runtime: reduce slice growth during append to 2x

The new inlined code for append assumed that it could pass the
desired new cap to growslice, not the number of new elements.
But growslice still interpreted the argument as the number of new elements,
making it always grow by >2x (more precisely, 2x+1 rounded up
to the next malloc block size). At the time, I had intended to change
the other callers to use the new cap as well, but it's too late for that.
Instead, introduce growslice_n for the old callers and keep growslice
for the inlined (common case) caller.

Fixes #11403.

Filed #11419 to merge them.

Change-Id: I1338b1e5b352f3be4e43641f44b652ef7195251b
Reviewed-on: https://go-review.googlesource.com/11541
Reviewed-by: Austin Clements <austin@google.com>

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

@@ -130,7 +130,8 @@ const runtimeimport = "" +
 	"func @\"\".selectgo (@\"\".sel·1 *byte)\n" +
 	"func @\"\".block ()\n" +
 	"func @\"\".makeslice (@\"\".typ·2 *byte, @\"\".nel·3 int64, @\"\".cap·4 int64) (@\"\".ary·1 []any)\n" +
-	"func @\"\".growslice (@\"\".typ·2 *byte, @\"\".old·3 []any, @\"\".n·4 int) (@\"\".ary·1 []any)\n" +
+	"func @\"\".growslice (@\"\".typ·2 *byte, @\"\".old·3 []any, @\"\".cap·4 int) (@\"\".ary·1 []any)\n" +
+	"func @\"\".growslice_n (@\"\".typ·2 *byte, @\"\".old·3 []any, @\"\".n·4 int) (@\"\".ary·1 []any)\n" +
 	"func @\"\".memmove (@\"\".to·1 *any, @\"\".frm·2 *any, @\"\".length·3 uintptr)\n" +
 	"func @\"\".memclr (@\"\".ptr·1 *byte, @\"\".length·2 uintptr)\n" +
 	"func @\"\".memequal (@\"\".x·2 *any, @\"\".y·3 *any, @\"\".size·4 uintptr) (? bool)\n" +

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

@@ -162,7 +162,8 @@ func selectgo(sel *byte)
 func block()
 
 func makeslice(typ *byte, nel int64, cap int64) (ary []any)
-func growslice(typ *byte, old []any, n int) (ary []any)
+func growslice(typ *byte, old []any, cap int) (ary []any)
+func growslice_n(typ *byte, old []any, n int) (ary []any)
 func memmove(to *any, frm *any, length uintptr)
 func memclr(ptr *byte, length uintptr)
 

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

@@ -2884,7 +2884,7 @@ func addstr(n *Node, init **NodeList) *Node {
 //   init {
 //     s := l1
 //     if n := len(l1) + len(l2) - cap(s); n > 0 {
-//       s = growslice(s, n)
+//       s = growslice_n(s, n)
 //     }
 //     s = s[:len(l1)+len(l2)]
 //     memmove(&s[len(l1)], &l2[0], len(l2)*sizeof(T))
@@ -2918,11 +2918,11 @@ func appendslice(n *Node, init **NodeList) *Node {
 
 	nif.Left = Nod(OGT, nt, Nodintconst(0))
 
-	// instantiate growslice(Type*, []any, int) []any
+	// instantiate growslice_n(Type*, []any, int) []any
-	fn := syslook("growslice", 1) //   growslice(<type>, old []T, n int64) (ret []T)
+	fn := syslook("growslice_n", 1) //   growslice_n(<type>, old []T, n int64) (ret []T)
 	substArgTypes(fn, s.Type.Type, s.Type.Type)
 
-	// s = growslice(T, s, n)
+	// s = growslice_n(T, s, n)
 	nif.Nbody = list1(Nod(OAS, s, mkcall1(fn, s.Type, &nif.Ninit, typename(s.Type), s, nt)))
 
 	l = list(l, nif)
@@ -2997,7 +2997,7 @@ func appendslice(n *Node, init **NodeList) *Node {
 //     s := src
 //     const argc = len(args) - 1
 //     if cap(s) - len(s) < argc {
-//	    s = growslice(s, argc)
+//	    s = growslice(s, len(s)+argc)
 //     }
 //     n := len(s)
 //     s = s[:n+argc]
@@ -3050,10 +3050,10 @@ func walkappend(n *Node, init **NodeList, dst *Node) *Node {
 	nx := Nod(OIF, nil, nil)       // if cap(s) - len(s) < argc
 	nx.Left = Nod(OLT, Nod(OSUB, Nod(OCAP, ns, nil), Nod(OLEN, ns, nil)), na)
 
-	fn := syslook("growslice", 1) //   growslice(<type>, old []T, n int) (ret []T)
+	fn := syslook("growslice", 1) //   growslice(<type>, old []T, mincap int) (ret []T)
 	substArgTypes(fn, ns.Type.Type, ns.Type.Type)
 
-	nx.Nbody = list1(Nod(OAS, ns, mkcall1(fn, ns.Type, &nx.Ninit, typename(ns.Type), ns, na)))
+	nx.Nbody = list1(Nod(OAS, ns, mkcall1(fn, ns.Type, &nx.Ninit, typename(ns.Type), ns, Nod(OADD, Nod(OLEN, ns, nil), na))))
 
 	l = list(l, nx)
 

src/runtime/runtime_test.go

@@ -261,3 +261,43 @@ func TestBadOpen(t *testing.T) {
 		t.Errorf("close()=%d, want -1", c)
 	}
 }
+
+func TestAppendGrowth(t *testing.T) {
+	var x []int64
+	check := func(want int) {
+		if cap(x) != want {
+			t.Errorf("len=%d, cap=%d, want cap=%d", len(x), cap(x), want)
+		}
+	}
+
+	check(0)
+	want := 1
+	for i := 1; i <= 100; i++ {
+		x = append(x, 1)
+		check(want)
+		if i&(i-1) == 0 {
+			want = 2 * i
+		}
+	}
+}
+
+var One = []int64{1}
+
+func TestAppendSliceGrowth(t *testing.T) {
+	var x []int64
+	check := func(want int) {
+		if cap(x) != want {
+			t.Errorf("len=%d, cap=%d, want cap=%d", len(x), cap(x), want)
+		}
+	}
+
+	check(0)
+	want := 1
+	for i := 1; i <= 100; i++ {
+		x = append(x, One...)
+		check(want)
+		if i&(i-1) == 0 {
+			want = 2 * i
+		}
+	}
+}

src/runtime/slice.go

@@ -33,12 +33,22 @@ func makeslice(t *slicetype, len64, cap64 int64) slice {
 	return slice{p, len, cap}
 }
 
-func growslice(t *slicetype, old slice, n int) slice {
+// growslice_n is a variant of growslice that takes the number of new elements
+// instead of the new minimum capacity.
+// TODO(rsc): This is used by append(slice, slice...).
+// The compiler should change that code to use growslice directly (issue #11419).
+func growslice_n(t *slicetype, old slice, n int) slice {
 	if n < 1 {
 		panic(errorString("growslice: invalid n"))
 	}
+	return growslice(t, old, old.cap+n)
+}
 
-	cap := old.cap + n
+// growslice handles slice growth during append.
+// It is passed the slice type, the old slice, and the desired new minimum capacity,
+// and it returns a new slice with at least that capacity, with the old data
+// copied into it.
+func growslice(t *slicetype, old slice, cap int) slice {
 	if cap < old.cap || t.elem.size > 0 && uintptr(cap) > _MaxMem/uintptr(t.elem.size) {
 		panic(errorString("growslice: cap out of range"))
 	}