both sides of the assignment are not dereferences, then partial |
| // overlap can't happen. Partial overlap can only occur only when the |
| // arrays referenced are strictly smaller parts of the same base array. |
| // If one side of the assignment is a full array, then partial overlap |
| // can't happen. (The arrays are either disjoint or identical.) |
| mayOverlap := n.X.Op() == ir.ODEREF && (n.Y != nil && n.Y.Op() == ir.ODEREF) |
| if n.Y != nil && n.Y.Op() == ir.ODEREF { |
| p := n.Y.(*ir.StarExpr).X |
| for p.Op() == ir.OCONVNOP { |
| p = p.(*ir.ConvExpr).X |
| } |
| if p.Op() == ir.OSPTR && p.(*ir.UnaryExpr).X.Type().IsString() { |
| // Pointer fields of strings point to unmodifiable memory. |
| // That memory can't overlap with the memory being written. |
| mayOverlap = false |
| } |
| } |
| |
| // Evaluate RHS. |
| rhs := n.Y |
| if rhs != nil { |
| switch rhs.Op() { |
| case ir.OSTRUCTLIT, ir.OARRAYLIT, ir.OSLICELIT: |
| // All literals with nonzero fields have already been |
| // rewritten during walk. Any that remain are just T{} |
| // or equivalents. Use the zero value. |
| if !ir.IsZero(rhs) { |
| s.Fatalf("literal with nonzero value in SSA: %v", rhs) |
| } |
| rhs = nil |
| case ir.OAPPEND: |
| rhs := rhs.(*ir.CallExpr) |
| // Check whether we're writing the result of an append back to the same slice. |
| // If so, we handle it specially to avoid write barriers on the fast |
| // (non-growth) path. |
| if !ir.SameSafeExpr(n.X, rhs.Args[0]) || base.Flag.N != 0 { |
| break |
| } |
| // If the slice can be SSA'd, it'll be on the stack, |
| // so there will be no write barriers, |
| // so there's no need to attempt to prevent them. |
| if s.canSSA(n.X) { |
| if base.Debug.Append > 0 { // replicating old diagnostic message |
| base.WarnfAt(n.Pos(), "append: len-only update (in local slice)") |
| } |
| break |
| } |
| if base.Debug.Append > 0 { |
| base.WarnfAt(n.Pos(), "append: len-only update") |
| } |
| s.append(rhs, true) |
| return |
| } |
| } |
| |
| if ir.IsBlank(n.X) { |
| // _ = rhs |
| // Just evaluate rhs for side-effects. |
| if rhs != nil { |
| s.expr(rhs) |
| } |
| return |
| } |
| |
| var t *types.Type |
| if n.Y != nil { |
| t = n.Y.Type() |
| } else { |
| t = n.X.Type() |
| } |
| |
| var r *ssa.Value |
| deref := !TypeOK(t) |
| if deref { |
| if rhs == nil { |
| r = nil // Signal assign to use OpZero. |
| } else { |
| r = s.addr(rhs) |
| } |
| } else { |
| if rhs == nil { |
| r = s.zeroVal(t) |
| } else { |
| r = s.expr(rhs) |
| } |
| } |
| |
| var skip skipMask |
| if rhs != nil && (rhs.Op() == ir.OSLICE || rhs.Op() == ir.OSLICE3 || rhs.Op() == ir.OSLICESTR) && ir.SameSafeExpr(rhs.(*ir.SliceExpr).X, n.X) { |
| // We're assigning a slicing operation back to its source. |
| // Don't write back fields we aren't changing. See issue #14855. |
| rhs := rhs.(*ir.SliceExpr) |
| i, j, k := rhs.Low, rhs.High, rhs.Max |
| if i != nil && (i.Op() == ir.OLITERAL && i.Val().Kind() == constant.Int && ir.Int64Val(i) == 0) { |
| // [0:...] is the same as [:...] |
| i = nil |
| } |
| // TODO: detect defaults for len/cap also. |
| // Currently doesn't really work because (*p)[:len(*p)] appears here as: |
| // tmp = len(*p) |
| // (*p)[:tmp] |
| // if j != nil && (j.Op == OLEN && SameSafeExpr(j.Left, n.Left)) { |
| // j = nil |
| // } |
| // if k != nil && (k.Op == OCAP && SameSafeExpr(k.Left, n.Left)) { |
| // k = nil |
| // } |
| if i == nil { |
| skip |= skipPtr |
| if j == nil { |
| skip |= skipLen |
| } |
| if k == nil { |
| skip |= skipCap |
| } |
| } |
| } |
| |
| s.assignWhichMayOverlap(n.X, r, deref, skip, mayOverlap) |
| |
| case ir.OIF: |
| n := n.(*ir.IfStmt) |
| if ir.IsConst(n.Cond, constant.Bool) { |
| s.stmtList(n.Cond.Init()) |
| if ir.BoolVal(n.Cond) { |
| s.stmtList(n.Body) |
| } else { |
| s.stmtList(n.Else) |
| } |
| break |
| } |
| |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| var likely int8 |
| if n.Likely { |
| likely = 1 |
| } |
| var bThen *ssa.Block |
| if len(n.Body) != 0 { |
| bThen = s.f.NewBlock(ssa.BlockPlain) |
| } else { |
| bThen = bEnd |
| } |
| var bElse *ssa.Block |
| if len(n.Else) != 0 { |
| bElse = s.f.NewBlock(ssa.BlockPlain) |
| } else { |
| bElse = bEnd |
| } |
| s.condBranch(n.Cond, bThen, bElse, likely) |
| |
| if len(n.Body) != 0 { |
| s.startBlock(bThen) |
| s.stmtList(n.Body) |
| if b := s.endBlock(); b != nil { |
| b.AddEdgeTo(bEnd) |
| } |
| } |
| if len(n.Else) != 0 { |
| s.startBlock(bElse) |
| s.stmtList(n.Else) |
| if b := s.endBlock(); b != nil { |
| b.AddEdgeTo(bEnd) |
| } |
| } |
| s.startBlock(bEnd) |
| |
| case ir.ORETURN: |
| n := n.(*ir.ReturnStmt) |
| s.stmtList(n.Results) |
| b := s.exit() |
| b.Pos = s.lastPos.WithIsStmt() |
| |
| case ir.OTAILCALL: |
| n := n.(*ir.TailCallStmt) |
| s.callResult(n.Call, callTail) |
| call := s.mem() |
| b := s.endBlock() |
| b.Kind = ssa.BlockRetJmp // could use BlockExit. BlockRetJmp is mostly for clarity. |
| b.SetControl(call) |
| |
| case ir.OCONTINUE, ir.OBREAK: |
| n := n.(*ir.BranchStmt) |
| var to *ssa.Block |
| if n.Label == nil { |
| // plain break/continue |
| switch n.Op() { |
| case ir.OCONTINUE: |
| to = s.continueTo |
| case ir.OBREAK: |
| to = s.breakTo |
| } |
| } else { |
| // labeled break/continue; look up the target |
| sym := n.Label |
| lab := s.label(sym) |
| switch n.Op() { |
| case ir.OCONTINUE: |
| to = lab.continueTarget |
| case ir.OBREAK: |
| to = lab.breakTarget |
| } |
| } |
| |
| b := s.endBlock() |
| b.Pos = s.lastPos.WithIsStmt() // Do this even if b is an empty block. |
| b.AddEdgeTo(to) |
| |
| case ir.OFOR: |
| // OFOR: for Ninit; Left; Right { Nbody } |
| // cond (Left); body (Nbody); incr (Right) |
| n := n.(*ir.ForStmt) |
| base.Assert(!n.DistinctVars) // Should all be rewritten before escape analysis |
| bCond := s.f.NewBlock(ssa.BlockPlain) |
| bBody := s.f.NewBlock(ssa.BlockPlain) |
| bIncr := s.f.NewBlock(ssa.BlockPlain) |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| |
| // ensure empty for loops have correct position; issue #30167 |
| bBody.Pos = n.Pos() |
| |
| // first, jump to condition test |
| b := s.endBlock() |
| b.AddEdgeTo(bCond) |
| |
| // generate code to test condition |
| s.startBlock(bCond) |
| if n.Cond != nil { |
| s.condBranch(n.Cond, bBody, bEnd, 1) |
| } else { |
| b := s.endBlock() |
| b.Kind = ssa.BlockPlain |
| b.AddEdgeTo(bBody) |
| } |
| |
| // set up for continue/break in body |
| prevContinue := s.continueTo |
| prevBreak := s.breakTo |
| s.continueTo = bIncr |
| s.breakTo = bEnd |
| var lab *ssaLabel |
| if sym := n.Label; sym != nil { |
| // labeled for loop |
| lab = s.label(sym) |
| lab.continueTarget = bIncr |
| lab.breakTarget = bEnd |
| } |
| |
| // generate body |
| s.startBlock(bBody) |
| s.stmtList(n.Body) |
| |
| // tear down continue/break |
| s.continueTo = prevContinue |
| s.breakTo = prevBreak |
| if lab != nil { |
| lab.continueTarget = nil |
| lab.breakTarget = nil |
| } |
| |
| // done with body, goto incr |
| if b := s.endBlock(); b != nil { |
| b.AddEdgeTo(bIncr) |
| } |
| |
| // generate incr |
| s.startBlock(bIncr) |
| if n.Post != nil { |
| s.stmt(n.Post) |
| } |
| if b := s.endBlock(); b != nil { |
| b.AddEdgeTo(bCond) |
| // It can happen that bIncr ends in a block containing only VARKILL, |
| // and that muddles the debugging experience. |
| if b.Pos == src.NoXPos { |
| b.Pos = bCond.Pos |
| } |
| } |
| |
| s.startBlock(bEnd) |
| |
| case ir.OSWITCH, ir.OSELECT: |
| // These have been mostly rewritten by the front end into their Nbody fields. |
| // Our main task is to correctly hook up any break statements. |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| |
| prevBreak := s.breakTo |
| s.breakTo = bEnd |
| var sym *types.Sym |
| var body ir.Nodes |
| if n.Op() == ir.OSWITCH { |
| n := n.(*ir.SwitchStmt) |
| sym = n.Label |
| body = n.Compiled |
| } else { |
| n := n.(*ir.SelectStmt) |
| sym = n.Label |
| body = n.Compiled |
| } |
| |
| var lab *ssaLabel |
| if sym != nil { |
| // labeled |
| lab = s.label(sym) |
| lab.breakTarget = bEnd |
| } |
| |
| // generate body code |
| s.stmtList(body) |
| |
| s.breakTo = prevBreak |
| if lab != nil { |
| lab.breakTarget = nil |
| } |
| |
| // walk adds explicit OBREAK nodes to the end of all reachable code paths. |
| // If we still have a current block here, then mark it unreachable. |
| if s.curBlock != nil { |
| m := s.mem() |
| b := s.endBlock() |
| b.Kind = ssa.BlockExit |
| b.SetControl(m) |
| } |
| s.startBlock(bEnd) |
| |
| case ir.OJUMPTABLE: |
| n := n.(*ir.JumpTableStmt) |
| |
| // Make blocks we'll need. |
| jt := s.f.NewBlock(ssa.BlockJumpTable) |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| |
| // The only thing that needs evaluating is the index we're looking up. |
| idx := s.expr(n.Idx) |
| unsigned := idx.Type.IsUnsigned() |
| |
| // Extend so we can do everything in uintptr arithmetic. |
| t := types.Types[types.TUINTPTR] |
| idx = s.conv(nil, idx, idx.Type, t) |
| |
| // The ending condition for the current block decides whether we'll use |
| // the jump table at all. |
| // We check that min <= idx <= max and jump around the jump table |
| // if that test fails. |
| // We implement min <= idx <= max with 0 <= idx-min <= max-min, because |
| // we'll need idx-min anyway as the control value for the jump table. |
| var min, max uint64 |
| if unsigned { |
| min, _ = constant.Uint64Val(n.Cases[0]) |
| max, _ = constant.Uint64Val(n.Cases[len(n.Cases)-1]) |
| } else { |
| mn, _ := constant.Int64Val(n.Cases[0]) |
| mx, _ := constant.Int64Val(n.Cases[len(n.Cases)-1]) |
| min = uint64(mn) |
| max = uint64(mx) |
| } |
| // Compare idx-min with max-min, to see if we can use the jump table. |
| idx = s.newValue2(s.ssaOp(ir.OSUB, t), t, idx, s.uintptrConstant(min)) |
| width := s.uintptrConstant(max - min) |
| cmp := s.newValue2(s.ssaOp(ir.OLE, t), types.Types[types.TBOOL], idx, width) |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.SetControl(cmp) |
| b.AddEdgeTo(jt) // in range - use jump table |
| b.AddEdgeTo(bEnd) // out of range - no case in the jump table will trigger |
| b.Likely = ssa.BranchLikely // TODO: assumes missing the table entirely is unlikely. True? |
| |
| // Build jump table block. |
| s.startBlock(jt) |
| jt.Pos = n.Pos() |
| if base.Flag.Cfg.SpectreIndex { |
| idx = s.newValue2(ssa.OpSpectreSliceIndex, t, idx, width) |
| } |
| jt.SetControl(idx) |
| |
| // Figure out where we should go for each index in the table. |
| table := make([]*ssa.Block, max-min+1) |
| for i := range table { |
| table[i] = bEnd // default target |
| } |
| for i := range n.Targets { |
| c := n.Cases[i] |
| lab := s.label(n.Targets[i]) |
| if lab.target == nil { |
| lab.target = s.f.NewBlock(ssa.BlockPlain) |
| } |
| var val uint64 |
| if unsigned { |
| val, _ = constant.Uint64Val(c) |
| } else { |
| vl, _ := constant.Int64Val(c) |
| val = uint64(vl) |
| } |
| // Overwrite the default target. |
| table[val-min] = lab.target |
| } |
| for _, t := range table { |
| jt.AddEdgeTo(t) |
| } |
| s.endBlock() |
| |
| s.startBlock(bEnd) |
| |
| case ir.OCHECKNIL: |
| n := n.(*ir.UnaryExpr) |
| p := s.expr(n.X) |
| s.nilCheck(p) |
| |
| case ir.OINLMARK: |
| n := n.(*ir.InlineMarkStmt) |
| s.newValue1I(ssa.OpInlMark, types.TypeVoid, n.Index, s.mem()) |
| |
| default: |
| s.Fatalf("unhandled stmt %v", n.Op()) |
