cmd/compile/internal/ssa: improve store combine optimization on arm64

Current implementation doesn't consider MOVDreg type operand and fail to combine
it into larger store. This patch fixes the issue.

Fixes #24242

Change-Id: I7d68697f80e76f48c3528ece01a602bf513248ec
Reviewed-on: https://go-review.googlesource.com/98397
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/compile/internal/ssa/gen/ARM64.rules

@@ -1520,6 +1520,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr0 w mem)
+(MOVBstore [i] {s} ptr0 (SRLconst [8] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr0 w mem)
 (MOVBstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] w) mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1530,6 +1535,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr0 w0 mem)
+(MOVBstore [i] {s} ptr0 (SRLconst [j] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr0 w0 mem)
 (MOVHstore [i] {s} ptr0 (SRLconst [16] w) x:(MOVHstore [i-2] {s} ptr1 w mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1540,6 +1550,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVWstore [i-2] {s} ptr0 w mem)
+(MOVHstore [i] {s} ptr0 (SRLconst [16] (MOVDreg w)) x:(MOVHstore [i-2] {s} ptr1 w mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVWstore [i-2] {s} ptr0 w mem)
 (MOVHstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVHstore [i-2] {s} ptr1 w0:(SRLconst [j-16] w) mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1589,6 +1604,17 @@
 	&& clobber(x1)
 	&& clobber(x2)
 	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w
+	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w))
+	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] (MOVDreg w))
+	x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] (MOVDreg w)) mem))))
+	&& x0.Uses == 1
+	&& x1.Uses == 1
+	&& x2.Uses == 1
+	&& clobber(x0)
+	&& clobber(x1)
+	&& clobber(x2)
+	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
 (MOVBstore [i] {s} ptr w
 	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w)
 	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w)
@@ -1608,10 +1634,18 @@
 	&& x.Uses == 1
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
 (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVWUreg w)) mem))
 	&& x.Uses == 1
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
 
 // FP simplification
 (FNEGS (FMULS x y)) -> (FNMULS x y)

