cmd/compile: optimize remaining convT2I calls

See #14874
Updates #6853

This change adds a compiler optimization for non pointer shaped convT2I.
Since itab symbols are now emitted by the compiler, the itab address can
be passed directly to convT2I instead of passing the iface type and a
cache pointer argument.

Compilebench results for the 5-commits series ending here:

name       old time/op     new time/op     delta
Template       336ms ± 4%      344ms ± 4%   +2.61%          (p=0.027 n=9+8)
Unicode        165ms ± 6%      173ms ± 7%   +5.11%          (p=0.014 n=9+9)
GoTypes        1.09s ± 1%      1.06s ± 2%   -3.29%          (p=0.000 n=9+9)
Compiler       5.09s ±10%      4.75s ±10%   -6.64%        (p=0.011 n=10+10)
MakeBash       31.1s ± 5%      30.3s ± 3%     ~           (p=0.089 n=10+10)

name       old text-bytes  new text-bytes  delta
HelloSize       558k ± 0%       558k ± 0%   +0.02%        (p=0.000 n=10+10)
CmdGoSize      6.24M ± 0%      6.11M ± 0%   -2.11%        (p=0.000 n=10+10)

name       old data-bytes  new data-bytes  delta
HelloSize      3.66k ± 0%      3.74k ± 0%   +2.41%        (p=0.000 n=10+10)
CmdGoSize       134k ± 0%       162k ± 0%  +20.76%        (p=0.000 n=10+10)

name       old bss-bytes   new bss-bytes   delta
HelloSize       126k ± 0%       126k ± 0%     ~     (all samples are equal)
CmdGoSize       149k ± 0%       146k ± 0%   -2.17%        (p=0.000 n=10+10)

name       old exe-bytes   new exe-bytes   delta
HelloSize       924k ± 0%       924k ± 0%   +0.05%        (p=0.000 n=10+10)
CmdGoSize      9.77M ± 0%      9.62M ± 0%   -1.47%        (p=0.000 n=10+10)

Change-Id: Ib230ddc04988824035c32287ae544a965fedd344
Reviewed-on: https://go-review.googlesource.com/20902Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Michel Lespinasse <walken@google.com>

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

@@ -46,11 +46,10 @@ const runtimeimport = "" +
 	"func @\"\".stringiter2 (? string, ? int) (@\"\".retk·1 int, @\"\".retv·2 rune)\n" +
 	"func @\"\".slicecopy (@\"\".to·2 any, @\"\".fr·3 any, @\"\".wid·4 uintptr \"unsafe-uintptr\") (? int)\n" +
 	"func @\"\".slicestringcopy (@\"\".to·2 any, @\"\".fr·3 any) (? int)\n" +
-	"func @\"\".typ2Itab (@\"\".typ·2 *byte, @\"\".typ2·3 *byte, @\"\".cache·4 **byte) (@\"\".ret·1 *byte)\n" +
 	"func @\"\".convI2E (@\"\".elem·2 any) (@\"\".ret·1 any)\n" +
 	"func @\"\".convI2I (@\"\".typ·2 *byte, @\"\".elem·3 any) (@\"\".ret·1 any)\n" +
 	"func @\"\".convT2E (@\"\".typ·2 *byte, @\"\".elem·3 *any, @\"\".buf·4 *any) (@\"\".ret·1 any)\n" +
-	"func @\"\".convT2I (@\"\".typ·2 *byte, @\"\".typ2·3 *byte, @\"\".cache·4 **byte, @\"\".elem·5 *any, @\"\".buf·6 *any) (@\"\".ret·1 any)\n" +
+	"func @\"\".convT2I (@\"\".tab·2 *byte, @\"\".elem·3 *any, @\"\".buf·4 *any) (@\"\".ret·1 any)\n" +
 	"func @\"\".assertE2E (@\"\".typ·1 *byte, @\"\".iface·2 any, @\"\".ret·3 *any)\n" +
 	"func @\"\".assertE2E2 (@\"\".typ·2 *byte, @\"\".iface·3 any, @\"\".ret·4 *any) (? bool)\n" +
 	"func @\"\".assertE2I (@\"\".typ·1 *byte, @\"\".iface·2 any, @\"\".ret·3 *any)\n" +

src/cmd/compile/internal/gc/builtin/runtime.go

@@ -60,11 +60,10 @@ func slicecopy(to any, fr any, wid uintptr) int
 func slicestringcopy(to any, fr any) int
 