| } |
| } |
| |
| // If true, share as many open-coded defer exits as possible (with the downside of |
| // worse line-number information) |
| const shareDeferExits = false |
| |
| // exit processes any code that needs to be generated just before returning. |
| // It returns a BlockRet block that ends the control flow. Its control value |
| // will be set to the final memory state. |
| func (s *state) exit() *ssa.Block { |
| if s.hasdefer { |
| if s.hasOpenDefers { |
| if shareDeferExits && s.lastDeferExit != nil && len(s.openDefers) == s.lastDeferCount { |
| if s.curBlock.Kind != ssa.BlockPlain { |
| panic("Block for an exit should be BlockPlain") |
| } |
| s.curBlock.AddEdgeTo(s.lastDeferExit) |
| s.endBlock() |
| return s.lastDeferFinalBlock |
| } |
| s.openDeferExit() |
| } else { |
| s.rtcall(ir.Syms.Deferreturn, true, nil) |
| } |
| } |
| |
| var b *ssa.Block |
| var m *ssa.Value |
| // Do actual return. |
| // These currently turn into self-copies (in many cases). |
| resultFields := s.curfn.Type().Results().FieldSlice() |
| results := make([]*ssa.Value, len(resultFields)+1, len(resultFields)+1) |
| m = s.newValue0(ssa.OpMakeResult, s.f.OwnAux.LateExpansionResultType()) |
| // Store SSAable and heap-escaped PPARAMOUT variables back to stack locations. |
| for i, f := range resultFields { |
| n := f.Nname.(*ir.Name) |
| if s.canSSA(n) { // result is in some SSA variable |
| if !n.IsOutputParamInRegisters() && n.Type().HasPointers() { |
| // We are about to store to the result slot. |
| s.vars[memVar] = s.newValue1A(ssa.OpVarDef, types.TypeMem, n, s.mem()) |
| } |
| results[i] = s.variable(n, n.Type()) |
| } else if !n.OnStack() { // result is actually heap allocated |
| // We are about to copy the in-heap result to the result slot. |
| if n.Type().HasPointers() { |
| s.vars[memVar] = s.newValue1A(ssa.OpVarDef, types.TypeMem, n, s.mem()) |
| } |
| ha := s.expr(n.Heapaddr) |
| s.instrumentFields(n.Type(), ha, instrumentRead) |
| results[i] = s.newValue2(ssa.OpDereference, n.Type(), ha, s.mem()) |
| } else { // result is not SSA-able; not escaped, so not on heap, but too large for SSA. |
| // Before register ABI this ought to be a self-move, home=dest, |
| // With register ABI, it's still a self-move if parameter is on stack (i.e., too big or overflowed) |
| // No VarDef, as the result slot is already holding live value. |
| results[i] = s.newValue2(ssa.OpDereference, n.Type(), s.addr(n), s.mem()) |
| } |
| } |
| |
| // Run exit code. Today, this is just racefuncexit, in -race mode. |
| // TODO(register args) this seems risky here with a register-ABI, but not clear it is right to do it earlier either. |
| // Spills in register allocation might just fix it. |
| s.stmtList(s.curfn.Exit) |
| |
| results[len(results)-1] = s.mem() |
| m.AddArgs(results...) |
| |
| b = s.endBlock() |
| b.Kind = ssa.BlockRet |
| b.SetControl(m) |
| if s.hasdefer && s.hasOpenDefers { |
| s.lastDeferFinalBlock = b |
| } |
| return b |
| } |
| |
| type opAndType struct { |
| op ir.Op |
| etype types.Kind |
| } |
| |
| var opToSSA = map[opAndType]ssa.Op{ |
| {ir.OADD, types.TINT8}: ssa.OpAdd8, |
| {ir.OADD, types.TUINT8}: ssa.OpAdd8, |
| {ir.OADD, types.TINT16}: ssa.OpAdd16, |
| {ir.OADD, types.TUINT16}: ssa.OpAdd16, |
| {ir.OADD, types.TINT32}: ssa.OpAdd32, |
| {ir.OADD, types.TUINT32}: ssa.OpAdd32, |
| {ir.OADD, types.TINT64}: ssa.OpAdd64, |
| {ir.OADD, types.TUINT64}: ssa.OpAdd64, |
| {ir.OADD, types.TFLOAT32}: ssa.OpAdd32F, |
| {ir.OADD, types.TFLOAT64}: ssa.OpAdd64F, |
| |
| {ir.OSUB, types.TINT8}: ssa.OpSub8, |
| {ir.OSUB, types.TUINT8}: ssa.OpSub8, |
| {ir.OSUB, types.TINT16}: ssa.OpSub16, |
| {ir.OSUB, types.TUINT16}: ssa.OpSub16, |
| {ir.OSUB, types.TINT32}: ssa.OpSub32, |
| {ir.OSUB, types.TUINT32}: ssa.OpSub32, |
| {ir.OSUB, types.TINT64}: ssa.OpSub64, |
| {ir.OSUB, types.TUINT64}: ssa.OpSub64, |
| {ir.OSUB, types.TFLOAT32}: ssa.OpSub32F, |
| {ir.OSUB, types.TFLOAT64}: ssa.OpSub64F, |
| |
| {ir.ONOT, types.TBOOL}: ssa.OpNot, |
| |
| {ir.ONEG, types.TINT8}: ssa.OpNeg8, |
| {ir.ONEG, types.TUINT8}: ssa.OpNeg8, |
| {ir.ONEG, types.TINT16}: ssa.OpNeg16, |
| {ir.ONEG, types.TUINT16}: ssa.OpNeg16, |
| {ir.ONEG, types.TINT32}: ssa.OpNeg32, |
| {ir.ONEG, types.TUINT32}: ssa.OpNeg32, |
| {ir.ONEG, types.TINT64}: ssa.OpNeg64, |
| {ir.ONEG, types.TUINT64}: ssa.OpNeg64, |
| {ir.ONEG, types.TFLOAT32}: ssa.OpNeg32F, |
| {ir.ONEG, types.TFLOAT64}: ssa.OpNeg64F, |
| |
| {ir.OBITNOT, types.TINT8}: ssa.OpCom8, |
| {ir.OBITNOT, types.TUINT8}: ssa.OpCom8, |
| {ir.OBITNOT, types.TINT16}: ssa.OpCom16, |
| {ir.OBITNOT, types.TUINT16}: ssa.OpCom16, |
| {ir.OBITNOT, types.TINT32}: ssa.OpCom32, |
| {ir.OBITNOT, types.TUINT32}: ssa.OpCom32, |
| {ir.OBITNOT, types.TINT64}: ssa.OpCom64, |
| {ir.OBITNOT, types.TUINT64}: ssa.OpCom64, |
| |
| {ir.OIMAG, types.TCOMPLEX64}: ssa.OpComplexImag, |
| {ir.OIMAG, types.TCOMPLEX128}: ssa.OpComplexImag, |
| {ir.OREAL, types.TCOMPLEX64}: ssa.OpComplexReal, |
| {ir.OREAL, types.TCOMPLEX128}: ssa.OpComplexReal, |
| |
| {ir.OMUL, types.TINT8}: ssa.OpMul8, |
| {ir.OMUL, types.TUINT8}: ssa.OpMul8, |
| {ir.OMUL, types.TINT16}: ssa.OpMul16, |
| {ir.OMUL, types.TUINT16}: ssa.OpMul16, |
| {ir.OMUL, types.TINT32}: ssa.OpMul32, |
| {ir.OMUL, types.TUINT32}: ssa.OpMul32, |
| {ir.OMUL, types.TINT64}: ssa.OpMul64, |
| {ir.OMUL, types.TUINT64}: ssa.OpMul64, |
| {ir.OMUL, types.TFLOAT32}: ssa.OpMul32F, |
| {ir.OMUL, types.TFLOAT64}: ssa.OpMul64F, |
| |
| {ir.ODIV, types.TFLOAT32}: ssa.OpDiv32F, |
| {ir.ODIV, types.TFLOAT64}: ssa.OpDiv64F, |
| |
| {ir.ODIV, types.TINT8}: ssa.OpDiv8, |
| {ir.ODIV, types.TUINT8}: ssa.OpDiv8u, |
| {ir.ODIV, types.TINT16}: ssa.OpDiv16, |
| {ir.ODIV, types.TUINT16}: ssa.OpDiv16u, |
| {ir.ODIV, types.TINT32}: ssa.OpDiv32, |
| {ir.ODIV, types.TUINT32}: ssa.OpDiv32u, |
| {ir.ODIV, types.TINT64}: ssa.OpDiv64, |
| {ir.ODIV, types.TUINT64}: ssa.OpDiv64u, |
| |
| {ir.OMOD, types.TINT8}: ssa.OpMod8, |
| {ir.OMOD, types.TUINT8}: ssa.OpMod8u, |
| {ir.OMOD, types.TINT16}: ssa.OpMod16, |
| {ir.OMOD, types.TUINT16}: ssa.OpMod16u, |
| {ir.OMOD, types.TINT32}: ssa.OpMod32, |
| {ir.OMOD, types.TUINT32}: ssa.OpMod32u, |
| {ir.OMOD, types.TINT64}: ssa.OpMod64, |
| {ir.OMOD, types.TUINT64}: ssa.OpMod64u, |
| |
| {ir.OAND, types.TINT8}: ssa.OpAnd8, |
| {ir.OAND, types.TUINT8}: ssa.OpAnd8, |
| {ir.OAND, types.TINT16}: ssa.OpAnd16, |
| {ir.OAND, types.TUINT16}: ssa.OpAnd16, |
| {ir.OAND, types.TINT32}: ssa.OpAnd32, |
| {ir.OAND, types.TUINT32}: ssa.OpAnd32, |
| {ir.OAND, types.TINT64}: ssa.OpAnd64, |
| {ir.OAND, types.TUINT64}: ssa.OpAnd64, |
| |
| {ir.OOR, types.TINT8}: ssa.OpOr8, |
| {ir.OOR, types.TUINT8}: ssa.OpOr8, |
| {ir.OOR, types.TINT16}: ssa.OpOr16, |
| {ir.OOR, types.TUINT16}: ssa.OpOr16, |
| {ir.OOR, types.TINT32}: ssa.OpOr32, |
| {ir.OOR, types.TUINT32}: ssa.OpOr32, |
| {ir.OOR, types.TINT64}: ssa.OpOr64, |
| {ir.OOR, types.TUINT64}: ssa.OpOr64, |
| |
| {ir.OXOR, types.TINT8}: ssa.OpXor8, |
| {ir.OXOR, types.TUINT8}: ssa.OpXor8, |
| {ir.OXOR, types.TINT16}: ssa.OpXor16, |
| {ir.OXOR, types.TUINT16}: ssa.OpXor16, |
| {ir.OXOR, types.TINT32}: ssa.OpXor32, |
| {ir.OXOR, types.TUINT32}: ssa.OpXor32, |
| {ir.OXOR, types.TINT64}: ssa.OpXor64, |
| {ir.OXOR, types.TUINT64}: ssa.OpXor64, |
| |
| {ir.OEQ, types.TBOOL}: ssa.OpEqB, |
| {ir.OEQ, types.TINT8}: ssa.OpEq8, |
| {ir.OEQ, types.TUINT8}: ssa.OpEq8, |
| {ir.OEQ, types.TINT16}: ssa.OpEq16, |
| {ir.OEQ, types.TUINT16}: ssa.OpEq16, |
| {ir.OEQ, types.TINT32}: ssa.OpEq32, |
| {ir.OEQ, types.TUINT32}: ssa.OpEq32, |
| {ir.OEQ, types.TINT64}: ssa.OpEq64, |
| {ir.OEQ, types.TUINT64}: ssa.OpEq64, |
| {ir.OEQ, types.TINTER}: ssa.OpEqInter, |
| {ir.OEQ, types.TSLICE}: ssa.OpEqSlice, |
| {ir.OEQ, types.TFUNC}: ssa.OpEqPtr, |
| {ir.OEQ, types.TMAP}: ssa.OpEqPtr, |
| {ir.OEQ, types.TCHAN}: ssa.OpEqPtr, |
| {ir.OEQ, types.TPTR}: ssa.OpEqPtr, |
| {ir.OEQ, types.TUINTPTR}: ssa.OpEqPtr, |
| {ir.OEQ, types.TUNSAFEPTR}: ssa.OpEqPtr, |
| {ir.OEQ, types.TFLOAT64}: ssa.OpEq64F, |
| {ir.OEQ, types.TFLOAT32}: ssa.OpEq32F, |
| |
| {ir.ONE, types.TBOOL}: ssa.OpNeqB, |
| {ir.ONE, types.TINT8}: ssa.OpNeq8, |
| {ir.ONE, types.TUINT8}: ssa.OpNeq8, |
| {ir.ONE, types.TINT16}: ssa.OpNeq16, |
| {ir.ONE, types.TUINT16}: ssa.OpNeq16, |
| {ir.ONE, types.TINT32}: ssa.OpNeq32, |
| {ir.ONE, types.TUINT32}: ssa.OpNeq32, |
| {ir.ONE, types.TINT64}: ssa.OpNeq64, |
| {ir.ONE, types.TUINT64}: ssa.OpNeq64, |
| {ir.ONE, types.TINTER}: ssa.OpNeqInter, |
| {ir.ONE, types.TSLICE}: ssa.OpNeqSlice, |
| {ir.ONE, types.TFUNC}: ssa.OpNeqPtr, |
| {ir.ONE, types.TMAP}: ssa.OpNeqPtr, |
| {ir.ONE, types.TCHAN}: ssa.OpNeqPtr, |
| {ir.ONE, types.TPTR}: ssa.OpNeqPtr, |
| {ir.ONE, types.TUINTPTR}: ssa.OpNeqPtr, |
| {ir.ONE, types.TUNSAFEPTR}: ssa.OpNeqPtr, |
| {ir.ONE, types.TFLOAT64}: ssa.OpNeq64F, |
| {ir.ONE, types.TFLOAT32}: ssa.OpNeq32F, |
| |
| {ir.OLT, types.TINT8}: ssa.OpLess8, |
| {ir.OLT, types.TUINT8}: ssa.OpLess8U, |
| {ir.OLT, types.TINT16}: ssa.OpLess16, |
| {ir.OLT, types.TUINT16}: ssa.OpLess16U, |
| {ir.OLT, types.TINT32}: ssa.OpLess32, |
| {ir.OLT, types.TUINT32}: ssa.OpLess32U, |
| {ir.OLT, types.TINT64}: ssa.OpLess64, |
| {ir.OLT, types.TUINT64}: ssa.OpLess64U, |
| {ir.OLT, types.TFLOAT64}: ssa.OpLess64F, |
| {ir.OLT, types.TFLOAT32}: ssa.OpLess32F, |
| |
| {ir.OLE, types.TINT8}: ssa.OpLeq8, |
| {ir.OLE, types.TUINT8}: ssa.OpLeq8U, |
| {ir.OLE, types.TINT16}: ssa.OpLeq16, |
| {ir.OLE, types.TUINT16}: ssa.OpLeq16U, |
| {ir.OLE, types.TINT32}: ssa.OpLeq32, |
| {ir.OLE, types.TUINT32}: ssa.OpLeq32U, |
| {ir.OLE, types.TINT64}: ssa.OpLeq64, |
| {ir.OLE, types.TUINT64}: ssa.OpLeq64U, |
| {ir.OLE, types.TFLOAT64}: ssa.OpLeq64F, |
| {ir.OLE, types.TFLOAT32}: ssa.OpLeq32F, |
| } |
| |
| func (s *state) concreteEtype(t *types.Type) types.Kind { |
| e := t.Kind() |
| switch e { |
| default: |
| return e |
| case types.TINT: |
| if s.config.PtrSize == 8 { |
| return types.TINT64 |
| } |
| return types.TINT32 |
| case types.TUINT: |
| if s.config.PtrSize == 8 { |
| return types.TUINT64 |
| } |
| return types.TUINT32 |
| case types.TUINTPTR: |
| if s.config.PtrSize == 8 { |
| return types.TUINT64 |
| } |
| return types.TUINT32 |
| } |
| } |
| |
| func (s *state) ssaOp(op ir.Op, t *types.Type) ssa.Op { |
| etype := s.concreteEtype(t) |
| x, ok := opToSSA[opAndType{op, etype}] |
| if !ok { |
| s.Fatalf("unhandled binary op %v %s", op, etype) |
| } |
| return x |
| } |
| |
| type opAndTwoTypes struct { |
| op ir.Op |
| etype1 types.Kind |
| etype2 types.Kind |
| } |
| |
| type twoTypes struct { |
| etype1 types.Kind |
| etype2 types.Kind |
| } |
| |
| type twoOpsAndType struct { |
| op1 ssa.Op |
| op2 ssa.Op |
| intermediateType types.Kind |
| } |
| |
| var fpConvOpToSSA = map[twoTypes]twoOpsAndType{ |
| |
| {types.TINT8, types.TFLOAT32}: {ssa.OpSignExt8to32, ssa.OpCvt32to32F, types.TINT32}, |
| {types.TINT16, types.TFLOAT32}: {ssa.OpSignExt16to32, ssa.OpCvt32to32F, types.TINT32}, |
| {types.TINT32, types.TFLOAT32}: {ssa.OpCopy, ssa.OpCvt32to32F, types.TINT32}, |
| {types.TINT64, types.TFLOAT32}: {ssa.OpCopy, ssa.OpCvt64to32F, types.TINT64}, |
| |
| {types.TINT8, types.TFLOAT64}: {ssa.OpSignExt8to32, ssa.OpCvt32to64F, types.TINT32}, |
| {types.TINT16, types.TFLOAT64}: {ssa.OpSignExt16to32, ssa.OpCvt32to64F, types.TINT32}, |
| {types.TINT32, types.TFLOAT64}: {ssa.OpCopy, ssa.OpCvt32to64F, types.TINT32}, |