src/cmd/compile/internal/ssa/rewriteARM64.go

@@ -138,7 +138,7 @@ func rewriteValueARM64(v *Value) bool {
 	case OpARM64MOVBreg:
 		return rewriteValueARM64_OpARM64MOVBreg_0(v)
 	case OpARM64MOVBstore:
-		return rewriteValueARM64_OpARM64MOVBstore_0(v) || rewriteValueARM64_OpARM64MOVBstore_10(v)
+		return rewriteValueARM64_OpARM64MOVBstore_0(v) || rewriteValueARM64_OpARM64MOVBstore_10(v) || rewriteValueARM64_OpARM64MOVBstore_20(v)
 	case OpARM64MOVBstorezero:
 		return rewriteValueARM64_OpARM64MOVBstorezero_0(v)
 	case OpARM64MOVDload:
@@ -158,7 +158,7 @@ func rewriteValueARM64(v *Value) bool {
 	case OpARM64MOVHreg:
 		return rewriteValueARM64_OpARM64MOVHreg_0(v)
 	case OpARM64MOVHstore:
-		return rewriteValueARM64_OpARM64MOVHstore_0(v)
+		return rewriteValueARM64_OpARM64MOVHstore_0(v) || rewriteValueARM64_OpARM64MOVHstore_10(v)
 	case OpARM64MOVHstorezero:
 		return rewriteValueARM64_OpARM64MOVHstorezero_0(v)
 	case OpARM64MOVQstorezero:
@@ -6221,6 +6221,53 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVBstore [i] {s} ptr0 (SRLconst [8] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w mem))
+	// cond: x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)
+	// result: (MOVHstore [i-1] {s} ptr0 w mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr0 := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpARM64SRLconst {
+			break
+		}
+		if v_1.AuxInt != 8 {
+			break
+		}
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		w := v_1_0.Args[0]
+		x := v.Args[2]
+		if x.Op != OpARM64MOVBstore {
+			break
+		}
+		if x.AuxInt != i-1 {
+			break
+		}
+		if x.Aux != s {
+			break
+		}
+		_ = x.Args[2]
+		ptr1 := x.Args[0]
+		if w != x.Args[1] {
+			break
+		}
+		mem := x.Args[2]
+		if !(x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)) {
+			break
+		}
+		v.reset(OpARM64MOVHstore)
+		v.AuxInt = i - 1
+		v.Aux = s
+		v.AddArg(ptr0)
+		v.AddArg(w)
+		v.AddArg(mem)
+		return true
+	}
 	// match: (MOVBstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] w) mem))
 	// cond: x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)
 	// result: (MOVHstore [i-1] {s} ptr0 w0 mem)
@@ -6325,6 +6372,62 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVBstore [i] {s} ptr0 (SRLconst [j] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] (MOVDreg w)) mem))
+	// cond: x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)
+	// result: (MOVHstore [i-1] {s} ptr0 w0 mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr0 := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpARM64SRLconst {
+			break
+		}
+		j := v_1.AuxInt
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		w := v_1_0.Args[0]
+		x := v.Args[2]
+		if x.Op != OpARM64MOVBstore {
+			break
+		}
+		if x.AuxInt != i-1 {
+			break
+		}
+		if x.Aux != s {
+			break
+		}
+		_ = x.Args[2]
+		ptr1 := x.Args[0]
+		w0 := x.Args[1]
+		if w0.Op != OpARM64SRLconst {
+			break
+		}
+		if w0.AuxInt != j-8 {
+			break
+		}
+		w0_0 := w0.Args[0]
+		if w0_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != w0_0.Args[0] {
+			break
+		}
+		mem := x.Args[2]
+		if !(x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)) {
+			break
+		}
+		v.reset(OpARM64MOVHstore)
+		v.AuxInt = i - 1
+		v.Aux = s
+		v.AddArg(ptr0)
+		v.AddArg(w0)
+		v.AddArg(mem)
+		return true
+	}
 	// match: (MOVBstore [i] {s} ptr w x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w) x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w) x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] w) x3:(MOVBstore [i-4] {s} ptr (SRLconst [32] w) x4:(MOVBstore [i-5] {s} ptr (SRLconst [40] w) x5:(MOVBstore [i-6] {s} ptr (SRLconst [48] w) x6:(MOVBstore [i-7] {s} ptr (SRLconst [56] w) mem))))))))
 	// cond: x0.Uses == 1 && x1.Uses == 1 && x2.Uses == 1 && x3.Uses == 1 && x4.Uses == 1 && x5.Uses == 1 && x6.Uses == 1 && clobber(x0) && clobber(x1) && clobber(x2) && clobber(x3) && clobber(x4) && clobber(x5) && clobber(x6)
 	// result: (MOVDstore [i-7] {s} ptr (REV <w.Type> w) mem)
@@ -6623,6 +6726,113 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVBstore [i] {s} ptr w x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] (MOVDreg w)) x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] (MOVDreg w)) mem))))
+	// cond: x0.Uses == 1 && x1.Uses == 1 && x2.Uses == 1 && clobber(x0) && clobber(x1) && clobber(x2)
+	// result: (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr := v.Args[0]
+		w := v.Args[1]
+		x0 := v.Args[2]
+		if x0.Op != OpARM64MOVBstore {
+			break
+		}
+		if x0.AuxInt != i-1 {
+			break
+		}
+		if x0.Aux != s {
+			break
+		}
+		_ = x0.Args[2]
+		if ptr != x0.Args[0] {
+			break
+		}
+		x0_1 := x0.Args[1]
+		if x0_1.Op != OpARM64SRLconst {
+			break
+		}
+		if x0_1.AuxInt != 8 {
+			break
+		}
+		x0_1_0 := x0_1.Args[0]
+		if x0_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != x0_1_0.Args[0] {
+			break
+		}
+		x1 := x0.Args[2]
+		if x1.Op != OpARM64MOVBstore {
+			break
+		}
+		if x1.AuxInt != i-2 {
+			break
+		}
+		if x1.Aux != s {
+			break
+		}
+		_ = x1.Args[2]
+		if ptr != x1.Args[0] {
+			break
+		}
+		x1_1 := x1.Args[1]
+		if x1_1.Op != OpARM64SRLconst {
+			break
+		}
+		if x1_1.AuxInt != 16 {
