diff --git a/sign/internal/dilithium/arm64.go b/sign/internal/dilithium/arm64.go
index dedcb596..83f1e867 100644
--- a/sign/internal/dilithium/arm64.go
+++ b/sign/internal/dilithium/arm64.go
@@ -81,7 +81,7 @@ func (p *Poly) Exceeds(bound uint32) bool {
 //
 // So it requires the coefficients of p  to be less than 2³²⁻ᴰ.
 func (p *Poly) MulBy2toD(q *Poly) {
-	p.mulBy2toDGeneric(q)
+	polyMulBy2toDARM64(p, q)
 }
 
 // Splits p into p1 and p0 such that [i]p1 * 2ᴰ + [i]p0 = [i]p
@@ -98,3 +98,6 @@ func polyAddARM64(p, a, b *Poly)
 
 //go:noescape
 func polyPackLe16ARM64(p *Poly, buf *byte)
+
+//go:noescape
+func polyMulBy2toDARM64(p, q *Poly)
diff --git a/sign/internal/dilithium/arm64.s b/sign/internal/dilithium/arm64.s
index f609e9a0..47eb7958 100644
--- a/sign/internal/dilithium/arm64.s
+++ b/sign/internal/dilithium/arm64.s
@@ -69,3 +69,25 @@ loop:
     BGT     loop
 
     RET
+
+// func polyMulBy2toDARM64(p, q *Poly)
+TEXT ·polyMulBy2toDARM64(SB), NOSPLIT|NOFRAME, $0-16
+    MOVD    p+0(FP), R0
+    MOVD    q+8(FP), R1
+
+    MOVW    $(const_N / 16), R2
+
+loop:
+    VLD1.P  (64)(R1), [V0.S4, V1.S4, V2.S4, V3.S4]
+
+    VSHL    $(const_D), V0.S4, V0.S4
+    VSHL    $(const_D), V1.S4, V1.S4
+    VSHL    $(const_D), V2.S4, V2.S4
+    VSHL    $(const_D), V3.S4, V3.S4
+
+    VST1.P  [V0.S4, V1.S4, V2.S4, V3.S4], (64)(R0)
+
+    SUBS    $1, R2, R2
+    BGT     loop
+
+    RET