| {types.TINT64, types.TFLOAT64}: {ssa.OpCopy, ssa.OpCvt64to64F, types.TINT64}, |
| |
| {types.TFLOAT32, types.TINT8}: {ssa.OpCvt32Fto32, ssa.OpTrunc32to8, types.TINT32}, |
| {types.TFLOAT32, types.TINT16}: {ssa.OpCvt32Fto32, ssa.OpTrunc32to16, types.TINT32}, |
| {types.TFLOAT32, types.TINT32}: {ssa.OpCvt32Fto32, ssa.OpCopy, types.TINT32}, |
| {types.TFLOAT32, types.TINT64}: {ssa.OpCvt32Fto64, ssa.OpCopy, types.TINT64}, |
| |
| {types.TFLOAT64, types.TINT8}: {ssa.OpCvt64Fto32, ssa.OpTrunc32to8, types.TINT32}, |
| {types.TFLOAT64, types.TINT16}: {ssa.OpCvt64Fto32, ssa.OpTrunc32to16, types.TINT32}, |
| {types.TFLOAT64, types.TINT32}: {ssa.OpCvt64Fto32, ssa.OpCopy, types.TINT32}, |
| {types.TFLOAT64, types.TINT64}: {ssa.OpCvt64Fto64, ssa.OpCopy, types.TINT64}, |
| // unsigned |
| {types.TUINT8, types.TFLOAT32}: {ssa.OpZeroExt8to32, ssa.OpCvt32to32F, types.TINT32}, |
| {types.TUINT16, types.TFLOAT32}: {ssa.OpZeroExt16to32, ssa.OpCvt32to32F, types.TINT32}, |
| {types.TUINT32, types.TFLOAT32}: {ssa.OpZeroExt32to64, ssa.OpCvt64to32F, types.TINT64}, // go wide to dodge unsigned |
| {types.TUINT64, types.TFLOAT32}: {ssa.OpCopy, ssa.OpInvalid, types.TUINT64}, // Cvt64Uto32F, branchy code expansion instead |
| |
| {types.TUINT8, types.TFLOAT64}: {ssa.OpZeroExt8to32, ssa.OpCvt32to64F, types.TINT32}, |
| {types.TUINT16, types.TFLOAT64}: {ssa.OpZeroExt16to32, ssa.OpCvt32to64F, types.TINT32}, |
| {types.TUINT32, types.TFLOAT64}: {ssa.OpZeroExt32to64, ssa.OpCvt64to64F, types.TINT64}, // go wide to dodge unsigned |
| {types.TUINT64, types.TFLOAT64}: {ssa.OpCopy, ssa.OpInvalid, types.TUINT64}, // Cvt64Uto64F, branchy code expansion instead |
| |
| {types.TFLOAT32, types.TUINT8}: {ssa.OpCvt32Fto32, ssa.OpTrunc32to8, types.TINT32}, |
| {types.TFLOAT32, types.TUINT16}: {ssa.OpCvt32Fto32, ssa.OpTrunc32to16, types.TINT32}, |
| {types.TFLOAT32, types.TUINT32}: {ssa.OpCvt32Fto64, ssa.OpTrunc64to32, types.TINT64}, // go wide to dodge unsigned |
| {types.TFLOAT32, types.TUINT64}: {ssa.OpInvalid, ssa.OpCopy, types.TUINT64}, // Cvt32Fto64U, branchy code expansion instead |
| |
| {types.TFLOAT64, types.TUINT8}: {ssa.OpCvt64Fto32, ssa.OpTrunc32to8, types.TINT32}, |
| {types.TFLOAT64, types.TUINT16}: {ssa.OpCvt64Fto32, ssa.OpTrunc32to16, types.TINT32}, |
| {types.TFLOAT64, types.TUINT32}: {ssa.OpCvt64Fto64, ssa.OpTrunc64to32, types.TINT64}, // go wide to dodge unsigned |
| {types.TFLOAT64, types.TUINT64}: {ssa.OpInvalid, ssa.OpCopy, types.TUINT64}, // Cvt64Fto64U, branchy code expansion instead |
| |
| // float |
| {types.TFLOAT64, types.TFLOAT32}: {ssa.OpCvt64Fto32F, ssa.OpCopy, types.TFLOAT32}, |
| {types.TFLOAT64, types.TFLOAT64}: {ssa.OpRound64F, ssa.OpCopy, types.TFLOAT64}, |
| {types.TFLOAT32, types.TFLOAT32}: {ssa.OpRound32F, ssa.OpCopy, types.TFLOAT32}, |
| {types.TFLOAT32, types.TFLOAT64}: {ssa.OpCvt32Fto64F, ssa.OpCopy, types.TFLOAT64}, |
| } |
| |
| // this map is used only for 32-bit arch, and only includes the difference |
| // on 32-bit arch, don't use int64<->float conversion for uint32 |
| var fpConvOpToSSA32 = map[twoTypes]twoOpsAndType{ |
| {types.TUINT32, types.TFLOAT32}: {ssa.OpCopy, ssa.OpCvt32Uto32F, types.TUINT32}, |
| {types.TUINT32, types.TFLOAT64}: {ssa.OpCopy, ssa.OpCvt32Uto64F, types.TUINT32}, |
| {types.TFLOAT32, types.TUINT32}: {ssa.OpCvt32Fto32U, ssa.OpCopy, types.TUINT32}, |
| {types.TFLOAT64, types.TUINT32}: {ssa.OpCvt64Fto32U, ssa.OpCopy, types.TUINT32}, |
| } |
| |
| // uint64<->float conversions, only on machines that have instructions for that |
| var uint64fpConvOpToSSA = map[twoTypes]twoOpsAndType{ |
| {types.TUINT64, types.TFLOAT32}: {ssa.OpCopy, ssa.OpCvt64Uto32F, types.TUINT64}, |
| {types.TUINT64, types.TFLOAT64}: {ssa.OpCopy, ssa.OpCvt64Uto64F, types.TUINT64}, |
| {types.TFLOAT32, types.TUINT64}: {ssa.OpCvt32Fto64U, ssa.OpCopy, types.TUINT64}, |
| {types.TFLOAT64, types.TUINT64}: {ssa.OpCvt64Fto64U, ssa.OpCopy, types.TUINT64}, |
| } |
| |
| var shiftOpToSSA = map[opAndTwoTypes]ssa.Op{ |
| {ir.OLSH, types.TINT8, types.TUINT8}: ssa.OpLsh8x8, |
| {ir.OLSH, types.TUINT8, types.TUINT8}: ssa.OpLsh8x8, |
| {ir.OLSH, types.TINT8, types.TUINT16}: ssa.OpLsh8x16, |
| {ir.OLSH, types.TUINT8, types.TUINT16}: ssa.OpLsh8x16, |
| {ir.OLSH, types.TINT8, types.TUINT32}: ssa.OpLsh8x32, |
| {ir.OLSH, types.TUINT8, types.TUINT32}: ssa.OpLsh8x32, |
| {ir.OLSH, types.TINT8, types.TUINT64}: ssa.OpLsh8x64, |
| {ir.OLSH, types.TUINT8, types.TUINT64}: ssa.OpLsh8x64, |
| |
| {ir.OLSH, types.TINT16, types.TUINT8}: ssa.OpLsh16x8, |
| {ir.OLSH, types.TUINT16, types.TUINT8}: ssa.OpLsh16x8, |
| {ir.OLSH, types.TINT16, types.TUINT16}: ssa.OpLsh16x16, |
| {ir.OLSH, types.TUINT16, types.TUINT16}: ssa.OpLsh16x16, |
| {ir.OLSH, types.TINT16, types.TUINT32}: ssa.OpLsh16x32, |
| {ir.OLSH, types.TUINT16, types.TUINT32}: ssa.OpLsh16x32, |
| {ir.OLSH, types.TINT16, types.TUINT64}: ssa.OpLsh16x64, |
| {ir.OLSH, types.TUINT16, types.TUINT64}: ssa.OpLsh16x64, |
| |
| {ir.OLSH, types.TINT32, types.TUINT8}: ssa.OpLsh32x8, |
| {ir.OLSH, types.TUINT32, types.TUINT8}: ssa.OpLsh32x8, |
| {ir.OLSH, types.TINT32, types.TUINT16}: ssa.OpLsh32x16, |
| {ir.OLSH, types.TUINT32, types.TUINT16}: ssa.OpLsh32x16, |
| {ir.OLSH, types.TINT32, types.TUINT32}: ssa.OpLsh32x32, |
| {ir.OLSH, types.TUINT32, types.TUINT32}: ssa.OpLsh32x32, |
| {ir.OLSH, types.TINT32, types.TUINT64}: ssa.OpLsh32x64, |
| {ir.OLSH, types.TUINT32, types.TUINT64}: ssa.OpLsh32x64, |
| |
| {ir.OLSH, types.TINT64, types.TUINT8}: ssa.OpLsh64x8, |
| {ir.OLSH, types.TUINT64, types.TUINT8}: ssa.OpLsh64x8, |
| {ir.OLSH, types.TINT64, types.TUINT16}: ssa.OpLsh64x16, |
| {ir.OLSH, types.TUINT64, types.TUINT16}: ssa.OpLsh64x16, |
| {ir.OLSH, types.TINT64, types.TUINT32}: ssa.OpLsh64x32, |
| {ir.OLSH, types.TUINT64, types.TUINT32}: ssa.OpLsh64x32, |
| {ir.OLSH, types.TINT64, types.TUINT64}: ssa.OpLsh64x64, |
| {ir.OLSH, types.TUINT64, types.TUINT64}: ssa.OpLsh64x64, |
| |
| {ir.ORSH, types.TINT8, types.TUINT8}: ssa.OpRsh8x8, |
| {ir.ORSH, types.TUINT8, types.TUINT8}: ssa.OpRsh8Ux8, |
| {ir.ORSH, types.TINT8, types.TUINT16}: ssa.OpRsh8x16, |
| {ir.ORSH, types.TUINT8, types.TUINT16}: ssa.OpRsh8Ux16, |
| {ir.ORSH, types.TINT8, types.TUINT32}: ssa.OpRsh8x32, |
| {ir.ORSH, types.TUINT8, types.TUINT32}: ssa.OpRsh8Ux32, |
| {ir.ORSH, types.TINT8, types.TUINT64}: ssa.OpRsh8x64, |
| {ir.ORSH, types.TUINT8, types.TUINT64}: ssa.OpRsh8Ux64, |
| |
| {ir.ORSH, types.TINT16, types.TUINT8}: ssa.OpRsh16x8, |
| {ir.ORSH, types.TUINT16, types.TUINT8}: ssa.OpRsh16Ux8, |
| {ir.ORSH, types.TINT16, types.TUINT16}: ssa.OpRsh16x16, |
| {ir.ORSH, types.TUINT16, types.TUINT16}: ssa.OpRsh16Ux16, |
| {ir.ORSH, types.TINT16, types.TUINT32}: ssa.OpRsh16x32, |
| {ir.ORSH, types.TUINT16, types.TUINT32}: ssa.OpRsh16Ux32, |
| {ir.ORSH, types.TINT16, types.TUINT64}: ssa.OpRsh16x64, |
| {ir.ORSH, types.TUINT16, types.TUINT64}: ssa.OpRsh16Ux64, |
| |
| {ir.ORSH, types.TINT32, types.TUINT8}: ssa.OpRsh32x8, |
| {ir.ORSH, types.TUINT32, types.TUINT8}: ssa.OpRsh32Ux8, |
| {ir.ORSH, types.TINT32, types.TUINT16}: ssa.OpRsh32x16, |
| {ir.ORSH, types.TUINT32, types.TUINT16}: ssa.OpRsh32Ux16, |
| {ir.ORSH, types.TINT32, types.TUINT32}: ssa.OpRsh32x32, |
| {ir.ORSH, types.TUINT32, types.TUINT32}: ssa.OpRsh32Ux32, |
| {ir.ORSH, types.TINT32, types.TUINT64}: ssa.OpRsh32x64, |
| {ir.ORSH, types.TUINT32, types.TUINT64}: ssa.OpRsh32Ux64, |
| |
| {ir.ORSH, types.TINT64, types.TUINT8}: ssa.OpRsh64x8, |
| {ir.ORSH, types.TUINT64, types.TUINT8}: ssa.OpRsh64Ux8, |
| {ir.ORSH, types.TINT64, types.TUINT16}: ssa.OpRsh64x16, |
| {ir.ORSH, types.TUINT64, types.TUINT16}: ssa.OpRsh64Ux16, |
| {ir.ORSH, types.TINT64, types.TUINT32}: ssa.OpRsh64x32, |
| {ir.ORSH, types.TUINT64, types.TUINT32}: ssa.OpRsh64Ux32, |
| {ir.ORSH, types.TINT64, types.TUINT64}: ssa.OpRsh64x64, |
| {ir.ORSH, types.TUINT64, types.TUINT64}: ssa.OpRsh64Ux64, |
| } |
| |
| func (s *state) ssaShiftOp(op ir.Op, t *types.Type, u *types.Type) ssa.Op { |
| etype1 := s.concreteEtype(t) |
| etype2 := s.concreteEtype(u) |
| x, ok := shiftOpToSSA[opAndTwoTypes{op, etype1, etype2}] |
| if !ok { |
| s.Fatalf("unhandled shift op %v etype=%s/%s", op, etype1, etype2) |
| } |
| return x |
| } |
| |
| func (s *state) uintptrConstant(v uint64) *ssa.Value { |
| if s.config.PtrSize == 4 { |
| return s.newValue0I(ssa.OpConst32, types.Types[types.TUINTPTR], int64(v)) |
| } |
| return s.newValue0I(ssa.OpConst64, types.Types[types.TUINTPTR], int64(v)) |
| } |
| |
| func (s *state) conv(n ir.Node, v *ssa.Value, ft, tt *types.Type) *ssa.Value { |
| if ft.IsBoolean() && tt.IsKind(types.TUINT8) { |
| // Bool -> uint8 is generated internally when indexing into runtime.staticbyte. |
| return s.newValue1(ssa.OpCvtBoolToUint8, tt, v) |
| } |
| if ft.IsInteger() && tt.IsInteger() { |
| var op ssa.Op |
| if tt.Size() == ft.Size() { |
| op = ssa.OpCopy |
| } else if tt.Size() < ft.Size() { |
| // truncation |
| switch 10*ft.Size() + tt.Size() { |
| case 21: |
| op = ssa.OpTrunc16to8 |
| case 41: |
| op = ssa.OpTrunc32to8 |
| case 42: |
| op = ssa.OpTrunc32to16 |
| case 81: |
| op = ssa.OpTrunc64to8 |
| case 82: |
| op = ssa.OpTrunc64to16 |
| case 84: |
| op = ssa.OpTrunc64to32 |
| default: |
| s.Fatalf("weird integer truncation %v -> %v", ft, tt) |
| } |
| } else if ft.IsSigned() { |
| // sign extension |
| switch 10*ft.Size() + tt.Size() { |
| case 12: |
| op = ssa.OpSignExt8to16 |
| case 14: |
| op = ssa.OpSignExt8to32 |
| case 18: |
| op = ssa.OpSignExt8to64 |
| case 24: |
| op = ssa.OpSignExt16to32 |
| case 28: |
| op = ssa.OpSignExt16to64 |
| case 48: |
| op = ssa.OpSignExt32to64 |
| default: |
| s.Fatalf("bad integer sign extension %v -> %v", ft, tt) |
| } |
| } else { |
| // zero extension |
| switch 10*ft.Size() + tt.Size() { |
| case 12: |
| op = ssa.OpZeroExt8to16 |
| case 14: |
| op = ssa.OpZeroExt8to32 |
| case 18: |
| op = ssa.OpZeroExt8to64 |
| case 24: |
| op = ssa.OpZeroExt16to32 |
| case 28: |
| op = ssa.OpZeroExt16to64 |
| case 48: |
| op = ssa.OpZeroExt32to64 |
| default: |
| s.Fatalf("weird integer sign extension %v -> %v", ft, tt) |
| } |
| } |
| return s.newValue1(op, tt, v) |
| } |
| |
| if ft.IsComplex() && tt.IsComplex() { |
| var op ssa.Op |
| if ft.Size() == tt.Size() { |
| switch ft.Size() { |
| case 8: |
| op = ssa.OpRound32F |
| case 16: |
| op = ssa.OpRound64F |
| default: |
| s.Fatalf("weird complex conversion %v -> %v", ft, tt) |
| } |
| } else if ft.Size() == 8 && tt.Size() == 16 { |
| op = ssa.OpCvt32Fto64F |
| } else if ft.Size() == 16 && tt.Size() == 8 { |
| op = ssa.OpCvt64Fto32F |
| } else { |
| s.Fatalf("weird complex conversion %v -> %v", ft, tt) |
| } |
| ftp := types.FloatForComplex(ft) |
| ttp := types.FloatForComplex(tt) |
| return s.newValue2(ssa.OpComplexMake, tt, |
| s.newValueOrSfCall1(op, ttp, s.newValue1(ssa.OpComplexReal, ftp, v)), |
| s.newValueOrSfCall1(op, ttp, s.newValue1(ssa.OpComplexImag, ftp, v))) |
| } |
| |
| if tt.IsComplex() { // and ft is not complex |
| // Needed for generics support - can't happen in normal Go code. |
| et := types.FloatForComplex(tt) |
| v = s.conv(n, v, ft, et) |
| return s.newValue2(ssa.OpComplexMake, tt, v, s.zeroVal(et)) |
| } |
| |
| if ft.IsFloat() || tt.IsFloat() { |
| conv, ok := fpConvOpToSSA[twoTypes{s.concreteEtype(ft), s.concreteEtype(tt)}] |
| if s.config.RegSize == 4 && Arch.LinkArch.Family != sys.MIPS && !s.softFloat { |
