cmd/compile: de-virtualize interface calls

With this change, code like

    h := sha1.New()
    h.Write(buf)
    sum := h.Sum()

gets compiled into static calls rather than
interface calls, because the compiler is able
to prove that 'h' is really a *sha1.digest.

The InterCall re-write rule hits a few dozen times
during make.bash, and hundreds of times during all.bash.

The most common pattern identified by the compiler
is a constructor like

    func New() Interface { return &impl{...} }

where the constructor gets inlined into the caller,
and the result is used immediately. Examples include
{sha1,md5,crc32,crc64,...}.New, base64.NewEncoder,
base64.NewDecoder, errors.New, net.Pipe, and so on.

Some existing benchmarks that change on darwin/amd64:

Crc64/ISO4KB-8        2.67µs ± 1%    2.66µs ± 0%  -0.36%  (p=0.015 n=10+10)
Crc64/ISO1KB-8         694ns ± 0%     690ns ± 1%  -0.59%  (p=0.001 n=10+10)
Adler32KB-8            473ns ± 1%     471ns ± 0%  -0.39%  (p=0.010 n=10+9)

On architectures like amd64, the reduction in code size
appears to contribute more to benchmark improvements than just
removing the indirect call, since that branch gets predicted
accurately when called in a loop.

Updates #19361

Change-Id: Ia9d30afdd5f6b4d38d38b14b88f308acae8ce7ed
Reviewed-on: https://go-review.googlesource.com/37751
Run-TryBot: Philip Hofer <phofer@umich.edu>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/gc/main.go

@@ -483,7 +483,11 @@ func Main() {
 			}
 		}
 
+		// Just before compilation, compile itabs found on
+		// the right side of OCONVIFACE so that methods
+		// can be de-virtualized during compilation.
 		Curfn = nil
+		peekitabs()
 
 		// Phase 8: Compile top level functions.
 		// Don't use range--walk can add functions to xtop.

src/cmd/compile/internal/gc/reflect.go

@@ -16,6 +16,15 @@ import (
 type itabEntry struct {
 	t, itype *Type
 	sym      *Sym
+
+	// symbol of the itab itself;
+	// filled in lazily after typecheck
+	lsym *obj.LSym
+
+	// symbols of each method in
+	// the itab, sorted by byte offset;
+	// filled in at the same time as lsym
+	entries []*obj.LSym
 }
 
 type ptabEntry struct {
@@ -415,7 +424,6 @@ func imethods(t *Type) []*Sig {
 		// Generate the method body, so that compiled
 		// code can refer to it.
 		isym := methodsym(method, t, 0)
-
 		if !isym.Siggen() {
 			isym.SetSiggen(true)
 			genwrapper(t, f, isym, 0)
@@ -1379,6 +1387,78 @@ ok:
 	return s
 }
 
+// for each itabEntry, gather the methods on
+// the concrete type that implement the interface
+func peekitabs() {
+	for i := range itabs {
+		tab := &itabs[i]
+		methods := genfun(tab.t, tab.itype)
+		if len(methods) == 0 {
+			continue
+		}
+		tab.lsym = Linksym(tab.sym)
+		tab.entries = methods
+	}
+}
+
+// for the given concrete type and interface
+// type, return the (sorted) set of methods
+// on the concrete type that implement the interface
+func genfun(t, it *Type) []*obj.LSym {
+	if t == nil || it == nil {
+		return nil
+	}
+	sigs := imethods(it)
+	methods := methods(t)
+	out := make([]*obj.LSym, 0, len(sigs))
+	if len(sigs) == 0 {
+		return nil
+	}
+
+	// both sigs and methods are sorted by name,
+	// so we can find the intersect in a single pass
+	for _, m := range methods {
+		if m.name == sigs[0].name {
+			out = append(out, Linksym(m.isym))
+			sigs = sigs[1:]
+			if len(sigs) == 0 {
+				break
+			}
+		}
+	}
+
+	return out
+}
+
+// itabsym uses the information gathered in
+// peekitabs to de-virtualize interface methods.
+// Since this is called by the SSA backend, it shouldn't
+// generate additional Nodes, Syms, etc.
+func itabsym(it *obj.LSym, offset int64) *obj.LSym {
+	var syms []*obj.LSym
+	if it == nil {
+		return nil
+	}
+
+	for i := range itabs {
+		e := &itabs[i]
+		if e.lsym == it {
+			syms = e.entries
+			break
+		}
+	}
+	if syms == nil {
+		return nil
+	}
+
+	// keep this arithmetic in sync with *itab layout
+	methodnum := int((offset - 3*int64(Widthptr) - 8) / int64(Widthptr))
+	if methodnum >= len(syms) {
+		return nil
+	}
+	return syms[methodnum]
+}
+
 func dumptypestructs() {
 	// copy types from externdcl list to signatlist
 	for _, n := range externdcl {

src/cmd/compile/internal/gc/ssa.go

@@ -4931,6 +4931,10 @@ func (e *ssaExport) SplitArray(name ssa.LocalSlot) ssa.LocalSlot {
 	return ssa.LocalSlot{N: n, Type: et, Off: name.Off}
 }
 
