1 files changed, 54 insertions, 15 deletions

diff --git a/libgo/go/runtime/memmove_test.go b/libgo/go/runtime/memmove_test.go
index 396c130..7c9d2ad 100644
--- a/libgo/go/runtime/memmove_test.go
+++ b/libgo/go/runtime/memmove_test.go
@@ -286,6 +286,9 @@ var bufSizes = []int{
 	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
 	32, 64, 128, 256, 512, 1024, 2048, 4096,
 }
+var bufSizesOverlap = []int{
+	32, 64, 128, 256, 512, 1024, 2048, 4096,
+}
 
 func BenchmarkMemmove(b *testing.B) {
 	benchmarkSizes(b, bufSizes, func(b *testing.B, n int) {
@@ -297,6 +300,15 @@ func BenchmarkMemmove(b *testing.B) {
 	})
 }
 
+func BenchmarkMemmoveOverlap(b *testing.B) {
+	benchmarkSizes(b, bufSizesOverlap, func(b *testing.B, n int) {
+		x := make([]byte, n+16)
+		for i := 0; i < b.N; i++ {
+			copy(x[16:n+16], x[:n])
+		}
+	})
+}
+
 func BenchmarkMemmoveUnalignedDst(b *testing.B) {
 	benchmarkSizes(b, bufSizes, func(b *testing.B, n int) {
 		x := make([]byte, n+1)
@@ -307,6 +319,15 @@ func BenchmarkMemmoveUnalignedDst(b *testing.B) {
 	})
 }
 
+func BenchmarkMemmoveUnalignedDstOverlap(b *testing.B) {
+	benchmarkSizes(b, bufSizesOverlap, func(b *testing.B, n int) {
+		x := make([]byte, n+16)
+		for i := 0; i < b.N; i++ {
+			copy(x[16:n+16], x[1:n+1])
+		}
+	})
+}
+
 func BenchmarkMemmoveUnalignedSrc(b *testing.B) {
 	benchmarkSizes(b, bufSizes, func(b *testing.B, n int) {
 		x := make([]byte, n)
@@ -317,6 +338,15 @@ func BenchmarkMemmoveUnalignedSrc(b *testing.B) {
 	})
 }
 
+func BenchmarkMemmoveUnalignedSrcOverlap(b *testing.B) {
+	benchmarkSizes(b, bufSizesOverlap, func(b *testing.B, n int) {
+		x := make([]byte, n+1)
+		for i := 0; i < b.N; i++ {
+			copy(x[1:n+1], x[:n])
+		}
+	})
+}
+
 func TestMemclr(t *testing.T) {
 	size := 512
 	if testing.Short() {
@@ -538,21 +568,30 @@ func BenchmarkCopyFat1024(b *testing.B) {
 	}
 }
 
+// BenchmarkIssue18740 ensures that memmove uses 4 and 8 byte load/store to move 4 and 8 bytes.
+// It used to do 2 2-byte load/stores, which leads to a pipeline stall
+// when we try to read the result with one 4-byte load.
 func BenchmarkIssue18740(b *testing.B) {
-	// This tests that memmove uses one 4-byte load/store to move 4 bytes.
-	// It used to do 2 2-byte load/stores, which leads to a pipeline stall
-	// when we try to read the result with one 4-byte load.
-	var buf [4]byte
-	for j := 0; j < b.N; j++ {
-		s := uint32(0)
-		for i := 0; i < 4096; i += 4 {
-			copy(buf[:], g[i:])
-			s += binary.LittleEndian.Uint32(buf[:])
-		}
-		sink = uint64(s)
+	benchmarks := []struct {
+		name  string
+		nbyte int
+		f     func([]byte) uint64
+	}{
+		{"2byte", 2, func(buf []byte) uint64 { return uint64(binary.LittleEndian.Uint16(buf)) }},
+		{"4byte", 4, func(buf []byte) uint64 { return uint64(binary.LittleEndian.Uint32(buf)) }},
+		{"8byte", 8, func(buf []byte) uint64 { return binary.LittleEndian.Uint64(buf) }},
+	}
+
+	var g [4096]byte
+	for _, bm := range benchmarks {
+		buf := make([]byte, bm.nbyte)
+		b.Run(bm.name, func(b *testing.B) {
+			for j := 0; j < b.N; j++ {
+				for i := 0; i < 4096; i += bm.nbyte {
+					copy(buf[:], g[i:])
+					sink += bm.f(buf[:])
+				}
+			}
+		})
 	}
 }
-
-// TODO: 2 byte and 8 byte benchmarks also.
-
-var g [4096]byte