| if conv1, ok1 := fpConvOpToSSA32[twoTypes{s.concreteEtype(ft), s.concreteEtype(tt)}]; ok1 { |
| conv = conv1 |
| } |
| } |
| if Arch.LinkArch.Family == sys.ARM64 || Arch.LinkArch.Family == sys.Wasm || Arch.LinkArch.Family == sys.S390X || s.softFloat { |
| if conv1, ok1 := uint64fpConvOpToSSA[twoTypes{s.concreteEtype(ft), s.concreteEtype(tt)}]; ok1 { |
| conv = conv1 |
| } |
| } |
| |
| if Arch.LinkArch.Family == sys.MIPS && !s.softFloat { |
| if ft.Size() == 4 && ft.IsInteger() && !ft.IsSigned() { |
| // tt is float32 or float64, and ft is also unsigned |
| if tt.Size() == 4 { |
| return s.uint32Tofloat32(n, v, ft, tt) |
| } |
| if tt.Size() == 8 { |
| return s.uint32Tofloat64(n, v, ft, tt) |
| } |
| } else if tt.Size() == 4 && tt.IsInteger() && !tt.IsSigned() { |
| // ft is float32 or float64, and tt is unsigned integer |
| if ft.Size() == 4 { |
| return s.float32ToUint32(n, v, ft, tt) |
| } |
| if ft.Size() == 8 { |
| return s.float64ToUint32(n, v, ft, tt) |
| } |
| } |
| } |
| |
| if !ok { |
| s.Fatalf("weird float conversion %v -> %v", ft, tt) |
| } |
| op1, op2, it := conv.op1, conv.op2, conv.intermediateType |
| |
| if op1 != ssa.OpInvalid && op2 != ssa.OpInvalid { |
| // normal case, not tripping over unsigned 64 |
| if op1 == ssa.OpCopy { |
| if op2 == ssa.OpCopy { |
| return v |
| } |
| return s.newValueOrSfCall1(op2, tt, v) |
| } |
| if op2 == ssa.OpCopy { |
| return s.newValueOrSfCall1(op1, tt, v) |
| } |
| return s.newValueOrSfCall1(op2, tt, s.newValueOrSfCall1(op1, types.Types[it], v)) |
| } |
| // Tricky 64-bit unsigned cases. |
| if ft.IsInteger() { |
| // tt is float32 or float64, and ft is also unsigned |
| if tt.Size() == 4 { |
| return s.uint64Tofloat32(n, v, ft, tt) |
| } |
| if tt.Size() == 8 { |
| return s.uint64Tofloat64(n, v, ft, tt) |
| } |
| s.Fatalf("weird unsigned integer to float conversion %v -> %v", ft, tt) |
| } |
| // ft is float32 or float64, and tt is unsigned integer |
| if ft.Size() == 4 { |
| return s.float32ToUint64(n, v, ft, tt) |
| } |
| if ft.Size() == 8 { |
| return s.float64ToUint64(n, v, ft, tt) |
| } |
| s.Fatalf("weird float to unsigned integer conversion %v -> %v", ft, tt) |
| return nil |
| } |
| |
| s.Fatalf("unhandled OCONV %s -> %s", ft.Kind(), tt.Kind()) |
| return nil |
| } |
| |
| // expr converts the expression n to ssa, adds it to s and returns the ssa result. |
| func (s *state) expr(n ir.Node) *ssa.Value { |
| return s.exprCheckPtr(n, true) |
| } |
| |
| func (s *state) exprCheckPtr(n ir.Node, checkPtrOK bool) *ssa.Value { |
| if ir.HasUniquePos(n) { |
| // ONAMEs and named OLITERALs have the line number |
| // of the decl, not the use. See issue 14742. |
| s.pushLine(n.Pos()) |
| defer s.popLine() |
| } |
| |
| s.stmtList(n.Init()) |
| switch n.Op() { |
| case ir.OBYTES2STRTMP: |
| n := n.(*ir.ConvExpr) |
| slice := s.expr(n.X) |
| ptr := s.newValue1(ssa.OpSlicePtr, s.f.Config.Types.BytePtr, slice) |
| len := s.newValue1(ssa.OpSliceLen, types.Types[types.TINT], slice) |
| return s.newValue2(ssa.OpStringMake, n.Type(), ptr, len) |
| case ir.OSTR2BYTESTMP: |
| n := n.(*ir.ConvExpr) |
| str := s.expr(n.X) |
| ptr := s.newValue1(ssa.OpStringPtr, s.f.Config.Types.BytePtr, str) |
| len := s.newValue1(ssa.OpStringLen, types.Types[types.TINT], str) |
| return s.newValue3(ssa.OpSliceMake, n.Type(), ptr, len, len) |
| case ir.OCFUNC: |
| n := n.(*ir.UnaryExpr) |
| aux := n.X.(*ir.Name).Linksym() |
| // OCFUNC is used to build function values, which must |
| // always reference ABIInternal entry points. |
| if aux.ABI() != obj.ABIInternal { |
| s.Fatalf("expected ABIInternal: %v", aux.ABI()) |
| } |
| return s.entryNewValue1A(ssa.OpAddr, n.Type(), aux, s.sb) |
| case ir.ONAME: |
| n := n.(*ir.Name) |
| if n.Class == ir.PFUNC { |
| // "value" of a function is the address of the function's closure |
| sym := staticdata.FuncLinksym(n) |
| return s.entryNewValue1A(ssa.OpAddr, types.NewPtr(n.Type()), sym, s.sb) |
| } |
| if s.canSSA(n) { |
| return s.variable(n, n.Type()) |
| } |
| return s.load(n.Type(), s.addr(n)) |
| case ir.OLINKSYMOFFSET: |
| n := n.(*ir.LinksymOffsetExpr) |
| return s.load(n.Type(), s.addr(n)) |
| case ir.ONIL: |
| n := n.(*ir.NilExpr) |
| t := n.Type() |
| switch { |
| case t.IsSlice(): |
| return s.constSlice(t) |
| case t.IsInterface(): |
| return s.constInterface(t) |
| default: |
| return s.constNil(t) |
| } |
| case ir.OLITERAL: |
| switch u := n.Val(); u.Kind() { |
| case constant.Int: |
| i := ir.IntVal(n.Type(), u) |
| switch n.Type().Size() { |
| case 1: |
| return s.constInt8(n.Type(), int8(i)) |
| case 2: |
| return s.constInt16(n.Type(), int16(i)) |
| case 4: |
| return s.constInt32(n.Type(), int32(i)) |
| case 8: |
| return s.constInt64(n.Type(), i) |
| default: |
| s.Fatalf("bad integer size %d", n.Type().Size()) |
| return nil |
| } |
| case constant.String: |
| i := constant.StringVal(u) |
| if i == "" { |
| return s.constEmptyString(n.Type()) |
| } |
| return s.entryNewValue0A(ssa.OpConstString, n.Type(), ssa.StringToAux(i)) |
| case constant.Bool: |
| return s.constBool(constant.BoolVal(u)) |
| case constant.Float: |
| f, _ := constant.Float64Val(u) |
| switch n.Type().Size() { |
| case 4: |
| return s.constFloat32(n.Type(), f) |
| case 8: |
| return s.constFloat64(n.Type(), f) |
| default: |
| s.Fatalf("bad float size %d", n.Type().Size()) |
| return nil |
| } |
| case constant.Complex: |
| re, _ := constant.Float64Val(constant.Real(u)) |
| im, _ := constant.Float64Val(constant.Imag(u)) |
| switch n.Type().Size() { |
| case 8: |
| pt := types.Types[types.TFLOAT32] |
| return s.newValue2(ssa.OpComplexMake, n.Type(), |
| s.constFloat32(pt, re), |
| s.constFloat32(pt, im)) |
| case 16: |
| pt := types.Types[types.TFLOAT64] |
| return s.newValue2(ssa.OpComplexMake, n.Type(), |
| s.constFloat64(pt, re), |
| s.constFloat64(pt, im)) |
| default: |
| s.Fatalf("bad complex size %d", n.Type().Size()) |
| return nil |
| } |
| default: |
| s.Fatalf("unhandled OLITERAL %v", u.Kind()) |
| return nil |
| } |
| case ir.OCONVNOP: |
| n := n.(*ir.ConvExpr) |
| to := n.Type() |
| from := n.X.Type() |
| |
| // Assume everything will work out, so set up our return value. |
| // Anything interesting that happens from here is a fatal. |
| x := s.expr(n.X) |
| if to == from { |
| return x |
| } |
| |
| // Special case for not confusing GC and liveness. |
| // We don't want pointers accidentally classified |
| // as not-pointers or vice-versa because of copy |
| // elision. |
| if to.IsPtrShaped() != from.IsPtrShaped() { |
| return s.newValue2(ssa.OpConvert, to, x, s.mem()) |
| } |
| |
| v := s.newValue1(ssa.OpCopy, to, x) // ensure that v has the right type |
| |
| // CONVNOP closure |
| if to.Kind() == types.TFUNC && from.IsPtrShaped() { |
| return v |
| } |
| |
| // named <--> unnamed type or typed <--> untyped const |
| if from.Kind() == to.Kind() { |
| return v |
| } |
| |
| // unsafe.Pointer <--> *T |
| if to.IsUnsafePtr() && from.IsPtrShaped() || from.IsUnsafePtr() && to.IsPtrShaped() { |
| if s.checkPtrEnabled && checkPtrOK && to.IsPtr() && from.IsUnsafePtr() { |
| s.checkPtrAlignment(n, v, nil) |
| } |
| return v |
| } |
| |
| // map <--> *hmap |
| if to.Kind() == types.TMAP && from.IsPtr() && |
| to.MapType().Hmap == from.Elem() { |
| return v |
| } |
| |
| types.CalcSize(from) |
| types.CalcSize(to) |
| if from.Size() != to.Size() { |
| s.Fatalf("CONVNOP width mismatch %v (%d) -> %v (%d)\n", from, from.Size(), to, to.Size()) |
| return nil |
| } |
| if etypesign(from.Kind()) != etypesign(to.Kind()) { |
| s.Fatalf("CONVNOP sign mismatch %v (%s) -> %v (%s)\n", from, from.Kind(), to, to.Kind()) |
| return nil |
| } |
| |
| if base.Flag.Cfg.Instrumenting { |
| // These appear to be fine, but they fail the |
| // integer constraint below, so okay them here. |
| // Sample non-integer conversion: map[string]string -> *uint8 |
| return v |
| } |
| |
| if etypesign(from.Kind()) == 0 { |
| s.Fatalf("CONVNOP unrecognized non-integer %v -> %v\n", from, to) |
| return nil |
| } |
| |
| // integer, same width, same sign |
| return v |
| |
| case ir.OCONV: |
| n := n.(*ir.ConvExpr) |
| x := s.expr(n.X) |
| return s.conv(n, x, n.X.Type(), n.Type()) |
| |
| case ir.ODOTTYPE: |
| n := n.(*ir.TypeAssertExpr) |
| res, _ := s.dottype(n, false) |
| return res |
| |
| case ir.ODYNAMICDOTTYPE: |
| n := n.(*ir.DynamicTypeAssertExpr) |
| res, _ := s.dynamicDottype(n, false) |
| return res |
| |
| // binary ops |
| case ir.OLT, ir.OEQ, ir.ONE, ir.OLE, ir.OGE, ir.OGT: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| if n.X.Type().IsComplex() { |
| pt := types.FloatForComplex(n.X.Type()) |
| op := s.ssaOp(ir.OEQ, pt) |
| r := s.newValueOrSfCall2(op, types.Types[types.TBOOL], s.newValue1(ssa.OpComplexReal, pt, a), s.newValue1(ssa.OpComplexReal, pt, b)) |
| i := s.newValueOrSfCall2(op, types.Types[types.TBOOL], s.newValue1(ssa.OpComplexImag, pt, a), s.newValue1(ssa.OpComplexImag, pt, b)) |
| c := s.newValue2(ssa.OpAndB, types.Types[types.TBOOL], r, i) |
| switch n.Op() { |
| case ir.OEQ: |
| return c |
| case ir.ONE: |
| return s.newValue1(ssa.OpNot, types.Types[types.TBOOL], c) |
| default: |
| s.Fatalf("ordered complex compare %v", n.Op()) |
| } |
| } |
| |
| // Convert OGE and OGT into OLE and OLT. |
| op := n.Op() |
| switch op { |
| case ir.OGE: |
| op, a, b = ir.OLE, b, a |
| case ir.OGT: |
| op, a, b = ir.OLT, b, a |
| } |
| if n.X.Type().IsFloat() { |
| // float comparison |
| return s.newValueOrSfCall2(s.ssaOp(op, n.X.Type()), types.Types[types.TBOOL], a, b) |
| } |
| // integer comparison |
| return s.newValue2(s.ssaOp(op, n.X.Type()), types.Types[types.TBOOL], a, b) |
| case ir.OMUL: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| if n.Type().IsComplex() { |
| mulop := ssa.OpMul64F |
| addop := ssa.OpAdd64F |
| subop := ssa.OpSub64F |
| pt := types.FloatForComplex(n.Type()) // Could be Float32 or Float64 |
| wt := types.Types[types.TFLOAT64] // Compute in Float64 to minimize cancellation error |
| |
| areal := s.newValue1(ssa.OpComplexReal, pt, a) |
| breal := s.newValue1(ssa.OpComplexReal, pt, b) |
| aimag := s.newValue1(ssa.OpComplexImag, pt, a) |
| bimag := s.newValue1(ssa.OpComplexImag, pt, b) |
| |
| if pt != wt { // Widen for calculation |
| areal = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, areal) |
| breal = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, breal) |
| aimag = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, aimag) |
| bimag = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, bimag) |
| } |
| |
| xreal := s.newValueOrSfCall2(subop, wt, s.newValueOrSfCall2(mulop, wt, areal, breal), s.newValueOrSfCall2(mulop, wt, aimag, bimag)) |
| ximag := s.newValueOrSfCall2(addop, wt, s.newValueOrSfCall2(mulop, wt, areal, bimag), s.newValueOrSfCall2(mulop, wt, aimag, breal)) |
| |
| if pt != wt { // Narrow to store back |
| xreal = s.newValueOrSfCall1(ssa.OpCvt64Fto32F, pt, xreal) |
| ximag = s.newValueOrSfCall1(ssa.OpCvt64Fto32F, pt, ximag) |
| } |
| |
| return s.newValue2(ssa.OpComplexMake, n.Type(), xreal, ximag) |
| } |
| |
| if n.Type().IsFloat() { |
| return s.newValueOrSfCall2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| } |
| |
| return s.newValue2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| |
| case ir.ODIV: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| if n.Type().IsComplex() { |