+func (e *ssaExport) DerefItab(it *obj.LSym, offset int64) *obj.LSym {
+	return itabsym(it, offset)
+}
+
 // namedAuto returns a new AUTO variable with the given name and type.
 // These are exposed to the debugger.
 func (e *ssaExport) namedAuto(name string, typ ssa.Type) ssa.GCNode {

src/cmd/compile/internal/gc/subr.go

@@ -1684,7 +1684,6 @@ func structargs(tl *Type, mustname bool) []*Node {
 //	rcvr - U
 //	method - M func (t T)(), a TFIELD type struct
 //	newnam - the eventual mangled name of this function
-
 func genwrapper(rcvr *Type, method *Field, newnam *Sym, iface int) {
 	if false && Debug['r'] != 0 {
 		fmt.Printf("genwrapper rcvrtype=%v method=%v newnam=%v\n", rcvr, method, newnam)
@@ -1720,6 +1719,7 @@ func genwrapper(rcvr *Type, method *Field, newnam *Sym, iface int) {
 	fn.Func.Nname = newname(newnam)
 	fn.Func.Nname.Name.Defn = fn
 	fn.Func.Nname.Name.Param.Ntype = t
+	fn.Func.Nname.Sym.SetExported(true) // prevent export; see closure.go
 	declare(fn.Func.Nname, PFUNC)
 	funchdr(fn)
 
@@ -1923,6 +1923,14 @@ func implements(t, iface *Type, m, samename **Field, ptr *int) bool {
 		}
 	}
 
+	// We're going to emit an OCONVIFACE.
+	// Call itabname so that (t, iface)
+	// gets added to itabs early, which allows
+	// us to de-virtualize calls through this
+	// type/interface pair later. See peekitabs in reflect.go
+	if isdirectiface(t0) && !iface.IsEmptyInterface() {
+		itabname(t0, iface)
+	}
 	return true
 }
 

src/cmd/compile/internal/ssa/config.go

@@ -121,6 +121,12 @@ type Frontend interface {
 	SplitArray(LocalSlot) LocalSlot              // array must be length 1
 	SplitInt64(LocalSlot) (LocalSlot, LocalSlot) // returns (hi, lo)
 
+	// DerefItab dereferences an itab function
+	// entry, given the symbol of the itab and
+	// the byte offset of the function pointer.
+	// It may return nil.
+	DerefItab(sym *obj.LSym, offset int64) *obj.LSym
+
 	// Line returns a string describing the given position.
 	Line(src.XPos) string
 

src/cmd/compile/internal/ssa/export_test.go

@@ -112,6 +112,7 @@ func (d DummyFrontend) TypeInt() Type     { return TypeInt64 }
 func (d DummyFrontend) TypeUintptr() Type                            { return TypeUInt64 }
 func (d DummyFrontend) TypeString() Type                             { panic("unimplemented") }
 func (d DummyFrontend) TypeBytePtr() Type                            { return TypeBytePtr }
+func (d DummyFrontend) DerefItab(sym *obj.LSym, off int64) *obj.LSym { return nil }
 
 func (d DummyFrontend) CanSSA(t Type) bool {
 	// There are no un-SSAable types in dummy land.

src/cmd/compile/internal/ssa/gen/generic.rules

@@ -1431,3 +1431,10 @@
 	&& c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value
 	&& warnRule(config.Debug_checknil() && v.Pos.Line() > 1, v, "removed nil check")
 	-> (Invalid)
+
+// De-virtualize interface calls into static calls.
+// Note that (ITab (IMake)) doesn't get
+// rewritten until after the first opt pass,
+// so this rule should trigger reliably.
+(InterCall [argsize] (Load (OffPtr [off] (ITab (IMake (Addr {itab} (SB)) _))) _) mem) && devirt(v, itab, off) != nil ->
+	(StaticCall [argsize] {devirt(v, itab, off)} mem)

src/cmd/compile/internal/ssa/rewrite.go

@@ -5,6 +5,7 @@
 package ssa
 
 import (
+	"cmd/internal/obj"
 	"crypto/sha1"
 	"fmt"
 	"math"
@@ -384,6 +385,25 @@ func uaddOvf(a, b int64) bool {
 	return uint64(a)+uint64(b) < uint64(a)
 }
 
