diff --git a/src/math/all_test.go b/src/math/all_test.go
index 39b5b33071..a145c5e60b 100644
--- a/src/math/all_test.go
+++ b/src/math/all_test.go
@@ -1014,6 +1014,8 @@ var vfexpSC = []float64{
 	1,
 	// near zero
 	3.725290298461915e-09,
+	// denormal
+	-740,
 }
 var expSC = []float64{
 	0,
@@ -1026,6 +1028,7 @@ var expSC = []float64{
 	Inf(1),
 	2.718281828459045,
 	1.0000000037252903,
+	4.2e-322,
 }
 
 var vfexp2SC = []float64{
diff --git a/src/math/exp_amd64.s b/src/math/exp_amd64.s
index 2f7fd76fe4..1c2ac8f676 100644
--- a/src/math/exp_amd64.s
+++ b/src/math/exp_amd64.s
@@ -96,14 +96,13 @@ TEXT ·Exp(SB),NOSPLIT,$0
 	MULSD   X1, X0
 	ADDSD exprodata<>+8(SB), X0
 	// return fr * 2**exponent
-lastStep:
-	MOVL    $0x3FF, AX // bias
-	ADDL    AX, BX
-	JLE     underflow
+ldexp:
+	ADDL    $0x3FF, BX // add bias
+	JLE     denormal
 	CMPL    BX, $0x7FF
 	JGE     overflow
-	MOVL    $52, CX
-	SHLQ    CX, BX
+lastStep:
+	SHLQ    $52, BX
 	MOVQ    BX, X1
 	MULSD   X1, X0
 	MOVSD   X0, ret+8(FP)
@@ -115,14 +114,22 @@ notFinite:
 	JNE     notNegInf
 	// -Inf, return 0
 underflow: // return 0
-	MOVQ    $0, AX
-	MOVQ    AX, ret+8(FP)
+	MOVQ    $0, ret+8(FP)
 	RET
 overflow: // return +Inf
 	MOVQ    $PosInf, BX
 notNegInf: // NaN or +Inf, return x
 	MOVQ    BX, ret+8(FP)
 	RET
+denormal:
+	CMPL    BX, $-52
+	JL      underflow
+	ADDL    $0x3FE, BX // add bias - 1
+	SHLQ    $52, BX
+	MOVQ    BX, X1
+	MULSD   X1, X0
+	MOVQ    $1, BX
+	JMP     lastStep
 
 avxfma:
 	MOVSD   $LN2U, X2
@@ -149,4 +156,4 @@ avxfma:
 	MULSD   X1, X0
 	VADDSD exprodata<>+16(SB), X0, X1
 	VFMADD213SD   exprodata<>+8(SB), X1, X0
-	JMP lastStep
+	JMP ldexp