| // TODO this is not executed because the front-end substitutes a runtime call. |
| // That probably ought to change; with modest optimization the widen/narrow |
| // conversions could all be elided in larger expression trees. |
| mulop := ssa.OpMul64F |
| addop := ssa.OpAdd64F |
| subop := ssa.OpSub64F |
| divop := ssa.OpDiv64F |
| pt := types.FloatForComplex(n.Type()) // Could be Float32 or Float64 |
| wt := types.Types[types.TFLOAT64] // Compute in Float64 to minimize cancellation error |
| |
| areal := s.newValue1(ssa.OpComplexReal, pt, a) |
| breal := s.newValue1(ssa.OpComplexReal, pt, b) |
| aimag := s.newValue1(ssa.OpComplexImag, pt, a) |
| bimag := s.newValue1(ssa.OpComplexImag, pt, b) |
| |
| if pt != wt { // Widen for calculation |
| areal = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, areal) |
| breal = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, breal) |
| aimag = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, aimag) |
| bimag = s.newValueOrSfCall1(ssa.OpCvt32Fto64F, wt, bimag) |
| } |
| |
| denom := s.newValueOrSfCall2(addop, wt, s.newValueOrSfCall2(mulop, wt, breal, breal), s.newValueOrSfCall2(mulop, wt, bimag, bimag)) |
| xreal := s.newValueOrSfCall2(addop, wt, s.newValueOrSfCall2(mulop, wt, areal, breal), s.newValueOrSfCall2(mulop, wt, aimag, bimag)) |
| ximag := s.newValueOrSfCall2(subop, wt, s.newValueOrSfCall2(mulop, wt, aimag, breal), s.newValueOrSfCall2(mulop, wt, areal, bimag)) |
| |
| // TODO not sure if this is best done in wide precision or narrow |
| // Double-rounding might be an issue. |
| // Note that the pre-SSA implementation does the entire calculation |
| // in wide format, so wide is compatible. |
| xreal = s.newValueOrSfCall2(divop, wt, xreal, denom) |
| ximag = s.newValueOrSfCall2(divop, wt, ximag, denom) |
| |
| if pt != wt { // Narrow to store back |
| xreal = s.newValueOrSfCall1(ssa.OpCvt64Fto32F, pt, xreal) |
| ximag = s.newValueOrSfCall1(ssa.OpCvt64Fto32F, pt, ximag) |
| } |
| return s.newValue2(ssa.OpComplexMake, n.Type(), xreal, ximag) |
| } |
| if n.Type().IsFloat() { |
| return s.newValueOrSfCall2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| } |
| return s.intDivide(n, a, b) |
| case ir.OMOD: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| return s.intDivide(n, a, b) |
| case ir.OADD, ir.OSUB: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| if n.Type().IsComplex() { |
| pt := types.FloatForComplex(n.Type()) |
| op := s.ssaOp(n.Op(), pt) |
| return s.newValue2(ssa.OpComplexMake, n.Type(), |
| s.newValueOrSfCall2(op, pt, s.newValue1(ssa.OpComplexReal, pt, a), s.newValue1(ssa.OpComplexReal, pt, b)), |
| s.newValueOrSfCall2(op, pt, s.newValue1(ssa.OpComplexImag, pt, a), s.newValue1(ssa.OpComplexImag, pt, b))) |
| } |
| if n.Type().IsFloat() { |
| return s.newValueOrSfCall2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| } |
| return s.newValue2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| case ir.OAND, ir.OOR, ir.OXOR: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| return s.newValue2(s.ssaOp(n.Op(), n.Type()), a.Type, a, b) |
| case ir.OANDNOT: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| b = s.newValue1(s.ssaOp(ir.OBITNOT, b.Type), b.Type, b) |
| return s.newValue2(s.ssaOp(ir.OAND, n.Type()), a.Type, a, b) |
| case ir.OLSH, ir.ORSH: |
| n := n.(*ir.BinaryExpr) |
| a := s.expr(n.X) |
| b := s.expr(n.Y) |
| bt := b.Type |
| if bt.IsSigned() { |
| cmp := s.newValue2(s.ssaOp(ir.OLE, bt), types.Types[types.TBOOL], s.zeroVal(bt), b) |
| s.check(cmp, ir.Syms.Panicshift) |
| bt = bt.ToUnsigned() |
| } |
| return s.newValue2(s.ssaShiftOp(n.Op(), n.Type(), bt), a.Type, a, b) |
| case ir.OANDAND, ir.OOROR: |
| // To implement OANDAND (and OOROR), we introduce a |
| // new temporary variable to hold the result. The |
| // variable is associated with the OANDAND node in the |
| // s.vars table (normally variables are only |
| // associated with ONAME nodes). We convert |
| // A && B |
| // to |
| // var = A |
| // if var { |
| // var = B |
| // } |
| // Using var in the subsequent block introduces the |
| // necessary phi variable. |
| n := n.(*ir.LogicalExpr) |
| el := s.expr(n.X) |
| s.vars[n] = el |
| |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.SetControl(el) |
| // In theory, we should set b.Likely here based on context. |
| // However, gc only gives us likeliness hints |
| // in a single place, for plain OIF statements, |
| // and passing around context is finnicky, so don't bother for now. |
| |
| bRight := s.f.NewBlock(ssa.BlockPlain) |
| bResult := s.f.NewBlock(ssa.BlockPlain) |
| if n.Op() == ir.OANDAND { |
| b.AddEdgeTo(bRight) |
| b.AddEdgeTo(bResult) |
| } else if n.Op() == ir.OOROR { |
| b.AddEdgeTo(bResult) |
| b.AddEdgeTo(bRight) |
| } |
| |
| s.startBlock(bRight) |
| er := s.expr(n.Y) |
| s.vars[n] = er |
| |
| b = s.endBlock() |
| b.AddEdgeTo(bResult) |
| |
| s.startBlock(bResult) |
| return s.variable(n, types.Types[types.TBOOL]) |
| case ir.OCOMPLEX: |
| n := n.(*ir.BinaryExpr) |
| r := s.expr(n.X) |
| i := s.expr(n.Y) |
| return s.newValue2(ssa.OpComplexMake, n.Type(), r, i) |
| |
| // unary ops |
| case ir.ONEG: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| if n.Type().IsComplex() { |
| tp := types.FloatForComplex(n.Type()) |
| negop := s.ssaOp(n.Op(), tp) |
| return s.newValue2(ssa.OpComplexMake, n.Type(), |
| s.newValue1(negop, tp, s.newValue1(ssa.OpComplexReal, tp, a)), |
| s.newValue1(negop, tp, s.newValue1(ssa.OpComplexImag, tp, a))) |
| } |
| return s.newValue1(s.ssaOp(n.Op(), n.Type()), a.Type, a) |
| case ir.ONOT, ir.OBITNOT: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| return s.newValue1(s.ssaOp(n.Op(), n.Type()), a.Type, a) |
| case ir.OIMAG, ir.OREAL: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| return s.newValue1(s.ssaOp(n.Op(), n.X.Type()), n.Type(), a) |
| case ir.OPLUS: |
| n := n.(*ir.UnaryExpr) |
| return s.expr(n.X) |
| |
| case ir.OADDR: |
| n := n.(*ir.AddrExpr) |
| return s.addr(n.X) |
| |
| case ir.ORESULT: |
| n := n.(*ir.ResultExpr) |
| if s.prevCall == nil || s.prevCall.Op != ssa.OpStaticLECall && s.prevCall.Op != ssa.OpInterLECall && s.prevCall.Op != ssa.OpClosureLECall { |
| panic("Expected to see a previous call") |
| } |
| which := n.Index |
| if which == -1 { |
| panic(fmt.Errorf("ORESULT %v does not match call %s", n, s.prevCall)) |
| } |
| return s.resultOfCall(s.prevCall, which, n.Type()) |
| |
| case ir.ODEREF: |
| n := n.(*ir.StarExpr) |
| p := s.exprPtr(n.X, n.Bounded(), n.Pos()) |
| return s.load(n.Type(), p) |
| |
| case ir.ODOT: |
| n := n.(*ir.SelectorExpr) |
| if n.X.Op() == ir.OSTRUCTLIT { |
| // All literals with nonzero fields have already been |
| // rewritten during walk. Any that remain are just T{} |
| // or equivalents. Use the zero value. |
| if !ir.IsZero(n.X) { |
| s.Fatalf("literal with nonzero value in SSA: %v", n.X) |
| } |
| return s.zeroVal(n.Type()) |
| } |
| // If n is addressable and can't be represented in |
| // SSA, then load just the selected field. This |
| // prevents false memory dependencies in race/msan/asan |
| // instrumentation. |
| if ir.IsAddressable(n) && !s.canSSA(n) { |
| p := s.addr(n) |
| return s.load(n.Type(), p) |
| } |
| v := s.expr(n.X) |
| return s.newValue1I(ssa.OpStructSelect, n.Type(), int64(fieldIdx(n)), v) |
| |
| case ir.ODOTPTR: |
| n := n.(*ir.SelectorExpr) |
| p := s.exprPtr(n.X, n.Bounded(), n.Pos()) |
| p = s.newValue1I(ssa.OpOffPtr, types.NewPtr(n.Type()), n.Offset(), p) |
| return s.load(n.Type(), p) |
| |
| case ir.OINDEX: |
| n := n.(*ir.IndexExpr) |
| switch { |
| case n.X.Type().IsString(): |
| if n.Bounded() && ir.IsConst(n.X, constant.String) && ir.IsConst(n.Index, constant.Int) { |
| // Replace "abc"[1] with 'b'. |
| // Delayed until now because "abc"[1] is not an ideal constant. |
| // See test/fixedbugs/issue11370.go. |
| return s.newValue0I(ssa.OpConst8, types.Types[types.TUINT8], int64(int8(ir.StringVal(n.X)[ir.Int64Val(n.Index)]))) |
| } |
| a := s.expr(n.X) |
| i := s.expr(n.Index) |
| len := s.newValue1(ssa.OpStringLen, types.Types[types.TINT], a) |
| i = s.boundsCheck(i, len, ssa.BoundsIndex, n.Bounded()) |
| ptrtyp := s.f.Config.Types.BytePtr |
| ptr := s.newValue1(ssa.OpStringPtr, ptrtyp, a) |
| if ir.IsConst(n.Index, constant.Int) { |
| ptr = s.newValue1I(ssa.OpOffPtr, ptrtyp, ir.Int64Val(n.Index), ptr) |
| } else { |
| ptr = s.newValue2(ssa.OpAddPtr, ptrtyp, ptr, i) |
| } |
| return s.load(types.Types[types.TUINT8], ptr) |
| case n.X.Type().IsSlice(): |
| p := s.addr(n) |
| return s.load(n.X.Type().Elem(), p) |
| case n.X.Type().IsArray(): |
| if TypeOK(n.X.Type()) { |
| // SSA can handle arrays of length at most 1. |
| bound := n.X.Type().NumElem() |
| a := s.expr(n.X) |
| i := s.expr(n.Index) |
| if bound == 0 { |
| // Bounds check will never succeed. Might as well |
| // use constants for the bounds check. |
| z := s.constInt(types.Types[types.TINT], 0) |
| s.boundsCheck(z, z, ssa.BoundsIndex, false) |
| // The return value won't be live, return junk. |
| // But not quite junk, in case bounds checks are turned off. See issue 48092. |
| return s.zeroVal(n.Type()) |
| } |
| len := s.constInt(types.Types[types.TINT], bound) |
| s.boundsCheck(i, len, ssa.BoundsIndex, n.Bounded()) // checks i == 0 |
| return s.newValue1I(ssa.OpArraySelect, n.Type(), 0, a) |
| } |
| p := s.addr(n) |
| return s.load(n.X.Type().Elem(), p) |
| default: |
| s.Fatalf("bad type for index %v", n.X.Type()) |
| return nil |
| } |
| |
| case ir.OLEN, ir.OCAP: |
| n := n.(*ir.UnaryExpr) |
| switch { |
| case n.X.Type().IsSlice(): |
| op := ssa.OpSliceLen |
| if n.Op() == ir.OCAP { |
| op = ssa.OpSliceCap |
| } |
| return s.newValue1(op, types.Types[types.TINT], s.expr(n.X)) |
| case n.X.Type().IsString(): // string; not reachable for OCAP |
| return s.newValue1(ssa.OpStringLen, types.Types[types.TINT], s.expr(n.X)) |
| case n.X.Type().IsMap(), n.X.Type().IsChan(): |
| return s.referenceTypeBuiltin(n, s.expr(n.X)) |
| default: // array |
| return s.constInt(types.Types[types.TINT], n.X.Type().NumElem()) |
| } |
| |
| case ir.OSPTR: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| if n.X.Type().IsSlice() { |
| if n.Bounded() { |
| return s.newValue1(ssa.OpSlicePtr, n.Type(), a) |
| } |
| return s.newValue1(ssa.OpSlicePtrUnchecked, n.Type(), a) |
| } else { |
| return s.newValue1(ssa.OpStringPtr, n.Type(), a) |
| } |
| |
| case ir.OITAB: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| return s.newValue1(ssa.OpITab, n.Type(), a) |
| |
| case ir.OIDATA: |
| n := n.(*ir.UnaryExpr) |
| a := s.expr(n.X) |
| return s.newValue1(ssa.OpIData, n.Type(), a) |
| |
| case ir.OEFACE: |
| n := n.(*ir.BinaryExpr) |
| tab := s.expr(n.X) |
| data := s.expr(n.Y) |
| return s.newValue2(ssa.OpIMake, n.Type(), tab, data) |
| |
| case ir.OSLICEHEADER: |
| n := n.(*ir.SliceHeaderExpr) |
| p := s.expr(n.Ptr) |
| l := s.expr(n.Len) |
| c := s.expr(n.Cap) |
| return s.newValue3(ssa.OpSliceMake, n.Type(), p, l, c) |
| |
| case ir.OSTRINGHEADER: |
| n := n.(*ir.StringHeaderExpr) |
| p := s.expr(n.Ptr) |
| l := s.expr(n.Len) |
| return s.newValue2(ssa.OpStringMake, n.Type(), p, l) |
| |
| case ir.OSLICE, ir.OSLICEARR, ir.OSLICE3, ir.OSLICE3ARR: |
| n := n.(*ir.SliceExpr) |