+// de-virtualize an InterCall
+// 'sym' is the symbol for the itab
+func devirt(v *Value, sym interface{}, offset int64) *obj.LSym {
+	f := v.Block.Func
+	ext, ok := sym.(*ExternSymbol)
+	if !ok {
+		return nil
+	}
+	lsym := f.Config.Frontend().DerefItab(ext.Sym, offset)
+	if f.pass.debug > 0 {
+		if lsym != nil {
+			f.Config.Warnl(v.Pos, "de-virtualizing call")
+		} else {
+			f.Config.Warnl(v.Pos, "couldn't de-virtualize call")
+		}
+	}
+	return lsym
+}
+
 // isSamePtr reports whether p1 and p2 point to the same address.
 func isSamePtr(p1, p2 *Value) bool {
 	if p1 == p2 {

src/cmd/compile/internal/ssa/rewritegeneric.go

@@ -124,6 +124,8 @@ func rewriteValuegeneric(v *Value, config *Config) bool {
 		return rewriteValuegeneric_OpGreater8U(v, config)
 	case OpIMake:
 		return rewriteValuegeneric_OpIMake(v, config)
+	case OpInterCall:
+		return rewriteValuegeneric_OpInterCall(v, config)
 	case OpIsInBounds:
 		return rewriteValuegeneric_OpIsInBounds(v, config)
 	case OpIsNonNil:
@@ -5736,6 +5738,52 @@ func rewriteValuegeneric_OpIMake(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValuegeneric_OpInterCall(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (InterCall [argsize] (Load (OffPtr [off] (ITab (IMake (Addr {itab} (SB)) _))) _) mem)
+	// cond: devirt(v, itab, off) != nil
+	// result: (StaticCall [argsize] {devirt(v, itab, off)} mem)
+	for {
+		argsize := v.AuxInt
+		v_0 := v.Args[0]
+		if v_0.Op != OpLoad {
+			break
+		}
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpOffPtr {
+			break
+		}
+		off := v_0_0.AuxInt
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpITab {
+			break
+		}
+		v_0_0_0_0 := v_0_0_0.Args[0]
+		if v_0_0_0_0.Op != OpIMake {
+			break
+		}
+		v_0_0_0_0_0 := v_0_0_0_0.Args[0]
+		if v_0_0_0_0_0.Op != OpAddr {
+			break
+		}
+		itab := v_0_0_0_0_0.Aux
+		v_0_0_0_0_0_0 := v_0_0_0_0_0.Args[0]
+		if v_0_0_0_0_0_0.Op != OpSB {
+			break
+		}
+		mem := v.Args[1]
+		if !(devirt(v, itab, off) != nil) {
+			break
+		}
+		v.reset(OpStaticCall)
+		v.AuxInt = argsize
+		v.Aux = devirt(v, itab, off)
+		v.AddArg(mem)
+		return true
+	}
+	return false
+}
 func rewriteValuegeneric_OpIsInBounds(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b

test/devirt.go

+// errorcheck -0 -d=ssa/opt/debug=3
+
+package main
+
+import (
+	"crypto/sha1"
+	"errors"
+	"fmt"
+	"sync"
+)
+
+func f0() {
+	v := errors.New("error string")
+	_ = v.Error() // ERROR "de-virtualizing call$"
+}
+
+func f1() {
+	h := sha1.New()
+	buf := make([]byte, 4)
+	h.Write(buf)   // ERROR "de-virtualizing call$"
+	_ = h.Sum(nil) // ERROR "de-virtualizing call$"
+}
+
+func f2() {
+	// trickier case: make sure we see this is *sync.rlocker
+	// instead of *sync.RWMutex,
+	// even though they are the same pointers
+	var m sync.RWMutex
+	r := m.RLocker()
+
+	// deadlock if the type of 'r' is improperly interpreted
+	// as *sync.RWMutex
+	r.Lock() // ERROR "de-virtualizing call$"
+	m.RLock()
+	r.Unlock() // ERROR "de-virtualizing call$"
+	m.RUnlock()
+}
+
+type multiword struct{ a, b, c int }
+
+func (m multiword) Error() string { return fmt.Sprintf("%d, %d, %d", m.a, m.b, m.c) }
+
+func f3() {
+	// can't de-virtualize this one yet;
+	// it passes through a call to iconvT2I
+	var err error
+	err = multiword{1, 2, 3}
+	if err.Error() != "1, 2, 3" {
+		panic("bad call")
+	}
+
+	// ... but we can do this one
+	err = &multiword{1, 2, 3}
+	if err.Error() != "1, 2, 3" { // ERROR "de-virtualizing call$"
+		panic("bad call")
+	}
+}
+
+func main() {
+	f0()
+	f1()
+	f2()
+	f3()
+}