diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 6df5405..dc086b7 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -337,6 +337,7 @@ void putX_X_XM_IMM()
 		{ 0x4F, "vrsqrt14ss", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, false },
 
 		{ 0x4F, "vrsqrtsh", T_66 | T_MAP6 | T_MUST_EVEX | T_EW0 | T_N2, false },
+		{ 0x51, "vsqrtsh", T_F3 | T_MAP5 | T_MUST_EVEX | T_EW0 | T_ER_X | T_N2, false },
 
 		{ 0x0B, "vrndscalesd", T_66 | T_0F3A | T_MUST_EVEX | T_EW1 | T_N8, true },
 		{ 0x0A, "vrndscaless", T_66 | T_0F3A | T_MUST_EVEX | T_EW0 | T_N4, true },
diff --git a/test/misc.cpp b/test/misc.cpp
index bb002ea..a435892 100644
--- a/test/misc.cpp
+++ b/test/misc.cpp
@@ -991,6 +991,9 @@ CYBOZU_TEST_AUTO(vaddph)
 			vsqrtph(ymm1|k4|T_z, ptr_b [rax+0x40]);
 			vsqrtph(zmm1|k4|T_z, ptr [rax+0x40]);
 			vsqrtph(zmm1|k4|T_z, ptr_b [rax+0x40]);
+
+			vsqrtsh(xmm1|k4|T_z, xmm5, ptr [rax+0x40]);
+			vsqrtsh(xmm1|k4|T_z|T_rd_sae, xmm5, xmm7);
 		}
 	} c;
 	const uint8_t tbl[] = {
@@ -1139,6 +1142,10 @@ CYBOZU_TEST_AUTO(vaddph)
 		0x62, 0xf5, 0x7c, 0xbc, 0x51, 0x48, 0x20,
 		0x62, 0xf5, 0x7c, 0xcc, 0x51, 0x48, 0x01,
 		0x62, 0xf5, 0x7c, 0xdc, 0x51, 0x48, 0x20,
+
+		// vsqrtsh
+		0x62, 0xf5, 0x56, 0x8c, 0x51, 0x48, 0x20,
+		0x62, 0xf5, 0x56, 0xbc, 0x51, 0xcf,
 	};
 	const size_t n = sizeof(tbl) / sizeof(tbl[0]);
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index c45db3c..5aac280 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -2224,6 +2224,7 @@ void vshuff64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) {
 void vshufi32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, 0x43, imm); }
 void vshufi64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x43, imm); }
 void vsqrtph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_MAP5 | T_EW0 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B16, 0x51); }
+void vsqrtsh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2 | T_F3 | T_MAP5 | T_EW0 | T_ER_X | T_MUST_EVEX, 0x51); }
 void vsubph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_ER_Z | T_B16, 0x5C); }
 void vsubsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_ER_X | T_N2, 0x5C); }
 void vucomish(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_MAP5 | T_MUST_EVEX | T_EW0 | T_SAE_X | T_N2, 0x2E); }