| check := s.checkPtrEnabled && n.Op() == ir.OSLICE3ARR && n.X.Op() == ir.OCONVNOP && n.X.(*ir.ConvExpr).X.Type().IsUnsafePtr() |
| v := s.exprCheckPtr(n.X, !check) |
| var i, j, k *ssa.Value |
| if n.Low != nil { |
| i = s.expr(n.Low) |
| } |
| if n.High != nil { |
| j = s.expr(n.High) |
| } |
| if n.Max != nil { |
| k = s.expr(n.Max) |
| } |
| p, l, c := s.slice(v, i, j, k, n.Bounded()) |
| if check { |
| // Emit checkptr instrumentation after bound check to prevent false positive, see #46938. |
| s.checkPtrAlignment(n.X.(*ir.ConvExpr), v, s.conv(n.Max, k, k.Type, types.Types[types.TUINTPTR])) |
| } |
| return s.newValue3(ssa.OpSliceMake, n.Type(), p, l, c) |
| |
| case ir.OSLICESTR: |
| n := n.(*ir.SliceExpr) |
| v := s.expr(n.X) |
| var i, j *ssa.Value |
| if n.Low != nil { |
| i = s.expr(n.Low) |
| } |
| if n.High != nil { |
| j = s.expr(n.High) |
| } |
| p, l, _ := s.slice(v, i, j, nil, n.Bounded()) |
| return s.newValue2(ssa.OpStringMake, n.Type(), p, l) |
| |
| case ir.OSLICE2ARRPTR: |
| // if arrlen > slice.len { |
| // panic(...) |
| // } |
| // slice.ptr |
| n := n.(*ir.ConvExpr) |
| v := s.expr(n.X) |
| nelem := n.Type().Elem().NumElem() |
| arrlen := s.constInt(types.Types[types.TINT], nelem) |
| cap := s.newValue1(ssa.OpSliceLen, types.Types[types.TINT], v) |
| s.boundsCheck(arrlen, cap, ssa.BoundsConvert, false) |
| op := ssa.OpSlicePtr |
| if nelem == 0 { |
| op = ssa.OpSlicePtrUnchecked |
| } |
| return s.newValue1(op, n.Type(), v) |
| |
| case ir.OCALLFUNC: |
| n := n.(*ir.CallExpr) |
| if ir.IsIntrinsicCall(n) { |
| return s.intrinsicCall(n) |
| } |
| fallthrough |
| |
| case ir.OCALLINTER: |
| n := n.(*ir.CallExpr) |
| return s.callResult(n, callNormal) |
| |
| case ir.OGETG: |
| n := n.(*ir.CallExpr) |
| return s.newValue1(ssa.OpGetG, n.Type(), s.mem()) |
| |
| case ir.OGETCALLERPC: |
| n := n.(*ir.CallExpr) |
| return s.newValue0(ssa.OpGetCallerPC, n.Type()) |
| |
| case ir.OGETCALLERSP: |
| n := n.(*ir.CallExpr) |
| return s.newValue1(ssa.OpGetCallerSP, n.Type(), s.mem()) |
| |
| case ir.OAPPEND: |
| return s.append(n.(*ir.CallExpr), false) |
| |
| case ir.OMIN, ir.OMAX: |
| return s.minMax(n.(*ir.CallExpr)) |
| |
| case ir.OSTRUCTLIT, ir.OARRAYLIT: |
| // All literals with nonzero fields have already been |
| // rewritten during walk. Any that remain are just T{} |
| // or equivalents. Use the zero value. |
| n := n.(*ir.CompLitExpr) |
| if !ir.IsZero(n) { |
| s.Fatalf("literal with nonzero value in SSA: %v", n) |
| } |
| return s.zeroVal(n.Type()) |
| |
| case ir.ONEW: |
| n := n.(*ir.UnaryExpr) |
| var rtype *ssa.Value |
| if x, ok := n.X.(*ir.DynamicType); ok && x.Op() == ir.ODYNAMICTYPE { |
| rtype = s.expr(x.RType) |
| } |
| return s.newObject(n.Type().Elem(), rtype) |
| |
| case ir.OUNSAFEADD: |
| n := n.(*ir.BinaryExpr) |
| ptr := s.expr(n.X) |
| len := s.expr(n.Y) |
| |
| // Force len to uintptr to prevent misuse of garbage bits in the |
| // upper part of the register (#48536). |
| len = s.conv(n, len, len.Type, types.Types[types.TUINTPTR]) |
| |
| return s.newValue2(ssa.OpAddPtr, n.Type(), ptr, len) |
| |
| default: |
| s.Fatalf("unhandled expr %v", n.Op()) |
| return nil |
| } |
| } |
| |
| func (s *state) resultOfCall(c *ssa.Value, which int64, t *types.Type) *ssa.Value { |
| aux := c.Aux.(*ssa.AuxCall) |
| pa := aux.ParamAssignmentForResult(which) |
| // TODO(register args) determine if in-memory TypeOK is better loaded early from SelectNAddr or later when SelectN is expanded. |
| // SelectN is better for pattern-matching and possible call-aware analysis we might want to do in the future. |
| if len(pa.Registers) == 0 && !TypeOK(t) { |
| addr := s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(t), which, c) |
| return s.rawLoad(t, addr) |
| } |
| return s.newValue1I(ssa.OpSelectN, t, which, c) |
| } |
| |
| func (s *state) resultAddrOfCall(c *ssa.Value, which int64, t *types.Type) *ssa.Value { |
| aux := c.Aux.(*ssa.AuxCall) |
| pa := aux.ParamAssignmentForResult(which) |
| if len(pa.Registers) == 0 { |
| return s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(t), which, c) |
| } |
| _, addr := s.temp(c.Pos, t) |
| rval := s.newValue1I(ssa.OpSelectN, t, which, c) |
| s.vars[memVar] = s.newValue3Apos(ssa.OpStore, types.TypeMem, t, addr, rval, s.mem(), false) |
| return addr |
| } |
| |
| // append converts an OAPPEND node to SSA. |
| // If inplace is false, it converts the OAPPEND expression n to an ssa.Value, |
| // adds it to s, and returns the Value. |
| // If inplace is true, it writes the result of the OAPPEND expression n |
| // back to the slice being appended to, and returns nil. |
| // inplace MUST be set to false if the slice can be SSA'd. |
| // Note: this code only handles fixed-count appends. Dotdotdot appends |
| // have already been rewritten at this point (by walk). |
| func (s *state) append(n *ir.CallExpr, inplace bool) *ssa.Value { |
| // If inplace is false, process as expression "append(s, e1, e2, e3)": |
| // |
| // ptr, len, cap := s |
| // len += 3 |
| // if uint(len) > uint(cap) { |
| // ptr, len, cap = growslice(ptr, len, cap, 3, typ) |
| // Note that len is unmodified by growslice. |
| // } |
| // // with write barriers, if needed: |
| // *(ptr+(len-3)) = e1 |
| // *(ptr+(len-2)) = e2 |
| // *(ptr+(len-1)) = e3 |
| // return makeslice(ptr, len, cap) |
| // |
| // |
| // If inplace is true, process as statement "s = append(s, e1, e2, e3)": |
| // |
| // a := &s |
| // ptr, len, cap := s |
| // len += 3 |
| // if uint(len) > uint(cap) { |
| // ptr, len, cap = growslice(ptr, len, cap, 3, typ) |
| // vardef(a) // if necessary, advise liveness we are writing a new a |
| // *a.cap = cap // write before ptr to avoid a spill |
| // *a.ptr = ptr // with write barrier |
| // } |
| // *a.len = len |
| // // with write barriers, if needed: |
| // *(ptr+(len-3)) = e1 |
| // *(ptr+(len-2)) = e2 |
| // *(ptr+(len-1)) = e3 |
| |
| et := n.Type().Elem() |
| pt := types.NewPtr(et) |
| |
| // Evaluate slice |
| sn := n.Args[0] // the slice node is the first in the list |
| var slice, addr *ssa.Value |
| if inplace { |
| addr = s.addr(sn) |
| slice = s.load(n.Type(), addr) |
| } else { |
| slice = s.expr(sn) |
| } |
| |
| // Allocate new blocks |
| grow := s.f.NewBlock(ssa.BlockPlain) |
| assign := s.f.NewBlock(ssa.BlockPlain) |
| |
| // Decomposse input slice. |
| p := s.newValue1(ssa.OpSlicePtr, pt, slice) |
| l := s.newValue1(ssa.OpSliceLen, types.Types[types.TINT], slice) |
| c := s.newValue1(ssa.OpSliceCap, types.Types[types.TINT], slice) |
| |
| // Add number of new elements to length. |
| nargs := s.constInt(types.Types[types.TINT], int64(len(n.Args)-1)) |
| l = s.newValue2(s.ssaOp(ir.OADD, types.Types[types.TINT]), types.Types[types.TINT], l, nargs) |
| |
| // Decide if we need to grow |
| cmp := s.newValue2(s.ssaOp(ir.OLT, types.Types[types.TUINT]), types.Types[types.TBOOL], c, l) |
| |
| // Record values of ptr/len/cap before branch. |
| s.vars[ptrVar] = p |
| s.vars[lenVar] = l |
| if !inplace { |
| s.vars[capVar] = c |
| } |
| |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.Likely = ssa.BranchUnlikely |
| b.SetControl(cmp) |
| b.AddEdgeTo(grow) |
| b.AddEdgeTo(assign) |
| |
| // Call growslice |
| s.startBlock(grow) |
| taddr := s.expr(n.X) |
| r := s.rtcall(ir.Syms.Growslice, true, []*types.Type{n.Type()}, p, l, c, nargs, taddr) |
| |
| // Decompose output slice |
| p = s.newValue1(ssa.OpSlicePtr, pt, r[0]) |
| l = s.newValue1(ssa.OpSliceLen, types.Types[types.TINT], r[0]) |
| c = s.newValue1(ssa.OpSliceCap, types.Types[types.TINT], r[0]) |
| |
| s.vars[ptrVar] = p |
| s.vars[lenVar] = l |
| s.vars[capVar] = c |
| if inplace { |
| if sn.Op() == ir.ONAME { |
| sn := sn.(*ir.Name) |
| if sn.Class != ir.PEXTERN { |
| // Tell liveness we're about to build a new slice |
| s.vars[memVar] = s.newValue1A(ssa.OpVarDef, types.TypeMem, sn, s.mem()) |
| } |
| } |
| capaddr := s.newValue1I(ssa.OpOffPtr, s.f.Config.Types.IntPtr, types.SliceCapOffset, addr) |
| s.store(types.Types[types.TINT], capaddr, c) |
| s.store(pt, addr, p) |
| } |
| |
| b = s.endBlock() |
| b.AddEdgeTo(assign) |
| |
| // assign new elements to slots |
| s.startBlock(assign) |
| p = s.variable(ptrVar, pt) // generates phi for ptr |
| l = s.variable(lenVar, types.Types[types.TINT]) // generates phi for len |
| if !inplace { |
| c = s.variable(capVar, types.Types[types.TINT]) // generates phi for cap |
| } |
| |
| if inplace { |
| // Update length in place. |
| // We have to wait until here to make sure growslice succeeded. |
| lenaddr := s.newValue1I(ssa.OpOffPtr, s.f.Config.Types.IntPtr, types.SliceLenOffset, addr) |
| s.store(types.Types[types.TINT], lenaddr, l) |
| } |
| |
| // Evaluate args |
| type argRec struct { |
| // if store is true, we're appending the value v. If false, we're appending the |
| // value at *v. |
| v *ssa.Value |
| store bool |
| } |
| args := make([]argRec, 0, len(n.Args[1:])) |
| for _, n := range n.Args[1:] { |
| if TypeOK(n.Type()) { |
| args = append(args, argRec{v: s.expr(n), store: true}) |
| } else { |
| v := s.addr(n) |
| args = append(args, argRec{v: v}) |
| } |
| } |
| |
| // Write args into slice. |
| oldLen := s.newValue2(s.ssaOp(ir.OSUB, types.Types[types.TINT]), types.Types[types.TINT], l, nargs) |
| p2 := s.newValue2(ssa.OpPtrIndex, pt, p, oldLen) |
| for i, arg := range args { |
| addr := s.newValue2(ssa.OpPtrIndex, pt, p2, s.constInt(types.Types[types.TINT], int64(i))) |
| if arg.store { |
| s.storeType(et, addr, arg.v, 0, true) |
| } else { |
| s.move(et, addr, arg.v) |
| } |
| } |
| |
| // The following deletions have no practical effect at this time |
| // because state.vars has been reset by the preceding state.startBlock. |
| // They only enforce the fact that these variables are no longer need in |
| // the current scope. |
| delete(s.vars, ptrVar) |
| delete(s.vars, lenVar) |
| if !inplace { |
| delete(s.vars, capVar) |
| } |
| |
| // make result |
| if inplace { |
| return nil |
| } |
| return s.newValue3(ssa.OpSliceMake, n.Type(), p, l, c) |
| } |
| |
| // minMax converts an OMIN/OMAX builtin call into SSA. |
| func (s *state) minMax(n *ir.CallExpr) *ssa.Value { |
| // The OMIN/OMAX builtin is variadic, but its semantics are |
| // equivalent to left-folding a binary min/max operation across the |
| // arguments list. |
| fold := func(op func(x, a *ssa.Value) *ssa.Value) *ssa.Value { |
| x := s.expr(n.Args[0]) |
| for _, arg := range n.Args[1:] { |
| x = op(x, s.expr(arg)) |
| } |
| return x |
| } |
| |
| typ := n.Type() |
| |
| if typ.IsFloat() || typ.IsString() { |
| // min/max semantics for floats are tricky because of NaNs and |
| // negative zero, so we let the runtime handle this instead. |
| // |
| // Strings are conceptually simpler, but we currently desugar |
| // string comparisons during walk, not ssagen. |
| |
| var name string |
| switch typ.Kind() { |
| case types.TFLOAT32: |
| switch n.Op() { |
| case ir.OMIN: |
| name = "fmin32" |
| case ir.OMAX: |
| name = "fmax32" |
| } |
| case types.TFLOAT64: |
| switch n.Op() { |
| case ir.OMIN: |
| name = "fmin64" |
| case ir.OMAX: |
| name = "fmax64" |
| } |
| case types.TSTRING: |
| switch n.Op() { |
| case ir.OMIN: |
| name = "strmin" |
| case ir.OMAX: |
| name = "strmax" |
| } |
| } |
| fn := typecheck.LookupRuntimeFunc(name) |
| |
| return fold(func(x, a *ssa.Value) *ssa.Value { |
