diff --git a/include/mcl/fp.hpp b/include/mcl/fp.hpp
index ef93acc..2d7d58a 100644
--- a/include/mcl/fp.hpp
+++ b/include/mcl/fp.hpp
@@ -66,6 +66,7 @@ public:
 	const Unit *getUnit() const { return v_; }
 	static inline size_t getUnitSize() { return op_.N; }
 	static inline size_t getBitSize() { return op_.bitSize; }
+	static inline const fp::Op& getOp() { return op_; }
 	void dump() const
 	{
 		const size_t N = op_.N;
diff --git a/include/mcl/fp_generator.hpp b/include/mcl/fp_generator.hpp
index 559e253..a896278 100644
--- a/include/mcl/fp_generator.hpp
+++ b/include/mcl/fp_generator.hpp
@@ -161,14 +161,11 @@ struct FpGenerator : Xbyak::CodeGenerator {
 	typedef int (*int2op)(uint64_t*, const uint64_t*);
 	bool3op addNC_;
 	bool3op subNC_;
-	void3op add_;
-	void3op sub_;
-	void3op mul_;
+	void3u mul_;
 	uint3opI mulI_;
 	void *montRedRaw_;
 	void2op montRed_;
 	void2op sqr_;
-	void2op neg_;
 	void2op shr1_;
 	int2op preInv_;
 	FpGenerator()
@@ -179,28 +176,24 @@ struct FpGenerator : Xbyak::CodeGenerator {
 		, isFullBit_(0)
 		, addNC_(0)
 		, subNC_(0)
-		, add_(0)
-		, sub_(0)
 		, mul_(0)
 		, mulI_(0)
 		, montRedRaw_(0)
 		, montRed_(0)
-		, neg_(0)
 		, shr1_(0)
 		, preInv_(0)
 	{
 		useMulx_ = cpu_.has(Xbyak::util::Cpu::tBMI2);
 	}
 	/*
-		@param p [in] pointer to prime
-		@param pn [in] length of prime
+		@param op [in] ; use op.p and op.N
 	*/
-	void init(Op& op, const uint64_t *p, int pn)
+	void init(Op& op)
 	{
-		if (pn < 2) throw cybozu::Exception("mcl:FpGenerator:small pn") << pn;
-		p_ = p;
-		rp_ = fp::getMontgomeryCoeff(p[0]);
-		pn_ = pn;
+		if (op.N < 2) throw cybozu::Exception("mcl:FpGenerator:small pn") << op.N;
+		p_ = op.p;
+		rp_ = fp::getMontgomeryCoeff(p_[0]);
+		pn_ = (int)op.N;
 		isFullBit_ = (p_[pn_ - 1] >> 63) != 0;
 //		printf("p=%p, pn_=%d, isFullBit_=%d\n", p_, pn_, isFullBit_);
 
@@ -212,19 +205,20 @@ struct FpGenerator : Xbyak::CodeGenerator {
 		subNC_ = getCurr<bool3op>();
 		gen_addSubNC(false);
 		align(16);
-		add_ = getCurr<void3op>();
+		op.fp_add = getCurr<void3u>();
 		gen_addMod();
 		align(16);
-		sub_ = getCurr<void3op>();
+		op.fp_sub = getCurr<void3u>();
 		gen_sub();
 		align(16);
-		neg_ = getCurr<void2op>();
+		op.fp_neg = getCurr<void2u>();
 		gen_neg();
 		align(16);
 		mulI_ = getCurr<uint3opI>();
 		gen_mulI();
 		align(16);
-		mul_ = getCurr<void3op>();
+		mul_ = getCurr<void3u>();
+		op.fp_mul = mul_;
 		gen_mul();
 		align(16);
 		montRed_ = getCurr<void2op>();
@@ -240,15 +234,11 @@ struct FpGenerator : Xbyak::CodeGenerator {
 		preInv_ = getCurr<int2op>();
 		gen_preInv();
 
-		op.fp_neg = Xbyak::CastTo<void2u>(neg_);
-		op.fp_add = Xbyak::CastTo<void3u>(add_);
-		op.fp_sub = Xbyak::CastTo<void3u>(sub_);
 		op.fp_addNC = Xbyak::CastTo<void3u>(addNC_);
 		op.fp_subNC = Xbyak::CastTo<void3u>(subNC_);
-		op.fp_mul = Xbyak::CastTo<void3u>(mul_);
 		op.fp_sqr = Xbyak::CastTo<void2u>(sqr_);
 
