diff --git a/src/cmd/compile/internal/gc/asm_test.go b/src/cmd/compile/internal/gc/asm_test.go
index c45615ae3a0..c2fc9862f3b 100644
--- a/src/cmd/compile/internal/gc/asm_test.go
+++ b/src/cmd/compile/internal/gc/asm_test.go
@@ -3245,6 +3245,24 @@ var linuxARM64Tests = []*asmTest{
 		pos: []string{"STP"},
 		neg: []string{"MOVB", "MOVH"},
 	},
+	{
+		fn: `
+		func $(a *[39]byte) {
+			*a = [39]byte{}
+		}
+		`,
+		pos: []string{"MOVD"},
+		neg: []string{"MOVB", "MOVH", "MOVW"},
+	},
+	{
+		fn: `
+		func $(a *[30]byte) {
+			*a = [30]byte{}
+		}
+		`,
+		pos: []string{"STP"},
+		neg: []string{"MOVB", "MOVH", "MOVW"},
+	},
 }
 
 var linuxMIPSTests = []*asmTest{
diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules
index 9f6ef57d434..c6057f24613 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules
@@ -399,10 +399,14 @@
 				(STP [0] ptr (MOVDconst [0]) (MOVDconst [0]) mem))))
 
 // strip off fractional word zeroing
-(Zero [s] ptr mem) && s%16 != 0 && s > 16 ->
-	(Zero [s-s%16]
-		(OffPtr <ptr.Type> ptr [s%16])
-		(Zero [s%16] ptr mem))
+(Zero [s] ptr mem) && s%16 != 0 && s%16 <= 8 && s > 16 ->
+	(Zero [8]
+		(OffPtr <ptr.Type> ptr [s-8])
+		(Zero [s-s%16] ptr mem))
+(Zero [s] ptr mem) && s%16 != 0 && s%16 > 8 && s > 16 ->
+	(Zero [16]
+		(OffPtr <ptr.Type> ptr [s-16])
+		(Zero [s-s%16] ptr mem))
 
 // medium zeroing uses a duff device
 // 4, 16, and 64 are magic constants, see runtime/mkduff.go
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index 1bb21d8a2c3..53331eda312 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -18551,24 +18551,48 @@ func rewriteValueARM64_OpZero_20(v *Value) bool {
 	config := b.Func.Config
 	_ = config
 	// match: (Zero [s] ptr mem)
-	// cond: s%16 != 0 && s > 16
-	// result: (Zero [s-s%16] (OffPtr <ptr.Type> ptr [s%16]) (Zero [s%16] ptr mem))
+	// cond: s%16 != 0 && s%16 <= 8 && s > 16
+	// result: (Zero [8] (OffPtr <ptr.Type> ptr [s-8]) (Zero [s-s%16] ptr mem))
 	for {
 		s := v.AuxInt
 		_ = v.Args[1]
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(s%16 != 0 && s > 16) {
+		if !(s%16 != 0 && s%16 <= 8 && s > 16) {
 			break
 		}
 		v.reset(OpZero)
-		v.AuxInt = s - s%16
+		v.AuxInt = 8
 		v0 := b.NewValue0(v.Pos, OpOffPtr, ptr.Type)
-		v0.AuxInt = s % 16
+		v0.AuxInt = s - 8
 		v0.AddArg(ptr)
 		v.AddArg(v0)
 		v1 := b.NewValue0(v.Pos, OpZero, types.TypeMem)
-		v1.AuxInt = s % 16
+		v1.AuxInt = s - s%16
+		v1.AddArg(ptr)
+		v1.AddArg(mem)
+		v.AddArg(v1)
+		return true
+	}
+	// match: (Zero [s] ptr mem)
+	// cond: s%16 != 0 && s%16 > 8 && s > 16
+	// result: (Zero [16] (OffPtr <ptr.Type> ptr [s-16]) (Zero [s-s%16] ptr mem))
+	for {
+		s := v.AuxInt
+		_ = v.Args[1]
+		ptr := v.Args[0]
+		mem := v.Args[1]
+		if !(s%16 != 0 && s%16 > 8 && s > 16) {
+			break
+		}
+		v.reset(OpZero)
+		v.AuxInt = 16
+		v0 := b.NewValue0(v.Pos, OpOffPtr, ptr.Type)
+		v0.AuxInt = s - 16
+		v0.AddArg(ptr)
+		v.AddArg(v0)
+		v1 := b.NewValue0(v.Pos, OpZero, types.TypeMem)
+		v1.AuxInt = s - s%16
 		v1.AddArg(ptr)
 		v1.AddArg(mem)
 		v.AddArg(v1)