| return s.rtcall(fn, true, []*types.Type{typ}, x, a)[0] |
| }) |
| } |
| |
| lt := s.ssaOp(ir.OLT, typ) |
| |
| return fold(func(x, a *ssa.Value) *ssa.Value { |
| switch n.Op() { |
| case ir.OMIN: |
| // a < x ? a : x |
| return s.ternary(s.newValue2(lt, types.Types[types.TBOOL], a, x), a, x) |
| case ir.OMAX: |
| // x < a ? a : x |
| return s.ternary(s.newValue2(lt, types.Types[types.TBOOL], x, a), a, x) |
| } |
| panic("unreachable") |
| }) |
| } |
| |
| // ternary emits code to evaluate cond ? x : y. |
| func (s *state) ternary(cond, x, y *ssa.Value) *ssa.Value { |
| // Note that we need a new ternaryVar each time (unlike okVar where we can |
| // reuse the variable) because it might have a different type every time. |
| ternaryVar := ssaMarker("ternary") |
| |
| bThen := s.f.NewBlock(ssa.BlockPlain) |
| bElse := s.f.NewBlock(ssa.BlockPlain) |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.SetControl(cond) |
| b.AddEdgeTo(bThen) |
| b.AddEdgeTo(bElse) |
| |
| s.startBlock(bThen) |
| s.vars[ternaryVar] = x |
| s.endBlock().AddEdgeTo(bEnd) |
| |
| s.startBlock(bElse) |
| s.vars[ternaryVar] = y |
| s.endBlock().AddEdgeTo(bEnd) |
| |
| s.startBlock(bEnd) |
| r := s.variable(ternaryVar, x.Type) |
| delete(s.vars, ternaryVar) |
| return r |
| } |
| |
| // condBranch evaluates the boolean expression cond and branches to yes |
| // if cond is true and no if cond is false. |
| // This function is intended to handle && and || better than just calling |
| // s.expr(cond) and branching on the result. |
| func (s *state) condBranch(cond ir.Node, yes, no *ssa.Block, likely int8) { |
| switch cond.Op() { |
| case ir.OANDAND: |
| cond := cond.(*ir.LogicalExpr) |
| mid := s.f.NewBlock(ssa.BlockPlain) |
| s.stmtList(cond.Init()) |
| s.condBranch(cond.X, mid, no, max8(likely, 0)) |
| s.startBlock(mid) |
| s.condBranch(cond.Y, yes, no, likely) |
| return |
| // Note: if likely==1, then both recursive calls pass 1. |
| // If likely==-1, then we don't have enough information to decide |
| // whether the first branch is likely or not. So we pass 0 for |
| // the likeliness of the first branch. |
| // TODO: have the frontend give us branch prediction hints for |
| // OANDAND and OOROR nodes (if it ever has such info). |
| case ir.OOROR: |
| cond := cond.(*ir.LogicalExpr) |
| mid := s.f.NewBlock(ssa.BlockPlain) |
| s.stmtList(cond.Init()) |
| s.condBranch(cond.X, yes, mid, min8(likely, 0)) |
| s.startBlock(mid) |
| s.condBranch(cond.Y, yes, no, likely) |
| return |
| // Note: if likely==-1, then both recursive calls pass -1. |
| // If likely==1, then we don't have enough info to decide |
| // the likelihood of the first branch. |
| case ir.ONOT: |
| cond := cond.(*ir.UnaryExpr) |
| s.stmtList(cond.Init()) |
| s.condBranch(cond.X, no, yes, -likely) |
| return |
| case ir.OCONVNOP: |
| cond := cond.(*ir.ConvExpr) |
| s.stmtList(cond.Init()) |
| s.condBranch(cond.X, yes, no, likely) |
| return |
| } |
| c := s.expr(cond) |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.SetControl(c) |
| b.Likely = ssa.BranchPrediction(likely) // gc and ssa both use -1/0/+1 for likeliness |
| b.AddEdgeTo(yes) |
| b.AddEdgeTo(no) |
| } |
| |
| type skipMask uint8 |
| |
| const ( |
| skipPtr skipMask = 1 << iota |
| skipLen |
| skipCap |
| ) |
| |
| // assign does left = right. |
| // Right has already been evaluated to ssa, left has not. |
| // If deref is true, then we do left = *right instead (and right has already been nil-checked). |
| // If deref is true and right == nil, just do left = 0. |
| // skip indicates assignments (at the top level) that can be avoided. |
| // mayOverlap indicates whether left&right might partially overlap in memory. Default is false. |
| func (s *state) assign(left ir.Node, right *ssa.Value, deref bool, skip skipMask) { |
| s.assignWhichMayOverlap(left, right, deref, skip, false) |
| } |
| func (s *state) assignWhichMayOverlap(left ir.Node, right *ssa.Value, deref bool, skip skipMask, mayOverlap bool) { |
| if left.Op() == ir.ONAME && ir.IsBlank(left) { |
| return |
| } |
| t := left.Type() |
| types.CalcSize(t) |
| if s.canSSA(left) { |
| if deref { |
| s.Fatalf("can SSA LHS %v but not RHS %s", left, right) |
| } |
| if left.Op() == ir.ODOT { |
| // We're assigning to a field of an ssa-able value. |
| // We need to build a new structure with the new value for the |
| // field we're assigning and the old values for the other fields. |
| // For instance: |
| // type T struct {a, b, c int} |
| // var T x |
| // x.b = 5 |
| // For the x.b = 5 assignment we want to generate x = T{x.a, 5, x.c} |
| |
| // Grab information about the structure type. |
| left := left.(*ir.SelectorExpr) |
| t := left.X.Type() |
| nf := t.NumFields() |
| idx := fieldIdx(left) |
| |
| // Grab old value of structure. |
| old := s.expr(left.X) |
| |
| // Make new structure. |
| new := s.newValue0(ssa.StructMakeOp(t.NumFields()), t) |
| |
| // Add fields as args. |
| for i := 0; i < nf; i++ { |
| if i == idx { |
| new.AddArg(right) |
| } else { |
| new.AddArg(s.newValue1I(ssa.OpStructSelect, t.FieldType(i), int64(i), old)) |
| } |
| } |
| |
| // Recursively assign the new value we've made to the base of the dot op. |
| s.assign(left.X, new, false, 0) |
| // TODO: do we need to update named values here? |
| return |
| } |
| if left.Op() == ir.OINDEX && left.(*ir.IndexExpr).X.Type().IsArray() { |
| left := left.(*ir.IndexExpr) |
| s.pushLine(left.Pos()) |
| defer s.popLine() |
| // We're assigning to an element of an ssa-able array. |
| // a[i] = v |
| t := left.X.Type() |
| n := t.NumElem() |
| |
| i := s.expr(left.Index) // index |
| if n == 0 { |
| // The bounds check must fail. Might as well |
| // ignore the actual index and just use zeros. |
| z := s.constInt(types.Types[types.TINT], 0) |
| s.boundsCheck(z, z, ssa.BoundsIndex, false) |
| return |
| } |
| if n != 1 { |
| s.Fatalf("assigning to non-1-length array") |
| } |
| // Rewrite to a = [1]{v} |
| len := s.constInt(types.Types[types.TINT], 1) |
| s.boundsCheck(i, len, ssa.BoundsIndex, false) // checks i == 0 |
| v := s.newValue1(ssa.OpArrayMake1, t, right) |
| s.assign(left.X, v, false, 0) |
| return |
| } |
| left := left.(*ir.Name) |
| // Update variable assignment. |
| s.vars[left] = right |
| s.addNamedValue(left, right) |
| return |
| } |
| |
| // If this assignment clobbers an entire local variable, then emit |
| // OpVarDef so liveness analysis knows the variable is redefined. |
| if base, ok := clobberBase(left).(*ir.Name); ok && base.OnStack() && skip == 0 && t.HasPointers() { |
| s.vars[memVar] = s.newValue1Apos(ssa.OpVarDef, types.TypeMem, base, s.mem(), !ir.IsAutoTmp(base)) |
| } |
| |
| // Left is not ssa-able. Compute its address. |
| addr := s.addr(left) |
| if ir.IsReflectHeaderDataField(left) { |
| // Package unsafe's documentation says storing pointers into |
| // reflect.SliceHeader and reflect.StringHeader's Data fields |
| // is valid, even though they have type uintptr (#19168). |
| // Mark it pointer type to signal the writebarrier pass to |
| // insert a write barrier. |
| t = types.Types[types.TUNSAFEPTR] |
| } |
| if deref { |
| // Treat as a mem->mem move. |
| if right == nil { |
| s.zero(t, addr) |
| } else { |
| s.moveWhichMayOverlap(t, addr, right, mayOverlap) |
| } |
| return |
| } |
| // Treat as a store. |
| s.storeType(t, addr, right, skip, !ir.IsAutoTmp(left)) |
| } |
| |
| // zeroVal returns the zero value for type t. |
| func (s *state) zeroVal(t *types.Type) *ssa.Value { |
| switch { |
| case t.IsInteger(): |
| switch t.Size() { |
| case 1: |
| return s.constInt8(t, 0) |
| case 2: |
| return s.constInt16(t, 0) |
| case 4: |
| return s.constInt32(t, 0) |
| case 8: |
| return s.constInt64(t, 0) |
| default: |
| s.Fatalf("bad sized integer type %v", t) |
| } |
| case t.IsFloat(): |
| switch t.Size() { |
| case 4: |
| return s.constFloat32(t, 0) |
| case 8: |
| return s.constFloat64(t, 0) |
| default: |
| s.Fatalf("bad sized float type %v", t) |
| } |
| case t.IsComplex(): |
| switch t.Size() { |
| case 8: |
| z := s.constFloat32(types.Types[types.TFLOAT32], 0) |
| return s.entryNewValue2(ssa.OpComplexMake, t, z, z) |
| case 16: |
| z := s.constFloat64(types.Types[types.TFLOAT64], 0) |
| return s.entryNewValue2(ssa.OpComplexMake, t, z, z) |
| default: |
| s.Fatalf("bad sized complex type %v", t) |
| } |
| |
| case t.IsString(): |
| return s.constEmptyString(t) |
| case t.IsPtrShaped(): |
| return s.constNil(t) |
| case t.IsBoolean(): |
| return s.constBool(false) |
| case t.IsInterface(): |
| return s.constInterface(t) |
| case t.IsSlice(): |
| return s.constSlice(t) |
| case t.IsStruct(): |
| n := t.NumFields() |
| v := s.entryNewValue0(ssa.StructMakeOp(t.NumFields()), t) |
| for i := 0; i < n; i++ { |
| v.AddArg(s.zeroVal(t.FieldType(i))) |
| } |
| return v |
| case t.IsArray(): |
| switch t.NumElem() { |
| case 0: |
| return s.entryNewValue0(ssa.OpArrayMake0, t) |
| case 1: |
| return s.entryNewValue1(ssa.OpArrayMake1, t, s.zeroVal(t.Elem())) |
| } |
| } |
| s.Fatalf("zero for type %v not implemented", t) |
| return nil |
| } |
| |
| type callKind int8 |
| |
| const ( |
| callNormal callKind = iota |
| callDefer |
| callDeferStack |
| callGo |
| callTail |
| ) |
| |
| type sfRtCallDef struct { |
| rtfn *obj.LSym |
| rtype types.Kind |
| } |
| |
| var softFloatOps map[ssa.Op]sfRtCallDef |
| |
| func softfloatInit() { |
| // Some of these operations get transformed by sfcall. |
| softFloatOps = map[ssa.Op]sfRtCallDef{ |
| ssa.OpAdd32F: {typecheck.LookupRuntimeFunc("fadd32"), types.TFLOAT32}, |
| ssa.OpAdd64F: {typecheck.LookupRuntimeFunc("fadd64"), types.TFLOAT64}, |
| ssa.OpSub32F: {typecheck.LookupRuntimeFunc("fadd32"), types.TFLOAT32}, |
| ssa.OpSub64F: {typecheck.LookupRuntimeFunc("fadd64"), types.TFLOAT64}, |
| ssa.OpMul32F: {typecheck.LookupRuntimeFunc("fmul32"), types.TFLOAT32}, |
| ssa.OpMul64F: {typecheck.LookupRuntimeFunc("fmul64"), types.TFLOAT64}, |
| ssa.OpDiv32F: {typecheck.LookupRuntimeFunc("fdiv32"), types.TFLOAT32}, |
| ssa.OpDiv64F: {typecheck.LookupRuntimeFunc("fdiv64"), types.TFLOAT64}, |
| |
| ssa.OpEq64F: {typecheck.LookupRuntimeFunc("feq64"), types.TBOOL}, |
| ssa.OpEq32F: {typecheck.LookupRuntimeFunc("feq32"), types.TBOOL}, |
| ssa.OpNeq64F: {typecheck.LookupRuntimeFunc("feq64"), types.TBOOL}, |
| ssa.OpNeq32F: {typecheck.LookupRuntimeFunc("feq32"), types.TBOOL}, |
| ssa.OpLess64F: {typecheck.LookupRuntimeFunc("fgt64"), types.TBOOL}, |
| ssa.OpLess32F: {typecheck.LookupRuntimeFunc("fgt32"), types.TBOOL}, |
| ssa.OpLeq64F: {typecheck.LookupRuntimeFunc("fge64"), types.TBOOL}, |
| ssa.OpLeq32F: {typecheck.LookupRuntimeFunc("fge32"), types.TBOOL}, |
| |
| ssa.OpCvt32to32F: {typecheck.LookupRuntimeFunc("fint32to32"), types.TFLOAT32}, |
| ssa.OpCvt32Fto32: {typecheck.LookupRuntimeFunc("f32toint32"), types.TINT32}, |
| ssa.OpCvt64to32F: {typecheck.LookupRuntimeFunc("fint64to32"), types.TFLOAT32}, |
| ssa.OpCvt32Fto64: {typecheck.LookupRuntimeFunc("f32toint64"), types.TINT64}, |
| ssa.OpCvt64Uto32F: {typecheck.LookupRuntimeFunc("fuint64to32"), types.TFLOAT32}, |
| ssa.OpCvt32Fto64U: {typecheck.LookupRuntimeFunc("f32touint64"), types.TUINT64}, |
| ssa.OpCvt32to64F: {typecheck.LookupRuntimeFunc("fint32to64"), types.TFLOAT64}, |
| ssa.OpCvt64Fto32: {typecheck.LookupRuntimeFunc("f64toint32"), types.TINT32}, |
