fix sqr

include/mcl/fp.hpp

@@ -98,6 +98,7 @@ public:
 		op_.useMont = mode == fp::FP_LLVM_MONT || mode == fp::FP_XBYAK;
 		if (mode == fp::FP_LLVM_MONT) {
 			op_.mul = montW;
+			op_.sqr = montSqrW;
 		}
 #if 0
 	fprintf(stderr, "mode=%d, useMont=%d"
@@ -278,7 +279,7 @@ public:
 	static inline void mul(FpT& z, const FpT& x, const FpT& y) { op_.mul(z.v_, x.v_, y.v_); }
 	static inline void inv(FpT& y, const FpT& x) { op_.invOp(y.v_, x.v_, op_); }
 	static inline void neg(FpT& y, const FpT& x) { op_.neg(y.v_, x.v_); }
-	static inline void square(FpT& y, const FpT& x) { op_.mul(y.v_, x.v_, x.v_); }
+	static inline void square(FpT& y, const FpT& x) { op_.sqr(y.v_, x.v_); }
 	static inline void div(FpT& z, const FpT& x, const FpT& y)
 	{
 		FpT rev;
@@ -433,10 +434,9 @@ public:
 	}
 	static inline void sqrW(Unit *y, const Unit *x)
 	{
-//		Unit xx[maxSize * 2];
-//		op_.sqrPreP(xx, x);
-//		op_.modP(y, xx, op_.p);
-		mulW(y, x, x);
+		Unit xx[maxSize * 2];
+		op_.sqrPreP(xx, x);
+		op_.modP(y, xx, op_.p);
 	}
 	static inline void negW(Unit *y, const Unit *x)
 	{
@@ -447,6 +447,10 @@ public:
 	{
 		op_.mont(z, x, y, op_.p, op_.rp);
 	}
+	static inline void montSqrW(Unit *y, const Unit *x)
+	{
+		op_.mont(y, x, x, op_.p, op_.rp);
+	}
 	void normalize() {} // dummy method
 };
 

include/mcl/fp_generator.hpp

@@ -219,9 +219,7 @@ struct FpGenerator : Xbyak::CodeGenerator {
 		gen_mul();
 		align(16);
 		sqr_ = getCurr<void2op>();
-		if (!gen_sqr()) {
-			sqr_ = 0;
-		}
+		gen_sqr();
 		align(16);
 		shr1_ = getCurr<void2op>();
 		gen_shr1();
@@ -547,13 +545,19 @@ struct FpGenerator : Xbyak::CodeGenerator {
 			throw cybozu::Exception("mcl:FpGenerator:gen_mul:not implemented for") << pn_;
 		}
 	}
-	bool gen_sqr()
+	void gen_sqr()
 	{
 		if (pn_ == 3) {
 			gen_montSqr3(p_, pp_);
-			return true;
+			return;
 		}
-		return false;
+		// sqr(y, x) = mul(y, x, x)
+#ifdef XBYAK_WIN
+		mov(r8, rdx);
+#else
+		mov(rdx, rsi);
+#endif
+		jmp((void*)mul_);
 	}
 	/*
 		input (pz[], px[], py[])

sample/bench.cpp

@@ -27,12 +27,13 @@ void benchFpSub(const char *pStr, const char *xStr, const char *yStr, mcl::fp::M
 	Fp x(xStr);
 	Fp y(yStr);
 
-	double addT, subT, mulT, invT;
+	double addT, subT, mulT, sqrT, invT;
 	CYBOZU_BENCH_T(addT, Fp::add, x, x, x);
 	CYBOZU_BENCH_T(subT, Fp::sub, x, x, y);
 	CYBOZU_BENCH_T(mulT, Fp::mul, x, x, x);
+	CYBOZU_BENCH_T(sqrT, Fp::square, x, x);
 	CYBOZU_BENCH_T(invT, x += y;Fp::inv, x, x); // avoid same jmp
-	printf("%10s bit % 3d add %8.2f sub %8.2f mul %8.2f inv %8.2f\n", s, (int)Fp::getBitSize(), addT, subT, mulT, invT);
+	printf("%10s bit % 3d add %8.2f sub %8.2f mul %8.2f sqr %8.2f inv %8.2f\n", s, (int)Fp::getBitSize(), addT, subT, mulT, sqrT, invT);
 }
 
 void benchFp(size_t bitSize, int mode)

src/fp.cpp

@@ -133,12 +133,11 @@ struct OpeFunc {
 	}
 	static inline void sqrPreC(Unit *y, const Unit *x)
 	{
-//		mpz_t mx, my;
-//		set_zero(my, y, N * 2);
-//		set_mpz_t(mx, x);
-//		mpz_mul(my, mx, mx);
-//		Gmp::getArray(y, N * 2, my);
-		mulPreC(y, x, x);
+		mpz_t mx, my;
+		set_zero(my, y, N * 2);
+		set_mpz_t(mx, x);
+		mpz_mul(my, mx, mx);
+		Gmp::getArray(y, N * 2, my);
 	}
 	// x[N * 2] -> y[N]
 	static inline void modC(Unit *y, const Unit *x, const Unit *p)
@@ -270,9 +269,7 @@ static void initForMont(Op& op, const Unit *p, Mode mode)
 	op.add = Xbyak::CastTo<void3u>(fg->add_);
 	op.sub = Xbyak::CastTo<void3u>(fg->sub_);
 	op.mul = Xbyak::CastTo<void3u>(fg->mul_);
-//	if (fg->sqr_) {
-//		op.sqr = Xbyak::CastTo<void2u>(fg->sqr_);
-//	}
+	op.sqr = Xbyak::CastTo<void2u>(fg->sqr_);
 	op.preInv = Xbyak::CastTo<int2u>(op.fg->preInv_);
 	op.invOp = &invOpForMont;