a little optimize Fp6::mul

include/mcl/fp_tower.hpp

@@ -978,6 +978,7 @@ template<class Fp>
 struct Fp6DblT {
 	typedef Fp2T<Fp> Fp2;
 	typedef Fp6T<Fp> Fp6;
+	typedef FpDblT<Fp> FpDbl;
 	typedef Fp2DblT<Fp> Fp2Dbl;
 	typedef Fp6DblT<Fp> Fp6Dbl;
 	typedef fp::Unit Unit;
@@ -994,6 +995,11 @@ struct Fp6DblT {
 		Fp2Dbl::sub(z.b, x.b, y.b);
 		Fp2Dbl::sub(z.c, x.c, y.c);
 	}
+	static void sub2(Fp2Dbl& y, const Fp2Dbl& x)
+	{
+		FpDbl::sub(y.a, y.a, x.a);
+		FpDbl::subPre(y.b, y.b, x.b);
+	}
 	/*
 		x = a + bv + cv^2, y = d + ev + fv^2, v^3 = xi
 		xy = (ad + (bf + ce)xi) + ((ae + bd) + cf xi)v + ((af + cd) + be)v^2
@@ -1010,7 +1016,7 @@ struct Fp6DblT {
 		const Fp2& d = y.a;
 		const Fp2& e = y.b;
 		const Fp2& f = y.c;
-#if 0
+#if 1
 		Fp2Dbl& ZA = z.a;
 		Fp2Dbl& ZB = z.b;
 		Fp2Dbl& ZC = z.c;
@@ -1028,15 +1034,12 @@ struct Fp6DblT {
 		Fp2Dbl::mulPre(BE, b, e);
 		Fp2Dbl::mulPre(CF, c, f);
 		Fp2Dbl::mulPre(AD, a, d);
-		Fp2Dbl::sub(ZA, ZA, BE);
-		Fp2Dbl::sub(ZA, ZA, CF);
-//		Fp2Dbl::sub_p_if_possible(ZA, ZA);
-		Fp2Dbl::sub(ZB, ZB, AD);
-		Fp2Dbl::sub(ZB, ZB, BE);
-//		Fp2Dbl::sub_p_if_possible(ZB, ZB);
-		Fp2Dbl::sub(ZC, ZC, AD);
-		Fp2Dbl::sub(ZC, ZC, CF);
-//		Fp2Dbl::sub_p_if_possible(ZC, ZC);
+		sub2(ZA, BE);
+		sub2(ZA, CF);
+		sub2(ZB, AD);
+		sub2(ZB, BE);
+		sub2(ZC, AD);
+		sub2(ZC, CF);
 		Fp2Dbl::mul_xi(ZA, ZA);
 		Fp2Dbl::add(ZA, ZA, AD);
 		Fp2Dbl::mul_xi(CF, CF);

test/common_test.hpp

@@ -129,14 +129,21 @@ void testMul2()
 void testABCDsub(const Fp2& a, const Fp2& b, const Fp2& c, const Fp2& d)
 {
 	Fp2 t1, t2;
-	Fp2::add(t1, a, b);
-	Fp2::add(t2, c, d);
+	Fp2::addPre(t1, a, b);
+	Fp2::addPre(t2, c, d);
 	Fp2Dbl T1, AC, BD;
 	Fp2Dbl::mulPre(T1, t1, t2);
 	Fp2Dbl::mulPre(AC, a, c);
 	Fp2Dbl::mulPre(BD, b, d);
+#if 0
 	Fp2Dbl::sub(T1, T1, AC);
 	Fp2Dbl::sub(T1, T1, BD);
+#else
+	FpDbl::sub(T1.a, T1.a, AC.a);
+	FpDbl::subPre(T1.b, T1.b, AC.b);
+	FpDbl::sub(T1.a, T1.a, BD.a);
+	FpDbl::subPre(T1.b, T1.b, BD.b);
+#endif
 	Fp2Dbl::mod(t1, T1);
 	CYBOZU_TEST_EQUAL(t1, a * d + b * c);
 }
@@ -145,13 +152,20 @@ void testABCD()
 {
 	puts("testMisc1");
 	// (a + b)(c + d) - ac - bd = ad + bc
-	Fp2 a, b, c, d;
-	a.a = -1;
-	a.b = -1;
-	b = a;
-	c = a;
-	d = a;
-	testABCDsub(a, b, c, d);
+	Fp2 a[4];
+	a[0].a = -1;
+	a[0].b = -1;
+	a[1] = a[0];
+	a[2] = a[0];
+	a[3] = a[0];
+	testABCDsub(a[0], a[1], a[2], a[3]);
+	for (int i = 0; i < 100; i++) {
+		for (int j = 0; j < 4; j++) {
+			a[j].a.setByCSPRNG();
+			a[j].b.setByCSPRNG();
+		}
+		testABCDsub(a[0], a[1], a[2], a[3]);
+	}
 }
 
 void testCommon(const G1& P, const G2& Q)