+			break
+		}
+		x1_1_0 := x1_1.Args[0]
+		if x1_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != x1_1_0.Args[0] {
+			break
+		}
+		x2 := x1.Args[2]
+		if x2.Op != OpARM64MOVBstore {
+			break
+		}
+		if x2.AuxInt != i-3 {
+			break
+		}
+		if x2.Aux != s {
+			break
+		}
+		_ = x2.Args[2]
+		if ptr != x2.Args[0] {
+			break
+		}
+		x2_1 := x2.Args[1]
+		if x2_1.Op != OpARM64SRLconst {
+			break
+		}
+		if x2_1.AuxInt != 24 {
+			break
+		}
+		x2_1_0 := x2_1.Args[0]
+		if x2_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != x2_1_0.Args[0] {
+			break
+		}
+		mem := x2.Args[2]
+		if !(x0.Uses == 1 && x1.Uses == 1 && x2.Uses == 1 && clobber(x0) && clobber(x1) && clobber(x2)) {
+			break
+		}
+		v.reset(OpARM64MOVWstore)
+		v.AuxInt = i - 3
+		v.Aux = s
+		v.AddArg(ptr)
+		v0 := b.NewValue0(v.Pos, OpARM64REVW, w.Type)
+		v0.AddArg(w)
+		v.AddArg(v0)
+		v.AddArg(mem)
+		return true
+	}
 	// match: (MOVBstore [i] {s} ptr w x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w) x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w) x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] w) mem))))
 	// cond: x0.Uses == 1 && x1.Uses == 1 && x2.Uses == 1 && clobber(x0) && clobber(x1) && clobber(x2)
 	// result: (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
@@ -6718,6 +6928,11 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	return false
+}
+func rewriteValueARM64_OpARM64MOVBstore_20(v *Value) bool {
+	b := v.Block
+	_ = b
 	// match: (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] w) mem))
 	// cond: x.Uses == 1 && clobber(x)
 	// result: (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
@@ -6816,6 +7031,57 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	// cond: x.Uses == 1 && clobber(x)
+	// result: (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr := v.Args[0]
+		w := v.Args[1]
+		x := v.Args[2]
+		if x.Op != OpARM64MOVBstore {
+			break
+		}
+		if x.AuxInt != i-1 {
+			break
+		}
+		if x.Aux != s {
+			break
+		}
+		_ = x.Args[2]
+		if ptr != x.Args[0] {
+			break
+		}
+		x_1 := x.Args[1]
+		if x_1.Op != OpARM64SRLconst {
+			break
+		}
+		if x_1.AuxInt != 8 {
+			break
+		}
+		x_1_0 := x_1.Args[0]
+		if x_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != x_1_0.Args[0] {
+			break
+		}
+		mem := x.Args[2]
+		if !(x.Uses == 1 && clobber(x)) {
+			break
+		}
+		v.reset(OpARM64MOVHstore)
+		v.AuxInt = i - 1
+		v.Aux = s
+		v.AddArg(ptr)
+		v0 := b.NewValue0(v.Pos, OpARM64REV16W, w.Type)
+		v0.AddArg(w)
+		v.AddArg(v0)
+		v.AddArg(mem)
+		return true
+	}
 	// match: (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVWUreg w)) mem))
 	// cond: x.Uses == 1 && clobber(x)
 	// result: (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
@@ -6867,6 +7133,57 @@ func rewriteValueARM64_OpARM64MOVBstore_10(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	// cond: x.Uses == 1 && clobber(x)
+	// result: (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr := v.Args[0]
+		w := v.Args[1]
+		x := v.Args[2]
+		if x.Op != OpARM64MOVBstore {
+			break
+		}
+		if x.AuxInt != i-1 {
+			break
+		}
+		if x.Aux != s {
+			break
+		}
+		_ = x.Args[2]
+		if ptr != x.Args[0] {
+			break
+		}
+		x_1 := x.Args[1]
+		if x_1.Op != OpARM64SRLconst {
+			break
+		}
+		if x_1.AuxInt != 8 {
+			break
+		}
+		x_1_0 := x_1.Args[0]
+		if x_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		if w != x_1_0.Args[0] {
+			break
+		}
+		mem := x.Args[2]
+		if !(x.Uses == 1 && clobber(x)) {
+			break
+		}
+		v.reset(OpARM64MOVHstore)
+		v.AuxInt = i - 1
+		v.Aux = s
+		v.AddArg(ptr)
+		v0 := b.NewValue0(v.Pos, OpARM64REV16W, w.Type)
+		v0.AddArg(w)
+		v.AddArg(v0)
+		v.AddArg(mem)
+		return true
+	}
 	return false
 }
 func rewriteValueARM64_OpARM64MOVBstorezero_0(v *Value) bool {
@@ -7821,6 +8138,56 @@ func rewriteValueARM64_OpARM64MOVHstore_0(v *Value) bool {
 		v.AddArg(mem)
 		return true
 	}
+	// match: (MOVHstore [i] {s} ptr0 (SRLconst [16] (MOVDreg w)) x:(MOVHstore [i-2] {s} ptr1 w mem))
+	// cond: x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)
+	// result: (MOVWstore [i-2] {s} ptr0 w mem)
+	for {
+		i := v.AuxInt
+		s := v.Aux
+		_ = v.Args[2]
+		ptr0 := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpARM64SRLconst {
+			break
+		}
+		if v_1.AuxInt != 16 {
+			break
+		}
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpARM64MOVDreg {
+			break
+		}
+		w := v_1_0.Args[0]
+		x := v.Args[2]
+		if x.Op != OpARM64MOVHstore {
+			break
+		}
+		if x.AuxInt != i-2 {
+			break
+		}
+		if x.Aux != s {
+			break
+		}
+		_ = x.Args[2]
+		ptr1 := x.Args[0]
+		if w != x.Args[1] {
+			break
+		}
+		mem := x.Args[2]
+		if !(x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)) {
+			break
+		}
+		v.reset(OpARM64MOVWstore)
+		v.AuxInt = i - 2
+		v.Aux = s
+		v.AddArg(ptr0)
+		v.AddArg(w)
+		v.AddArg(mem)
+		return true
+	}
+	return false
+}
+func rewriteValueARM64_OpARM64MOVHstore_10(v *Value) bool {
 	// match: (MOVHstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVHstore [i-2] {s} ptr1 w0:(SRLconst [j-16] w) mem))
 	// cond: x.Uses == 1 && isSamePtr(ptr0, ptr1) && clobber(x)
 	// result: (MOVWstore [i-2] {s} ptr0 w0 mem)