-		if (pn <= 4) {
+		if (op.N <= 4) {
 			if (montRed_) {
 				op.fp_mod = Xbyak::CastTo<void2u>(montRed_);
 			}
diff --git a/src/fp.cpp b/src/fp.cpp
index f3b1ef6..0142ec7 100644
--- a/src/fp.cpp
+++ b/src/fp.cpp
@@ -363,7 +363,7 @@ static void initForMont(Op& op, const Unit *p, Mode mode)
 #ifdef MCL_USE_XBYAK
 	FpGenerator *fg = op.fg;
 	if (fg == 0) return;
-	fg->init(op, p, (int)N);
+	fg->init(op);
 
 	if (N <= 4) {
 		op.fp_invOp = &invOpForMontC;
diff --git a/test/fp_generator_test.cpp b/test/fp_generator_test.cpp
index 5930727..1143881 100644
--- a/test/fp_generator_test.cpp
+++ b/test/fp_generator_test.cpp
@@ -93,10 +93,10 @@ static inline std::ostream& operator<<(std::ostream& os, const Int& x)
 	return os << x.getStr();
 }
 
-void testAddSub(const mcl::fp::FpGenerator& fg, int pn)
+void testAddSub(const mcl::fp::Op& op)
 {
 	Fp x, y;
-	Int mx(pn), my(pn);
+	Int mx(op.N), my(op.N);
 	x.setStr("0x8811aabb23427cc");
 	y.setStr("0x8811aabb23427cc11");
 	mx.set(x);
@@ -104,24 +104,24 @@ void testAddSub(const mcl::fp::FpGenerator& fg, int pn)
 	for (int i = 0; i < 30; i++) {
 		CYBOZU_TEST_EQUAL(mx, x);
 		x += x;
-		fg.add_(mx.v, mx.v, mx.v);
+		op.fp_add(mx.v, mx.v, mx.v);
 	}
 	for (int i = 0; i < 30; i++) {
 		CYBOZU_TEST_EQUAL(mx, x);
 		x += y;
-		fg.add_(mx.v, mx.v, my.v);
+		op.fp_add(mx.v, mx.v, my.v);
 	}
 	for (int i = 0; i < 30; i++) {
 		CYBOZU_TEST_EQUAL(my, y);
 		y -= x;
-		fg.sub_(my.v, my.v, mx.v);
+		op.fp_sub(my.v, my.v, mx.v);
 	}
 }
 
-void testNeg(const mcl::fp::FpGenerator& fg, int pn)
+void testNeg(const mcl::fp::Op& op)
 {
 	Fp x;
-	Int mx(pn), my(pn);
+	Int mx(op.N), my(op.N);
 	const char *tbl[] = {
 		"0",
 		"0x12346",
@@ -132,7 +132,7 @@ void testNeg(const mcl::fp::FpGenerator& fg, int pn)
 		x.setStr(tbl[i]);
 		mx.set(x);
 		x = -x;
-		fg.neg_(mx.v, mx.v);
+		op.fp_neg(mx.v, mx.v);
 		CYBOZU_TEST_EQUAL(mx, x);
 	}
 }
@@ -183,18 +183,13 @@ void testShr1(const mcl::fp::FpGenerator& fg, int pn)
 
 void test(const char *pStr)
 {
-	Fp::setModulo(pStr, 16);
-	uint64_t p[MAX_N];
-	strToArray(p, MAX_N, pStr);
-	const int pn = mcl::fp::getNonZeroArraySize(p, MAX_N);
-	printf("pn=%d\n", pn);
-	mcl::fp::FpGenerator fg;
-	mcl::fp::Op op;
-	fg.init(op, p, pn);
-	testAddSub(fg, pn);
-	testNeg(fg, pn);
-	testMulI(fg, pn);
-	testShr1(fg, pn);
+	Fp::setModulo(pStr, 16, mcl::fp::FP_XBYAK);
+	const mcl::fp::Op& op = Fp::getOp();
+	const int pn = (int)op.N;
+	testAddSub(op);
+	testNeg(op);
+	testMulI(*op.fg, pn);
+	testShr1(*op.fg, pn);
 }
 
 CYBOZU_TEST_AUTO(all)