| ssa.OpCvt64to64F: {typecheck.LookupRuntimeFunc("fint64to64"), types.TFLOAT64}, |
| ssa.OpCvt64Fto64: {typecheck.LookupRuntimeFunc("f64toint64"), types.TINT64}, |
| ssa.OpCvt64Uto64F: {typecheck.LookupRuntimeFunc("fuint64to64"), types.TFLOAT64}, |
| ssa.OpCvt64Fto64U: {typecheck.LookupRuntimeFunc("f64touint64"), types.TUINT64}, |
| ssa.OpCvt32Fto64F: {typecheck.LookupRuntimeFunc("f32to64"), types.TFLOAT64}, |
| ssa.OpCvt64Fto32F: {typecheck.LookupRuntimeFunc("f64to32"), types.TFLOAT32}, |
| } |
| } |
| |
| // TODO: do not emit sfcall if operation can be optimized to constant in later |
| // opt phase |
| func (s *state) sfcall(op ssa.Op, args ...*ssa.Value) (*ssa.Value, bool) { |
| f2i := func(t *types.Type) *types.Type { |
| switch t.Kind() { |
| case types.TFLOAT32: |
| return types.Types[types.TUINT32] |
| case types.TFLOAT64: |
| return types.Types[types.TUINT64] |
| } |
| return t |
| } |
| |
| if callDef, ok := softFloatOps[op]; ok { |
| switch op { |
| case ssa.OpLess32F, |
| ssa.OpLess64F, |
| ssa.OpLeq32F, |
| ssa.OpLeq64F: |
| args[0], args[1] = args[1], args[0] |
| case ssa.OpSub32F, |
| ssa.OpSub64F: |
| args[1] = s.newValue1(s.ssaOp(ir.ONEG, types.Types[callDef.rtype]), args[1].Type, args[1]) |
| } |
| |
| // runtime functions take uints for floats and returns uints. |
| // Convert to uints so we use the right calling convention. |
| for i, a := range args { |
| if a.Type.IsFloat() { |
| args[i] = s.newValue1(ssa.OpCopy, f2i(a.Type), a) |
| } |
| } |
| |
| rt := types.Types[callDef.rtype] |
| result := s.rtcall(callDef.rtfn, true, []*types.Type{f2i(rt)}, args...)[0] |
| if rt.IsFloat() { |
| result = s.newValue1(ssa.OpCopy, rt, result) |
| } |
| if op == ssa.OpNeq32F || op == ssa.OpNeq64F { |
| result = s.newValue1(ssa.OpNot, result.Type, result) |
| } |
| return result, true |
| } |
| return nil, false |
| } |
| |
| var intrinsics map[intrinsicKey]intrinsicBuilder |
| |
| // An intrinsicBuilder converts a call node n into an ssa value that |
| // implements that call as an intrinsic. args is a list of arguments to the func. |
| type intrinsicBuilder func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value |
| |
| type intrinsicKey struct { |
| arch *sys.Arch |
| pkg string |
| fn string |
| } |
| |
| func InitTables() { |
| intrinsics = map[intrinsicKey]intrinsicBuilder{} |
| |
| var all []*sys.Arch |
| var p4 []*sys.Arch |
| var p8 []*sys.Arch |
| var lwatomics []*sys.Arch |
| for _, a := range &sys.Archs { |
| all = append(all, a) |
| if a.PtrSize == 4 { |
| p4 = append(p4, a) |
| } else { |
| p8 = append(p8, a) |
| } |
| if a.Family != sys.PPC64 { |
| lwatomics = append(lwatomics, a) |
| } |
| } |
| |
| // add adds the intrinsic b for pkg.fn for the given list of architectures. |
| add := func(pkg, fn string, b intrinsicBuilder, archs ...*sys.Arch) { |
| for _, a := range archs { |
| intrinsics[intrinsicKey{a, pkg, fn}] = b |
| } |
| } |
| // addF does the same as add but operates on architecture families. |
| addF := func(pkg, fn string, b intrinsicBuilder, archFamilies ...sys.ArchFamily) { |
| m := 0 |
| for _, f := range archFamilies { |
| if f >= 32 { |
| panic("too many architecture families") |
| } |
| m |= 1 << uint(f) |
| } |
| for _, a := range all { |
| if m>>uint(a.Family)&1 != 0 { |
| intrinsics[intrinsicKey{a, pkg, fn}] = b |
| } |
| } |
| } |
| // alias defines pkg.fn = pkg2.fn2 for all architectures in archs for which pkg2.fn2 exists. |
| alias := func(pkg, fn, pkg2, fn2 string, archs ...*sys.Arch) { |
| aliased := false |
| for _, a := range archs { |
| if b, ok := intrinsics[intrinsicKey{a, pkg2, fn2}]; ok { |
| intrinsics[intrinsicKey{a, pkg, fn}] = b |
| aliased = true |
| } |
| } |
| if !aliased { |
| panic(fmt.Sprintf("attempted to alias undefined intrinsic: %s.%s", pkg, fn)) |
| } |
| } |
| |
| /******** runtime ********/ |
| if !base.Flag.Cfg.Instrumenting { |
| add("runtime", "slicebytetostringtmp", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| // Compiler frontend optimizations emit OBYTES2STRTMP nodes |
| // for the backend instead of slicebytetostringtmp calls |
| // when not instrumenting. |
| return s.newValue2(ssa.OpStringMake, n.Type(), args[0], args[1]) |
| }, |
| all...) |
| } |
| addF("runtime/internal/math", "MulUintptr", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| if s.config.PtrSize == 4 { |
| return s.newValue2(ssa.OpMul32uover, types.NewTuple(types.Types[types.TUINT], types.Types[types.TUINT]), args[0], args[1]) |
| } |
| return s.newValue2(ssa.OpMul64uover, types.NewTuple(types.Types[types.TUINT], types.Types[types.TUINT]), args[0], args[1]) |
| }, |
| sys.AMD64, sys.I386, sys.Loong64, sys.MIPS64, sys.RISCV64, sys.ARM64) |
| alias("runtime", "mulUintptr", "runtime/internal/math", "MulUintptr", all...) |
| add("runtime", "KeepAlive", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| data := s.newValue1(ssa.OpIData, s.f.Config.Types.BytePtr, args[0]) |
| s.vars[memVar] = s.newValue2(ssa.OpKeepAlive, types.TypeMem, data, s.mem()) |
| return nil |
| }, |
| all...) |
| add("runtime", "getclosureptr", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return s.newValue0(ssa.OpGetClosurePtr, s.f.Config.Types.Uintptr) |
| }, |
| all...) |
| |
| add("runtime", "getcallerpc", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return s.newValue0(ssa.OpGetCallerPC, s.f.Config.Types.Uintptr) |
| }, |
| all...) |
| |
| add("runtime", "getcallersp", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return s.newValue1(ssa.OpGetCallerSP, s.f.Config.Types.Uintptr, s.mem()) |
| }, |
| all...) |
| |
| addF("runtime", "publicationBarrier", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue1(ssa.OpPubBarrier, types.TypeMem, s.mem()) |
| return nil |
| }, |
| sys.ARM64, sys.PPC64) |
| |
| brev_arch := []sys.ArchFamily{sys.AMD64, sys.I386, sys.ARM64, sys.ARM, sys.S390X} |
| if buildcfg.GOPPC64 >= 10 { |
| // Use only on Power10 as the new byte reverse instructions that Power10 provide |
| // make it worthwhile as an intrinsic |
| brev_arch = append(brev_arch, sys.PPC64) |
| } |
| /******** runtime/internal/sys ********/ |
| addF("runtime/internal/sys", "Bswap32", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return s.newValue1(ssa.OpBswap32, types.Types[types.TUINT32], args[0]) |
| }, |
| brev_arch...) |
| addF("runtime/internal/sys", "Bswap64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return s.newValue1(ssa.OpBswap64, types.Types[types.TUINT64], args[0]) |
| }, |
| brev_arch...) |
| |
| /****** Prefetch ******/ |
| makePrefetchFunc := func(op ssa.Op) func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| return func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue2(op, types.TypeMem, args[0], s.mem()) |
| return nil |
| } |
| } |
| |
| // Make Prefetch intrinsics for supported platforms |
| // On the unsupported platforms stub function will be eliminated |
| addF("runtime/internal/sys", "Prefetch", makePrefetchFunc(ssa.OpPrefetchCache), |
| sys.AMD64, sys.ARM64, sys.PPC64) |
| addF("runtime/internal/sys", "PrefetchStreamed", makePrefetchFunc(ssa.OpPrefetchCacheStreamed), |
| sys.AMD64, sys.ARM64, sys.PPC64) |
| |
| /******** runtime/internal/atomic ********/ |
| addF("runtime/internal/atomic", "Load", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoad32, types.NewTuple(types.Types[types.TUINT32], types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT32], v) |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "Load8", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoad8, types.NewTuple(types.Types[types.TUINT8], types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT8], v) |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "Load64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoad64, types.NewTuple(types.Types[types.TUINT64], types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT64], v) |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "LoadAcq", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoadAcq32, types.NewTuple(types.Types[types.TUINT32], types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT32], v) |
| }, |
| sys.PPC64, sys.S390X) |
| addF("runtime/internal/atomic", "LoadAcq64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoadAcq64, types.NewTuple(types.Types[types.TUINT64], types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT64], v) |
| }, |
| sys.PPC64) |
| addF("runtime/internal/atomic", "Loadp", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue2(ssa.OpAtomicLoadPtr, types.NewTuple(s.f.Config.Types.BytePtr, types.TypeMem), args[0], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, s.f.Config.Types.BytePtr, v) |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| |
| addF("runtime/internal/atomic", "Store", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStore32, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "Store8", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStore8, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "Store64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStore64, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "StorepNoWB", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStorePtrNoWB, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "StoreRel", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStoreRel32, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.PPC64, sys.S390X) |
| addF("runtime/internal/atomic", "StoreRel64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| s.vars[memVar] = s.newValue3(ssa.OpAtomicStoreRel64, types.TypeMem, args[0], args[1], s.mem()) |
| return nil |
| }, |
| sys.PPC64) |
| |
| addF("runtime/internal/atomic", "Xchg", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue3(ssa.OpAtomicExchange32, types.NewTuple(types.Types[types.TUINT32], types.TypeMem), args[0], args[1], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT32], v) |
| }, |
| sys.AMD64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| addF("runtime/internal/atomic", "Xchg64", |
| func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| v := s.newValue3(ssa.OpAtomicExchange64, types.NewTuple(types.Types[types.TUINT64], types.TypeMem), args[0], args[1], s.mem()) |
| s.vars[memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) |
| return s.newValue1(ssa.OpSelect0, types.Types[types.TUINT64], v) |
| }, |
| sys.AMD64, sys.Loong64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) |
| |
| type atomicOpEmitter func(s *state, n *ir.CallExpr, args []*ssa.Value, op ssa.Op, typ types.Kind) |
| |
| makeAtomicGuardedIntrinsicARM64 := func(op0, op1 ssa.Op, typ, rtyp types.Kind, emit atomicOpEmitter) intrinsicBuilder { |
| |
| return func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value { |
| // Target Atomic feature is identified by dynamic detection |
| addr := s.entryNewValue1A(ssa.OpAddr, types.Types[types.TBOOL].PtrTo(), ir.Syms.ARM64HasATOMICS, s.sb) |
| v := s.load(types.Types[types.TBOOL], addr) |
| b := s.endBlock() |
| b.Kind = ssa.BlockIf |
| b.SetControl(v) |
| bTrue := s.f.NewBlock(ssa.BlockPlain) |
| bFalse := s.f.NewBlock(ssa.BlockPlain) |
| bEnd := s.f.NewBlock(ssa.BlockPlain) |
| b.AddEdgeTo(bTrue) |
| b.AddEdgeTo(bFalse) |
| b.Likely = ssa.BranchLikely |
| |
| // We have atomic instructions - use it directly. |
| s.startBlock(bTrue) |
| emit(s, n, args, op1, typ) |
|