test/codegen/memcombine.go

@@ -110,28 +110,28 @@ func store_le64_idx(b []byte, idx int) {
 
 func store_le32(b []byte) {
 	// amd64:`MOVL\s`
-	// arm64(DISABLED):`MOVW`,-`MOV[BH]`
+	// arm64:`MOVW`,-`MOV[BH]`
 	// ppc64le:`MOVW\s`
 	binary.LittleEndian.PutUint32(b, sink32)
 }
 
 func store_le32_idx(b []byte, idx int) {
 	// amd64:`MOVL\s`
-	// arm64(DISABLED):`MOVW`,-`MOV[BH]`
+	// arm64:`MOVW`,-`MOV[BH]`
 	// ppc64le:`MOVW\s`
 	binary.LittleEndian.PutUint32(b[idx:], sink32)
 }
 
 func store_le16(b []byte) {
 	// amd64:`MOVW\s`
-	// arm64(DISABLED):`MOVH`,-`MOVB`
+	// arm64:`MOVH`,-`MOVB`
 	// ppc64le(DISABLED):`MOVH\s`
 	binary.LittleEndian.PutUint16(b, sink16)
 }
 
 func store_le16_idx(b []byte, idx int) {
 	// amd64:`MOVW\s`
-	// arm64(DISABLED):`MOVH`,-`MOVB`
+	// arm64:`MOVH`,-`MOVB`
 	// ppc64le(DISABLED):`MOVH\s`
 	binary.LittleEndian.PutUint16(b[idx:], sink16)
 }
@@ -150,24 +150,24 @@ func store_be64_idx(b []byte, idx int) {
 
 func store_be32(b []byte) {
 	// amd64:`BSWAPL`,-`SHR.`
-	// arm64(DISABLED):`MOVW`,`REVW`,-`MOV[BH]`
+	// arm64:`MOVW`,`REVW`,-`MOV[BH]`
 	binary.BigEndian.PutUint32(b, sink32)
 }
 
 func store_be32_idx(b []byte, idx int) {
 	// amd64:`BSWAPL`,-`SHR.`
-	// arm64(DISABLED):`MOVW`,`REVW`,-`MOV[BH]`
+	// arm64:`MOVW`,`REVW`,-`MOV[BH]`
 	binary.BigEndian.PutUint32(b[idx:], sink32)
 }
 
 func store_be16(b []byte) {
 	// amd64:`ROLW\s\$8`,-`SHR.`
-	// arm64(DISABLED):`MOVH`,`REV16W`,-`MOVB`
+	// arm64:`MOVH`,`REV16W`,-`MOVB`
 	binary.BigEndian.PutUint16(b, sink16)
 }
 
 func store_be16_idx(b []byte, idx int) {
 	// amd64:`ROLW\s\$8`,-`SHR.`
-	// arm64(DISABLED):`MOVH`,`REV16W`,-`MOVB`
+	// arm64:`MOVH`,`REV16W`,-`MOVB`
 	binary.BigEndian.PutUint16(b[idx:], sink16)
 }