 // interface conversions
-func typ2Itab(typ *byte, typ2 *byte, cache **byte) (ret *byte)
 func convI2E(elem any) (ret any)
 func convI2I(typ *byte, elem any) (ret any)
 func convT2E(typ *byte, elem, buf *any) (ret any)
-func convT2I(typ *byte, typ2 *byte, cache **byte, elem, buf *any) (ret any)
+func convT2I(tab *byte, elem, buf *any) (ret any)
 
 // interface type assertions  x.(T)
 func assertE2E(typ *byte, iface any, ret *any)

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

@@ -248,9 +248,7 @@ var localpkg *Pkg // package being compiled
 
 var importpkg *Pkg // package being imported
 
-var itabpkg *Pkg // fake pkg for itab cache
-
-var itab2pkg *Pkg // fake pkg for itab entries
+var itabpkg *Pkg // fake pkg for itab entries
 
 var itablinkpkg *Pkg // fake package for runtime itab entries
 

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

@@ -109,17 +109,14 @@ func Main() {
 
 	// pseudo-package, for scoping
 	builtinpkg = mkpkg("go.builtin")
-
 	builtinpkg.Prefix = "go.builtin" // not go%2ebuiltin
 
 	// pseudo-package, accessed by import "unsafe"
 	unsafepkg = mkpkg("unsafe")
-
 	unsafepkg.Name = "unsafe"
 
 	// real package, referred to by generated runtime calls
 	Runtimepkg = mkpkg("runtime")
-
 	Runtimepkg.Name = "runtime"
 
 	// pseudo-packages used in symbol tables
@@ -127,10 +124,6 @@ func Main() {
 	itabpkg.Name = "go.itab"
 	itabpkg.Prefix = "go.itab" // not go%2eitab
 
-	itab2pkg = mkpkg("go.itab2")
-	itab2pkg.Name = "go.itab2"
-	itab2pkg.Prefix = "go.itab2" // not go%2eitab2
-
 	typelinkpkg = mkpkg("go.typelink")
 	typelinkpkg.Name = "go.typelink"
 	typelinkpkg.Prefix = "go.typelink" // not go%2etypelink
@@ -140,12 +133,10 @@ func Main() {
 	itablinkpkg.Prefix = "go.itablink" // not go%2eitablink
 
 	trackpkg = mkpkg("go.track")
-
 	trackpkg.Name = "go.track"
 	trackpkg.Prefix = "go.track" // not go%2etrack
 
 	typepkg = mkpkg("type")
-
 	typepkg.Name = "type"
 
 	goroot = obj.Getgoroot()

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

@@ -951,7 +951,7 @@ func itabname(t, itype *Type) *Node {
 	if t == nil || (Isptr[t.Etype] && t.Type == nil) || isideal(t) {
 		Fatalf("itabname %v", t)
 	}
-	s := Pkglookup(Tconv(t, FmtLeft)+","+Tconv(itype, FmtLeft), itab2pkg)
+	s := Pkglookup(Tconv(t, FmtLeft)+","+Tconv(itype, FmtLeft), itabpkg)
 	if s.Def == nil {
 		n := newname(s)
 		n.Type = Types[TUINT8]

