add fp2_mul_xi for xi_a = 1

include/mcl/fp_tower.hpp

@@ -242,9 +242,9 @@ public:
 	static void (*neg)(Fp2T& y, const Fp2T& x);
 	static void (*mul)(Fp2T& z, const Fp2T& x, const Fp2T& y);
 	static void (*sqr)(Fp2T& y, const Fp2T& x);
+	static void (*mul_xi)(Fp2T& y, const Fp2T& x);
 	static void addPre(Fp2T& z, const Fp2T& x, const Fp2T& y) { Fp::addPre(z.a, x.a, y.a); Fp::addPre(z.b, x.b, y.b); }
 	static void inv(Fp2T& y, const Fp2T& x) { Fp::op_.fp2_inv(y.a.v_, x.a.v_); }
-	static void mul_xi(Fp2T& y, const Fp2T& x) { Fp::op_.fp2_mul_xi(y.a.v_, x.a.v_); }
 	static void divBy2(Fp2T& y, const Fp2T& x)
 	{
 		Fp::divBy2(y.a, x.a);
@@ -405,11 +405,21 @@ public:
 		sqr = (void (*)(Fp2T& y, const Fp2T& x))op.fp2_sqrA_;
 		if (sqr == 0) sqr = fp2_sqrC;
 		op.fp2_inv = fp2_invW;
-		if (xi_a == 1) {
-			op.fp2_mul_xi = fp2_mul_xi_1_1i;
-		} else {
-			op.fp2_mul_xi = fp2_mul_xiW;
+		if (op.fp2_mul_xi == 0) {
+			if (xi_a == 1) {
+				/*
+					current fp_generator.hpp generates mul_xi for xi_a = 1
+				*/
+				if (op.fp2_mul_xiA_) {
+					op.fp2_mul_xi = op.fp2_mul_xiA_;
+				} else {
+					op.fp2_mul_xi = fp2_mul_xi_1_1i;
+				}
+			} else {
+				op.fp2_mul_xi = fp2_mul_xiW;
+			}
 		}
+		mul_xi = (void (*)(Fp2T&, const Fp2T&))op.fp2_mul_xi;
 		const Fp2T xi(xi_a, 1);
 		const mpz_class& p = Fp::getOp().mp;
 		Fp2T::pow(g[0], xi, (p - 1) / 6); // g = xi^((p-1)/6)
@@ -621,6 +631,7 @@ template<class Fp_> void (*Fp2T<Fp_>::sub)(Fp2T& z, const Fp2T& x, const Fp2T& y
 template<class Fp_> void (*Fp2T<Fp_>::neg)(Fp2T& y, const Fp2T& x);
 template<class Fp_> void (*Fp2T<Fp_>::mul)(Fp2T& z, const Fp2T& x, const Fp2T& y);
 template<class Fp_> void (*Fp2T<Fp_>::sqr)(Fp2T& y, const Fp2T& x);
+template<class Fp_> void (*Fp2T<Fp_>::mul_xi)(Fp2T& y, const Fp2T& x);
 
 template<class Fp>
 struct Fp2DblT {

include/mcl/op.hpp

@@ -227,6 +227,7 @@ struct Op {
 	void4u fp2_mulNF;
 	void2u fp2_inv;
 	void2u fp2_mul_xi;
+	void2u fp2_mul_xiA_;
 	uint32_t (*hash)(void *out, uint32_t maxOutSize, const void *msg, uint32_t msgSize);
 
 	PrimeMode primeMode;
@@ -306,6 +307,7 @@ struct Op {
 		fp2_mulNF = 0;
 		fp2_inv = 0;
 		fp2_mul_xi = 0;
+		fp2_mul_xiA_ = 0;
 
 		primeMode = PM_GENERIC;
 		isFullBit = false;

src/fp_generator.hpp

@@ -386,6 +386,9 @@ struct Code : Xbyak::CodeGenerator {
 			align(16);
 			op.fp2_sqrA_ = getCurr<void2u>();
 			gen_fp2_sqr4();
+			align(16);
+			op.fp2_mul_xiA_ = getCurr<void2u>();
+			gen_fp2_mul_xi4();
 		}
 	}
 	void gen_addSubPre(bool isAdd, int n)
@@ -2873,6 +2876,20 @@ private:
 		gen_raw_fp_sub(sf.p[0], sf.p[1], sf.p[2], sf.t, false);
 		gen_raw_fp_sub(sf.p[0] + FpByte_, sf.p[1] + FpByte_, sf.p[2] + FpByte_, sf.t, false);
 	}
+	/*
+		for only xi_a = 1
+	*/
+	void gen_fp2_mul_xi4()
+	{
+		assert(!isFullBit_);
+		StackFrame sf(this, 2, 8, 8 * 4);
+		gen_raw_fp_add(rsp, sf.p[1], sf.p[1] + FpByte_, sf.t, false);
+		gen_raw_fp_sub(sf.p[0], sf.p[1], sf.p[1] + FpByte_, sf.t, false);
+		for (int i = 0; i < 4; i++) {
+			mov(rax, ptr [rsp + i * 8]);
+			mov(ptr[sf.p[0] + FpByte_ + i * 8], rax);
+		}
+	}
 	void gen_fp2_neg4()
 	{
 		assert(!isFullBit_);