cmd/compile: handle loops better during stack allocation of slices

Don't use the move2heap optimization if the move2heap is inside
a loop deeper than the declaration of the slice. We really only want
to do the move2heap operation once.

Change-Id: I4a68d01609c2c9d4e0abe4580839e70059393a81
Reviewed-on: https://go-review.googlesource.com/c/go/+/722440
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Junyang Shao <shaojunyang@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

src/cmd/compile/internal/slice/slice.go

@@ -152,6 +152,11 @@ func analyze(fn *ir.Func) {
 		// least weight 2. (Note: appends in loops have weight >= 2.)
 		appendWeight int
 
+		// Loop depth at declaration point.
+		// Use for heuristics only, it is not guaranteed to be correct
+		// in the presence of gotos.
+		declDepth int
+
 		// Whether we ever do cap(s), or other operations that use cap(s)
 		// (possibly implicitly), like s[i:j].
 		capUsed bool
@@ -209,6 +214,20 @@ func analyze(fn *ir.Func) {
 			i.s.Opt = nil
 			return
 		}
+		if loopDepth > i.declDepth {
+			// Conservatively, we disable this optimization when the
+			// transition is inside a loop. This can result in adding
+			// overhead unnecessarily in cases like:
+			// func f(n int, p *[]byte) {
+			//     var s []byte
+			//     for i := range n {
+			//         *p = s
+			//         s = append(s, 0)
+			//     }
+			// }
+			i.s.Opt = nil
+			return
+		}
 		i.transition = loc
 	}
 
@@ -237,7 +256,7 @@ func analyze(fn *ir.Func) {
 					// s = append(s, ...) is ok
 					i.okUses += 2
 					i.appends = append(i.appends, y)
-					i.appendWeight += 1 + loopDepth
+					i.appendWeight += 1 + (loopDepth - i.declDepth)
 				}
 				// TODO: s = append(nil, ...)?
 			}
@@ -277,6 +296,7 @@ func analyze(fn *ir.Func) {
 			n := n.(*ir.Decl)
 			if i := tracking(n.X); i != nil {
 				i.okUses++
+				i.declDepth = loopDepth
 			}
 		case ir.OINDEX:
 			n := n.(*ir.IndexExpr)

test/codegen/append.go

@@ -185,6 +185,38 @@ func Append17(n int) []int {
 	return r
 }
 
+func Append18(n int, p *[]int) {
+	var r []int
+	for i := range n {
+		// amd64:-`.*moveSliceNoCapNoScan`
+		*p = r
+		// amd64:`.*growslice`
+		r = append(r, i)
+	}
+}
+
+func Append19(n int, p [][]int) {
+	for j := range p {
+		var r []int
+		for i := range n {
+			// amd64:`.*growslice`
+			r = append(r, i)
+		}
+		// amd64:`.*moveSliceNoCapNoScan`
+		p[j] = r
+	}
+}
+
+func Append20(n int, p [][]int) {
+	for j := range p {
+		var r []int
+		// amd64:`.*growslice`
+		r = append(r, 0)
+		// amd64:-`.*moveSliceNoCapNoScan`
+		p[j] = r
+	}
+}
+
 //go:noinline
 func useSlice(s []int) {
 }