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

@@ -1003,63 +1003,15 @@ opswitch:
 		}
 
 		var ll []*Node
-		if !Isinter(n.Left.Type) {
-			ll = append(ll, typename(n.Left.Type))
-		}
-		if !isnilinter(n.Type) {
-			ll = append(ll, typename(n.Type))
-		}
-		if !Isinter(n.Left.Type) && !isnilinter(n.Type) {
-			sym := Pkglookup(Tconv(n.Left.Type, FmtLeft)+"."+Tconv(n.Type, FmtLeft), itabpkg)
-			if sym.Def == nil {
-				l := Nod(ONAME, nil, nil)
-				l.Sym = sym
-				l.Type = Ptrto(Types[TUINT8])
-				l.Addable = true
-				l.Class = PEXTERN
-				l.Xoffset = 0
-				sym.Def = l
-				ggloblsym(sym, int32(Widthptr), obj.DUPOK|obj.NOPTR)
+		if isnilinter(n.Type) {
+			if !Isinter(n.Left.Type) {
+				ll = append(ll, typename(n.Left.Type))
 			}
-
-			l := Nod(OADDR, sym.Def, nil)
-			l.Addable = true
-			ll = append(ll, l)
-
-			if isdirectiface(n.Left.Type) {
-				// For pointer types, we can make a special form of optimization
-				//
-				// These statements are put onto the expression init list:
-				// 	Itab *tab = atomicloadtype(&cache);
-				// 	if(tab == nil)
-				// 		tab = typ2Itab(type, itype, &cache);
-				//
-				// The CONVIFACE expression is replaced with this:
-				// 	OEFACE{tab, ptr};
-				l := temp(Ptrto(Types[TUINT8]))
-
-				n1 := Nod(OAS, l, sym.Def)
-				n1 = typecheck(n1, Etop)
-				init.Append(n1)
-
-				fn := syslook("typ2Itab")
-				n1 = Nod(OCALL, fn, nil)
-				n1.List.Set(ll)
-				n1 = typecheck(n1, Erv)
-				n1 = walkexpr(n1, init)
-
-				n2 := Nod(OIF, nil, nil)
-				n2.Left = Nod(OEQ, l, nodnil())
-				n2.Nbody.Set1(Nod(OAS, l, n1))
-				n2.Likely = -1
-				n2 = typecheck(n2, Etop)
-				init.Append(n2)
-
-				l = Nod(OEFACE, l, n.Left)
-				l.Typecheck = n.Typecheck
-				l.Type = n.Type
-				n = l
-				break
+		} else {
+			if Isinter(n.Left.Type) {
+				ll = append(ll, typename(n.Type))
+			} else {
+				ll = append(ll, itabname(n.Left.Type, n.Type))
 			}
 		}
 

src/runtime/iface.go

@@ -147,12 +147,6 @@ func itabsinit() {
 	unlock(&ifaceLock)
 }
 
-func typ2Itab(t *_type, inter *interfacetype, cache **itab) *itab {
-	tab := getitab(inter, t, false)
-	atomicstorep(unsafe.Pointer(cache), unsafe.Pointer(tab))
-	return tab
-}
-
 func convT2E(t *_type, elem unsafe.Pointer, x unsafe.Pointer) (e eface) {
 	if raceenabled {
 		raceReadObjectPC(t, elem, getcallerpc(unsafe.Pointer(&t)), funcPC(convT2E))
@@ -176,18 +170,14 @@ func convT2E(t *_type, elem unsafe.Pointer, x unsafe.Pointer) (e eface) {
 	return
 }
 
-func convT2I(t *_type, inter *interfacetype, cache **itab, elem unsafe.Pointer, x unsafe.Pointer) (i iface) {
+func convT2I(tab *itab, elem unsafe.Pointer, x unsafe.Pointer) (i iface) {
+	t := tab._type
 	if raceenabled {
 		raceReadObjectPC(t, elem, getcallerpc(unsafe.Pointer(&t)), funcPC(convT2I))
 	}
 	if msanenabled {
 		msanread(elem, t.size)
 	}
-	tab := (*itab)(atomic.Loadp(unsafe.Pointer(cache)))
-	if tab == nil {
-		tab = getitab(inter, t, false)
-		atomicstorep(unsafe.Pointer(cache), unsafe.Pointer(tab))
-	}
 	if isDirectIface(t) {
 		i.tab = tab
 		typedmemmove(t, unsafe.Pointer(&i.data), elem)