diff --git a/asm/f64/add_amd64.s b/asm/f64/add_amd64.s
index a23c7436..6568765d 100644
--- a/asm/f64/add_amd64.s
+++ b/asm/f64/add_amd64.s
@@ -9,15 +9,28 @@
 // func Add(dst, s []float64)
 TEXT ·Add(SB), NOSPLIT, $0
 	MOVQ    dst_base+0(FP), DI
-	MOVQ    dst_len+8(FP), DX
+	MOVQ    dst_len+8(FP), CX
 	MOVQ    s_base+24(FP), SI
-	CMPQ    s_len+32(FP), DX
-	CMOVQLE s_len+32(FP), DX
-	CMPQ    DX, $0
+	CMPQ    s_len+32(FP), CX
+	CMOVQLE s_len+32(FP), CX
+	CMPQ    CX, $0
 	JE      add_end
 	XORQ    AX, AX
-	CMPQ    DX, $4
-	JL      add_tail
+	MOVQ    DI, BX
+	ANDQ    $15, BX
+	JZ      add_no_trim
+	MOVSD   (DI)(AX*8), X0
+	ADDSD   (SI)(AX*8), X0
+	MOVSD   X0, (DI)(AX*8)
+	INCQ    AX
+	DECQ    CX
+	JE      add_end
+
+add_no_trim:
+	MOVQ CX, BX
+	ANDQ $3, BX
+	SHRQ $2, CX
+	JZ   add_tail_start
 
 add_loop:
 	MOVUPS (SI)(AX*8), X0
@@ -27,19 +40,19 @@ add_loop:
 	ADDPD  16(DI)(AX*8), X1
 	MOVUPS X1, 16(DI)(AX*8)
 	ADDQ   $4, AX
-	SUBQ   $4, DX
-	CMPQ   DX, $4
-	JGE    add_loop
-	CMPQ   DX, $0
+	LOOPNE add_loop
+	CMPQ   BX, $0
 	JE     add_end
 
+add_tail_start:
+	MOVQ BX, CX
+
 add_tail:
-	MOVSD (DI)(AX*8), X0
-	ADDSD (SI)(AX*8), X0
-	MOVSD X0, (DI)(AX*8)
-	INCQ  AX
-	DECQ  DX
-	JNZ   add_tail
+	MOVSD  (DI)(AX*8), X0
+	ADDSD  (SI)(AX*8), X0
+	MOVSD  X0, (DI)(AX*8)
+	INCQ   AX
+	LOOPNE add_tail
 
 add_end:
 	RET
diff --git a/asm/f64/stubs_test.go b/asm/f64/stubs_test.go
index 3c6c8602..17890412 100644
--- a/asm/f64/stubs_test.go
+++ b/asm/f64/stubs_test.go
@@ -32,6 +32,9 @@ func TestAdd(t *testing.T) {
 		{[]float64{0, 1, 2, 3, 4},
 			[]float64{-inf, 4, nan, 8, 9},
 			[]float64{-inf, 5, nan, 11, 13}},
+		{make([]float64, 50)[1:49],
+			make([]float64, 50)[1:49],
+			make([]float64, 50)[1:49]},
 	} {
 		Add(v.dst, v.src)
 		for i := range v.expect {
@@ -70,7 +73,7 @@ func TestCumSum(t *testing.T) {
 	}{
 		{[]float64{0}, []float64{1}, []float64{1}},
 		{[]float64{nan}, []float64{nan}, []float64{nan}},
-		{[]float64{0, 0, 0}, []float64{1, 2, 3, 4}, []float64{1, 3, 6}},
+		{[]float64{0, 0, 0}, []float64{1, 2, 3}, []float64{1, 3, 6}},
 		{[]float64{0, 0, 0, 0}, []float64{1, 2, 3}, []float64{1, 3, 6}},
 		{[]float64{0, 0, 0, 0}, []float64{1, 2, 3, 4}, []float64{1, 3, 6, 10}},
 		{[]float64{1, nan, nan, 1, 1},